Valuation of Ecological Benefits: Improving the Science
Behind Policy Decisions
INTRODUCTORY REMARKS BY
MIKE SHAPIRO
DEPUTY ASSISTANT ADMINISTRATOR,
U.S. EPA, OFFICE OF WATER
A WORKSHOP SPONSORED BY THE U.S. ENVIRONMENTAL PROTECTION
AGENCY'S NATIONAL CENTER FOR ENVIRONMENTAL ECONOMICS (NCEE)
AND NATIONAL CENTER FOR ENVIRONMENTAL RESEARCH (NCER)
October 26-27, 2004
Wyndham Washington Hotel
Washington, DC
Prepared by Alpha-Gamma Technologies, Inc.
4700 Falls of Neuse Road, Suite 350, Raleigh, NC 27609
-------
ACKNOWLEDGEMENTS
This report has been prepared by Alpha-Gamma Technologies, Inc. with funding from
the National Center for Environmental Economics (NCEE). Alpha-Gamma wishes to
thank NCEE's Cynthia Morgan and the Project Officer, Ronald Wiley, for their guidance
and assistance throughout this project.
DISCLAIMER
These proceedings are being distributed in the interest of increasing public understanding
and knowledge of the issues discussed at the workshop and have been prepared
independently of the workshop. Although the proceedings have been funded in part by
the United States Environmental Protection Agency under Contract No. 68-W-01-055 to
Alpha-Gamma Technologies, Inc., the contents of this document may not necessarily
reflect the views of the Agency and no official endorsement should be inferred.
2
-------
Transcript of Introductory Remarks
Mike Shapiro, Deputy Assistant Administrator
U.S. EPA, Office of Water
It's a pleasure to be here. My role really is to come here as, basically, your primary
customer, from our perspective. As Robin (Jenkins) mentioned, the Office of Water has
a strong interest in advancing the art of assessing the benefits from ecological
improvements. That stems very directly from the mission of our office and our
experience in implementing our statutory mandates over the last three decades. The
Clean Water Act, which is one of the two main statutes that our office implements,
provides the national goal that "wherever attainable, an interim goal of water quality
which provides for the protection and propagation of fish, shellfish, and wildlife, and
provides for recreation in and on the water be achieved." That's just the interim goal, but
we haven't gotten there quite yet, even though I believe 1983 was the date we were
supposed to achieve it. There are a whole variety of reasons for that. That part of the
statute clearly was aspirational—no one sued us over missing that deadline. However, I
think it emphasizes a couple of key aspects of implementing the Clean Water Act.
First and foremost, I think the notion that the importance of the quality of aquatic systems
beyond just straight contribution to human health is an overriding concern in the Clean
Water Act, much more so than in a number of the other statutes that I've been charged
with administering during my tenure at EPA. So, built into the decision making process
really is a very direct requirement that we look at ecological values holistically in order to
provide guidance and criteria in establishing basic water quality standards, as well as in
developing mandatory national regulations—for example, under the effluent guidelines
program. Probably more significantly, it's built into how we have to evolve a strategy for
managing and maintaining water quality in this country. So, at each layer in the decision
making process—from establishing strategic approaches to setting and achieving water
quality goals to developing implementation programs and establishing specific criteria or
regulations—we have to consider (and more often than not, it's become an overriding
driver) ecological factors in our decision making and how we set priorities and how we
go about evolving strategies and in specific actions that we have to take.
As the challenge of implementing our water programs becomes greater and greater, we
have to cope with very difficult and expensive issues to manage—such as contributions
from non-point source pollution, such as getting increasingly more stringent controls over
point sources that have already been through one or two generations of regulation. We
also find that increasingly we have to demonstrate not just that that's a good thing to do,
but that we can, first of all, evaluate the ecological impacts of what we plan to do, and
secondly, and increasingly more significantly, quantify the benefits of those impacts if we
are going to have an effective public policy debate over the evolution of water programs
in this country. And that's occurring, I think, for fundamentally the right reasons: the
nation is making huge investments in environmental quality across all the media. We
face a significant challenge in all our media over the next several decades to not only
3
-------
maintain the gains of the past but to achieve the goals that we seek for protecting human
health and the environment, and we want to know how to make the best decisions. We
want to know as a society where to invest resources and how to use those resources that
we do invest in the most effective way possible. You are a very impressive community
of researchers, and we are asking for your help in developing some of the basic tools that
we in the Office of Water, as well as EPA's other program offices, need to inform those
decisions.
It's no secret to any of you that as compared to assessing and valuing human health
effects we've trailed far behind in the area of ecological benefits, both because some of
the underlying ecosystem modeling has had to be developed and because of some of the
very challenging valuation issues posed by evaluating and valuing ecosystem services. I
think that gap is beginning to close, and in fact I think one of the impressive things about
this two-day meeting is the research products that are already beginning to become
available that many of you will be speaking about over the course of the coming days. I
think that some of that work, which is cutting edge, will pave the way to closing the gaps
that we're facing.
I think that the Office of Water, together with ORD (Office of Research and
Development) and OPEI (Office of Policy, Economics, and Innovation), have really
undertaken a sustained effort to support research in this area. Robin mentioned a number
of the aspects of that sustained effort, and you'll be hearing more about them this
morning. One is the Ecological Benefits Assessment Strategic Plan, which is largely an
internal-driven document that has laid out the fundamental priorities and principles
underlying our sustained support for ecological research. There's also the Science
Advisory Board Committee on Valuing Protection of Ecological Systems, which you'll
be hearing about shortly. That group is charged with providing us input and guidance as
we try to develop our research programs and apply new methodologies. We're also
anxiously awaiting the National Research Council's Report on Assessing and Valuing the
Services of Aquatic Systems, which again you'll be hearing more about later today.
We're very much looking forward to the guidance that that NAS panel provides, and the
support and interest in that panel is not exclusive to EPA—it also has been supported by
the Army Corps of Engineers and the Department of Agriculture, two of our key federal
partners in managing and implementing water quality programs.
So, the interest in this area and the importance of the work that you're doing (although I
like to speak of myself as your primary customer) really goes well beyond the Office of
Water and well beyond the Environmental Protection Agency as, increasingly,
management of the nation's water resources has to become a coordinated effort, certainly
across several key federal agencies as well as other levels of government.
I'd just like to close by emphasizing one point. In my tenure at EPA I think one of the
dramatic challenges we've faced across the years has been the speed by which we can
convert cutting edge research into tools that we can use in our day-to-day business of
environmental decision making, whether through rule-making or through guidance or
through a variety of other mechanisms. As I'm sure many of you are well aware, our
4
-------
practice lags what is available in the research community by a number of years—often,
too many years in my view. So, as each of you thinks about the work you're doing and
the research and new tools that you're developing, I hope that you bear in mind the
importance of the development part of research and development. We need you to help
us by applying your methods increasingly to examples that are representative of the kind
of work that we do so that we can close the gap between research and practice as quickly
and effectively as possible. We encourage you to establish the appropriate tools and
verify your work so that we can really take what you're doing and incorporate it into our
decision making as quickly as possible. Hopefully, that will be a theme that connects
many of the different sessions that we have. We're very eager to follow the work that all
of you are doing and to apply it, as appropriate, to the decisions that we have to make
every day in the Office of Water and at the Environmental Protection Agency.
I'd certainly like to thank all of you for your contributions to date and look forward to the
result of the next two days as well as the ongoing research that you're conducting.
Thank you.
5
-------
Valuation of Ecological Benefits: Improving the Science
Behind Policy Decisions
PROCEEDINGS OF
SESSION I: SELECTED ECOLOGICAL VALUATION ACTIVITIES AT EPA
A WORKSHOP SPONSORED BY THE U.S. ENVIRONMENTAL PROTECTION
AGENCY'S NATIONAL CENTER FOR ENVIRONMENTAL ECONOMICS (NCEE)
AND NATIONAL CENTER FOR ENVIRONMENTAL RESEARCH (NCER)
October 26-27, 2004
Wyndham Washington Hotel
Washington, DC
Prepared by Alpha-Gamma Technologies, Inc.
4700 Falls of Neuse Road, Suite 350, Raleigh, NC 27609
-------
ACKNOWLEDGEMENTS
This report has been prepared by Alpha-Gamma Technologies, Inc. with funding from
the National Center for Environmental Economics (NCEE). Alpha-Gamma wishes to
thank NCEE's Cynthia Morgan and the Project Officer, Ronald Wiley, for their guidance
and assistance throughout this project.
DISCLAIMER
These proceedings are being distributed in the interest of increasing public understanding
and knowledge of the issues discussed at the workshop and have been prepared
independently of the workshop. Although the proceedings have been funded in part by
the United States Environmental Protection Agency under Contract No. 68-W-01-055 to
Alpha-Gamma Technologies, Inc., the contents of this document may not necessarily
reflect the views of the Agency and no official endorsement should be inferred.
11
-------
TABLE OF CONTENTS
Session I: Selected Ecological Valuation Activities at EPA
Draft Ecological Benefits Assessment Strategic Plan
Nicole Owens, U.S. EPA, National Center for Environmental Economics 1
Valuing Ecological Protection: A Tangle, a Web, or a Fabric?
Angela Nugent, Science Advisory Board Staff Office 11
Valuing Ecosystem Services Toward Better Environmental Decision-
Making
Mark Gibson, National Academy of Sciences 19
Summary of Q&A Discussion Following Session I 33
in
-------
Draft Ecological Benefits
Assessment Strategic Plan
Nicole Owens
October 2004
The following material describes the Draft Ecological
Benefits Assessment Strategic Plan. The Plan has not
undergone a final review and should not be construed to
represent Agency Policy.
2 Workgroup Members
OAR
Linda Chappell
OPPTS
Lynne Blake-Hedges
TJ Wyatt
OPEI
Rich Iovarina
Sabrina Lovel!
Steve Newbold
Nicole Owens
~ ORD
Randy Bruins
- Wayne Munns
Will Wheeler
~ OSWER
David Charters
~ OW
- Joel Corona
- Doug Norton
~ Contractor support, ICF
Margaret McVey
1
-------
Q EPA and Ecological Benefits
~ The mission of EPA is to protect human health
and the environment.
~ To that end, EPA
Develops and enforces regulations
Sponsors and develops voluntary programs and partnerships
Conducts and sponsors environmental research
~ Identifying, quantifying, and monetizing
ecological benefits can improve decision-making.
~ Benefit-cost analysis required by executive order
and statute.
~ EPA is increasingly asked to provide concrete
support for programmatic decisions
| Ecological Benefits
~ Any improvements in human well-being
that are derived from ecosystem
services.
~ Difficult to quantify and monetize.
~ Current state of the practice values a limited
set of ecological services affected by Agency
actions.
-------
Q Goal of the Strategic Plan
"To improve the Agency's ability to
identify, quantify, and value ecological
benefits in order to improve decision-
making and better communicate the
results of Agency actions."
| Objectives of the Strategic Plan
~ "Clearly describe some of the major technical
and institutional issues that prevent the
Agency from conducting accurate and
comprehensive ecological benefits
assessments on a routine basis."
~ "Identify directions for future research, data
collection, and development of analytical
tools."
3
-------
Q Objectives of the Strategic Plan
~ "Propose activities to foster increased collaboration
and coordination among the Agency's ecologists,
economists, and other analysts in conducting
ecological benefits assessments."
~ "Propose institutional mechanisms to facilitate
adaptive implementation of this Strategic Plan,
including periodic adjustments of the Plan to reflect
progress in the state of knowledge."
The Plan will help Offices develop program-
specific Action Plans to guide investment in the
development of the methods, models, and data
needed to conduct accurate and comprehensive
ecological benefits assessments.
| Contents of the Plan
Section 1 - Introduction
1.1 Objectives of the Plan
1.2 The Role of Benefits Assessment in Agency Decision-making
1.3 Focusing on Ecological Benefits
1.4 Intended Audience and Scope of This Plan
1.5 Organization of the Plan
Section 2 - Background
2.1 Nature of the Challenge
2.2 Past EPA Efforts
2.3 Ongoing EPA Efforts
2.4 This Effort and Looking Forward
Section 3 - Linking Ecological and Economic Assessments
3.1 Current State of the Practice
3.1.1 Ecological Assessments
3.1.2 Economic Benefits Assessments
3.2 Towards an Integrated Ecological Benefits Assessment Process
4
-------
Contents Cont'd
I Section 4 - Improving Ecological Benefits Assessments
| 4.1 Identifying Issues
4.1.1 Information Gathering Activities
4.1.2 Identifying Priority Issues and Actions
4.2 Cross-Cutting Issues
4.2.1 Interdisciplinary Assessment
4.2.2 Internal and External Coordination
4.2.3 Addressing Uncertainty in Ecological Benefits Assessments
4.3 Problem Formulation
4.4 Evaluating the Effectiveness of Management Options
4.4.1 Behavioral Responses to Management Actions
4.4.2 Effectiveness of Pollution Control, Remediation, or Restoration Measures
4.5 Analyzing Ecological Changes
4.5.1 Establishing Baselines for Ecological Condition
4.5.2 Assessing Changes in Ecological Populations
4.5.3 Assessing Ecosystem Processes
4.6 Estimating Monetary Values of Ecological Changes
4.6.1 Conducting Original Valuation Studies
4.6.2 Benefit Transfer
4.7 Supplemental Approaches
4.7.1 Weighting/Ranking Procedures
4.7.2 Approaches Based on Properties of Ecological-economic Systems
4.7.3 Hybrid Approaches
Contents Cont'd
Section 5 - Implementation
5.1 Identifying Future Investments
5.2 Aligning Resources
5.3 Sustaining Improvement Efforts
5
-------
¦
Ecological and Economic
Assessments are Usually Separate
Ecological knowledge, methods, models, and data
Problem formulation: selection of
assessment end points and
development of a conceptual model _
and analysis plan
Analysis of exposure and responses to
stressors, and characterization of
effects on ecological endpoints
ECOLOGICAL RISK ASSESSMENT
Decision-
makers
¦
Ecological and Economic
Assessments are Usually Separate
Characterization of
problem and
management options
Assessment of
changes in
ecological
conditions
Valuation of changes in a
limited set of goods and services
ECONOMIC ASSESSMENT
Economic knowledge, methods, models, and data
r
\
Decision
makers
V
/
6
-------
¦
A More Integrated Process
Ecological knowledge, methods, models, and data
Problem formulation:
selection of assessment
endpoints and
development of a
conceptual model and
analysis plan
Assessment of
effects of
management
actions on
sources of
stressors
Analysis of exposure
and responses
to stressors, and
characterization of
effects on ecological
endpoints
Valuation of effects
on goods and
INTEGRATED ECOLOGICAL
BENEFITS ASSESSMENT
Decision-
makers
Economic knowledge, methods, models, and data
| Section 4
~ Describes some of the major ways in which EPA
could improve its capabilities for conducting
rigorous and comprehensive ecological benefits
assessments on a routine basis.
~ Describes key issues associated with ecological
benefits assessments and actions that should lead
to improvements.
~ The actions address directions for future
research, data collection, development of
analytical tools, and institutional changes.
-------
Q Example Issues and Actions
Issue: Communication between ecologists and economists within
EPA.
Action: Provide formal and informal opportunities for
communication among disciplines.
Action: Provide basic training in the fundamentals of other
disciplines.
Issue: Collaboration between ecologists and economists.
Action: Explore methods for expanding the use of ecological risk
assessment information in economic benefits assessments.
Action: Require multi-disciplinary participation in assessing
ecological benefits.
Action: Develop guidelines for planning and conducting
ecological benefits assessments.
| Example Issues and Actions
Issue: Ability to predict changes in ecosystem processes in response
to changing environmental stressors. Action: Identify which
ecosystem processes are most important to benefits
assessments at EPA.
Action: Identify which of the important ecosystem processes
need further research to allow model development.
Action: Develop a catalogue of existing relevant ecosystem
process models at different geographic scales to support benefits
assessment.
Action: Expand portfolio of models to address the ecosystem
processes important to benefits assessment at multiple
geographic scales.
Action: Address data needs for those models.
Action: Evaluate other options for estimating changes in
ecosystem processes.
8
-------
Q Example Issues and Actions
Issue: Using existing valuations studies for benefit transfer.
Action: Encourage researchers to estimate values for a wider
variety of ecological resources.
Action: Encourage researchers to use standardized measures of
ecological resources in valuation studies.
Action: Encourage researchers to estimate and report values for
a greater range of ecological changes.
Action: Support the development of new publication outlets.
| Section 5
~ Identifies Agency actions needed to
implement the Plan.
- Further define research and development needs
and communicate those needs by developing
office-specific Action Plans.
Develop a systematic method to guide
prioritization of the investment opportunities
identified in the Plan and individual program office
Action Plans.
Track progress and integrating ecological benefits
assessment into the Agency's base programs.
9
-------
Status
SAB Committee on Valuing the Protection of
Ecological Systems and Services review
January 2005
SAB Review Draft available late fall 2004
Want a copy?
Sign-up sheet on registration table
- Email owens.nicole@epa.gov
-------
Valuing Ecological Protection: A
Tangle, a Web, or a Fabric?
A Look at the Work of the
SAB Committee on Valuing
the Protection of Ecological
Systems and Services
EPA SAB Staff Office
History of Project
SAB's Executive Committee conceived Project (2002)
Project within SAB mission
- To provide external, independent advice on the scientific and
technical aspects of environmental issues to help inform
environmental decision-making
- Advice directed at the technical bases of EPA policies,
regulations, research, and science programs
Project supported as an SAB priority by EPA's
Science Policy Council
EPA SAB Staff Office
11
-------
Charge to the Committee
• To assess:
- Agency needs and the state of the art and
science of valuing protection of ecological
systems and services
- To identify key areas for improving knowledge,
methodologies, practice, and research
• A multi-disciplinary, multi-year effort
EPA SAB Staff Office
Formation of Committee
• Committee formed in August 2003
• 24 members - with experience, expertise, and range
of views in different fields
- 7 Economists
- 8 Environmental Scientists (Ecologists/Biologists)
- 9 "Others":
• Decision Science
• Engineering
• Law
• Philosophy
• Political Science
• Psychology
EPA SAB Staff Office
12
-------
Strategy and Steering
• "Value" - untangling its
meanings
• Important to provide
advice to help EPA
make decisions
• Steering Committee
established - February
2004
EPA SAB Staff Office
Focusing on 4 Types of EPA Needs
1. Valuation of ecological benefits for National
Rulemaking
2. Assessing options, priority setting for
Regional decision-making
3. Assessing ecological benefits for GPRA
compliance
4. Communicating ecological benefits to the
public
EPA SAB Staff Office
13
-------
Key Technical Issues
Addressing non-use values
Expressing data quality and uncertainty
Appropriate use of monetized, quantified, and
qualitative methods
Assumptions about:
- Elasticity and substitution
- Transferability
- Stability of ecological systems
- Discounting benefits
Appropriate role of public in developing scientific
information
EPA SAB Staff Office
Planned Committee Activities
Advisory on Agency draft Ecological Benefits
Assessment Strategic Plan
Public meetings/workshops focused on EPA
decision needs
"Example Exercises" to meet those needs
Public meetings on key technical issues
Learning from/building on work of others
Final report (2005-2006) addressing EPA's needs
EPA SAB Staff Office
14
-------
US EPA Science Advisory Board (SAB) Committee on Valuing the Protection of
Ecological Systems and Services
Fact-Sheet
Charge
The SAB initiated this project to assess Agency needs and the state of the art and science
of valuing protection of ecological systems and services, and then to identify key areas
for improving knowledge, methodologies, practice, and research.
Committee Membership
The Committee is an inter-disciplinary group (24 Members) of ecologists, economists,
engineers, other environmental specialists, and related disciplines. A committee roster is
attached to this fact sheet. The Committee has organized a Steering Group to assist the
Chair and the Designated Federal Officer, Dr. Angela Nugent, in planning the work of
the Committee.
Approach
To fulfill this charge, the SAB Committee appointed by the Administrator will conduct a
multiyear initiative with the goal of providing a first approximation of the advice needed
by the Environmental Protection Agency.
• They will also advise the Agency on its draft Ecological Benefits Assessment
Strategic Plan
• They will host workshops on science-based approaches to valuing the protection
of ecological systems and services used in practice by groups outside EPA: in
other federal agencies, state governments, environmental groups, business entities
and international organizations.
• The Committee will focus on specific EPA decision-making needs by reviewing a
range of EPA analyses supporting those needs and by intensively working on
related "examples."
• At the conclusion of the two-year initiative, the Committee will issue a final
report assessing overall Agency needs and provide advice for strengthening the
Agency's approaches for valuing the protection of ecological systems and
services, their use by decision makers, and the key research areas needed to
strengthen the science base.
Specific Areas of Focus on EPA Decision-Making Needs
• Needs for benefit assessments supporting regulations protecting ecological
systems and services
• Regional needs for assessing and communicating the value of protecting
ecological systems and services
• Needs for assessing and communicating to Congress, the Executive Branch, and
the public the value of EPA's programs protecting ecological systems and services
under the Government Performance and Results Act
• Needs for information/communication products to communicate to the general
public about EPA regulatory decisions protecting ecological systems and services
and information/communication products encouraging voluntary actions to
protect ecological systems and services
15
-------
Status of Work
• The Committee held an "Initial Background Workshop" on October 27, 2004. The
purpose was to identify the range of EPA's needs for science-based information
on valuing the protection of ecological systems and services.
• Minutes are posted on the web at:
http://www.epa.gov/sciencel/04minutes/cvpess_102703m.pdf.
• The Committee held a "Workshop on Different Approaches and Methods for
Valuing the Protection of Ecological Systems and Services" on April 13-14, 2004.
• Minutes are posted on the web at:
http://www.epa.gov/sciencel/04minutes/valueprotecosys41304min.pdf
• The Committee held an Advisory Meeting focused on support documents for
national rulemakings on June 14-15, 2004.
• Minutes are posted on the web at:
http://www.epa.gov/sciencel/04minutes/cvpess_061404m.pdf
• The Committee held an advisory meeting in San Francisco on Sept. 13, 14, and 15
focused on regional science needs, work-products, and activities by holding panel
discussions, briefings, and break-out groups.
• Minutes are posted on the web at:
http://www.epa.gov/sciencel/04minutes/cvpess_091304m.pdf
• The Committee will hold an advisory meeting on Jan. 25 and 26, 2005. The
purpose of this meeting will be to review EPA's Draft Ecological Benefits
Assessment Strategic Plan and Related Charge Questions and then to discuss
science needs, work-products, and activities related to requirements under the
Government Performance and Results Act for valuing the protection of ecological
systems and Services.
• Background materials for the meeting will be posted on the SAB web site
(www.epa.gov/sab) as they become available.
For Additional Information
Please contact the Designated Federal Officer, Dr. Angela Nugent by email at
nugent.angela@epa.gov or by phone at 202-343-9981.
16
-------
U.S. Environmental Protection Agency
Science Advisory Board
Committee on Valuing the Protection of Ecological Systems and Services
CHAIR
Dr. Domenico Grasso, Rosemary Bradford Hewlett Professor and Chair, Picker Engineering
Program, Smith College, Northampton, MA
MEMBERS
Dr. William Louis Ascher, Dean of the Faculty, Bauer Center, Claremont McKenna College,
Claremont, CA
Dr. Gregory Biddinger, Environmental Programs Coordinator, ExxonMobil Biomedical
Sciences, Inc, Houston, TX
Dr. Ann Bostrom, Associate Professor, School of Public Policy, Georgia Institute of
Technology, Atlanta, GA
Dr. James Boyd, Senior Fellow, Director, Energy & Natural Resources Division, Resources for
the Future, Washington, DC
Dr. Robert Costanza, Professor/Director, Gund Institute for Ecological Economics, School of
Natural Resources, University of Vermont, Burlington, VT
Dr. Terry Daniel, Professor of Psychology and Natural Resources, Department of Psychology,
Environmental Perception Laboratory, University of Arizona, Tucson, AZ
Dr. A. Myrick Freeman, Research Professor of Economics, Department of Economics,
Bowdoin College, Brunswick, ME
Dr. Dennis Grossman, Vice President for Science, Science Division, NatureServe, Arlington,
VA
Dr. Geoffrey Heal, Paul Garrett Professor of Public Policy and Business Responsibility,
Columbia Business School, Columbia University, New York, NY
Dr. Robert Huggett, Consultant and Professor Emeritus, College of William and Mary,,
Dr. Douglas E. MacLean, Professor, Department of Philosophy, University of North Carolina,
Chapel Hill, NC
17
-------
Dr. Harold Mooney, Paul S. Achilles Professor of Environmental Biology, Department of
Biological Sciences, Stanford University, Stanford, CA
Dr. Louis F. Pitelka, Director and Professor, Appalachian Laboratory, University of Maryland
Center for Environmental Science, Frostburg, MD
Dr. Stephen Polasky, Fesler-Lampert Professor of Ecological/Environmental Economics,
Department of Applied Economics, University of Minnesota, St. Paul, MN
Dr. Paul G. Risser, Chancellor, Oklahoma State Regents for Higher Education, Oklahoma City,
OK
Dr. Holmes Rolston, University Distinguished Professor, Department of Philosophy, Colorado
State University, Fort Collins, CO
Dr. Joan Roughgarden, Professor, Biological Sciences and Evolutionary Biology, Stanford
University, Stanford, CA
Dr. Mark Sagoff, Senior Research Scholar, Institute for Philosophy and Public Policy, School
of Public Affairs, University of Maryland, College Park, MD
Dr. Kathleen Segerson, Professor, Department of Economics, University of Connecticut, Storrs,
CT
Dr. Paul Slovic, Professor, Department of Psychology, Decision Research, Eugene, OR
Dr. V. Kerry Smith, University Distinguished Professor, Department of Agricultural and
Resource Economics, College of Agriculture and Life Sciences, North Carolina State University,
Raleigh, NC
Dr. Robert Stavins, Albert Pratt Professor of Business and Government, Environment and
Natural Resources Program, John F. Kennedy School of Government, Harvard University,
Cambridge, MA
Dr. Barton H. (Buzz) Thompson, Jr., Robert E. Paradise Professor of Natural Resources Law
and Vice Dean , Stanford Law School, Stanford University, Stanford, CA
SCIENCE ADVISORY BOARD STAFF
Dr. Angela Nugent, Designated Federal Officer, 1200 Pennsylvania Avenue, NW, Washington,
DC, Phone: 202-343-9981, Fax: 202-233-0643, (nugent.angela@epa.gov)
18
-------
Valuing Ecosystem Services
Toward Better Environmental Decision-Making
National Research Council
Mark Gibson
Study Director
Presentation to U.S. EPA Workshop:
Improving the Valuation of Ecological Benefits
October 26, 2004
Available on-line at http:/www.nap.edu/cataloq/11139.htmi
Committee and Process
~ 5 meetings, 1 consensus report, extensive external review process
~ Committee on Assessing and Valuing the Services of Aquatic and
Related Terrestrial Ecosystems:
• GEOFFREY M. HEAL, Chair, Columbia University
• EDWARD B. BARBIER, University of Wyoming
• KEVIN J. BOYLE, University of Maine
• ALAN P. COVICH, University of Georgia
• STEVEN P. GLOSS, U.S. Geological Survey
• CARLTON H HERSHNER, JR., Virginia Institute of Marine Science
• JOHN P. HOEHN, Michigan State University
• STEPHEN POLASKY, University of Minnesota
• CATHERINE M. PRINGLE. University of Georgia
• KATHLEEN SEGERSON, University of Connecticut
• KRISTIN SHRADER-FRECHETTE, University of Notre Dame
19
-------
Report Reviewers
• Mark Brinson, East Carolina University
• J. Baird Callicott, University of North Texas
• Nancy Grimm, Arizona State University
• Michael Hanemann, University of California, Berkeley
• Peter Kareiva, The Nature Conservancy
• Raymond Knopp, Resources for the Future
• Sandra Postel, Global Water Policy Project
• Robert Stavins, Harvard University
Statement of Task
The committee will evaluate methods for assessing
services and the associated economic values of aquatic
and related terrestrial ecosystems. The committee's
work will focus on identifying and assessing existing
economic methods to quantitatively determine the
intrinsic value of these ecosystems in support of
improved environmental decision-making, including
situations where ecosystem services can be only partially
valued. The committee will also address several
key questions, including;
20
-------
Statement of Task
(continued)
What is the relationship between ecosystem services and the
more widely studied ecosystem functions?
For a broad array of ecosystem types, what services can be
defined, how can they be measured, and is the knowledge of
these services sufficient to support an assessment of their value
to society?
What lessons can be learned from a comparative review of past
attempts to value ecosystem services—particularly, are there
significant differences between eastern and western U.S.
perspectives on these issues?
What kinds of research or syntheses would most rapidly
advance the ability of natural resource managers and decision-
makers to recognize, measure, and value ecosystem services?
Considering existing limitations, error, and bias in the
understanding and measurement of ecosystem values, how can
available information best be used to improve the quality of
natural resource planning, management, and regulation?
Report Organization
EXECUTIVE SUMMARY
1. INTRODUCTION
2 THE MEANING OF VALUE AND USE OF ECONOMIC VALUATION IN
THE ENVIRONMENTAL POLICY DECISION-MAKING PROCESS
3 AQUATIC AND RELATED TERRESTRIAL ECOSYSTEMS
4 METHODS OF NONMARKET VALUATION
5 TRANSLATING ECOSYSTEM FUNCTIONS TO THE VALUE OF
• ECOSYSTEM SERVICES: CASE STUDIES
6 JUDGMENT, UNCERTAINTY, AND VALUATION
7 ECOSYSTEM VALUATION; SYNTHESIS AND FUTURE DIRECTIONS
APPENDIXES
A Summary of Related NRC Reports
B Household Production Function Models
C Production Function Models
D Committee and Staff Biographical Information
21
-------
Introduction and Overview
• The study was conceived in 1997 at a strategic planning session
of Water Science and Technology Board of the NRC
• In early November 1999 the NRC organized and hosted a
planning workshop to assess the feasibility of and need for an
NRC study of the functions and associated economic values of
aquatic and terrestrial ecosystems
• The report focuses on the goods and services provided by
aquatic and related terrestrial ecosystems and reflects an
intentional effort to focus on management and valuation issues
confronting state and federal agencies for these ecosystems
• Because the principles and practices of valuing ecosystem goods
and services are rarely sensitive to whether the underlying
ecosystem is aquatic or terrestrial, the report's various
conclusions and recommendations are likely to be directly or at
least indirectly applicable to the valuation of the goods and
services provided by any ecosystem
Connections Between Ecosystem Structure and
Function, Goods and Services, Policies, and Values
Nonuse values
e.g., existence, species preservation,
biodiversity, cultural heritage
Consumptive use
e.g., harvesting, water supply (irrigatii
drinking), genetic and medicinal
| Nonconsumptive use
Indirect
e.g., UVB protection, habitat
support, flood control, pollution
control, erosion prevention
22
-------
The Meaning of Value and Use of Economic Valuation
• Recent philosophical debates regarding ecosystem value generally
derive from two points of view (1) values of ecosystems and their
services are non-anthropocentric and (2) all values are
anthropocentric
• Although economic valuation does not capture all sources or types of
value, it is much broader than usually presumed. It recognizes that
economic value can stern from the use of an environmental resource
(use values), or from its existence even in the absence of use (nonuse
value)
' The broad array of values included under this approach is captured by
using the total economic value (TEV) framework. The TEV framework
helps to provide a checklist of potential impacts and effects that need
to be considered in valuing ecosystem services
• A valuation question can be framed in terms of two alternative
measures of value, willingness to pay (WTP) and willingness to accept
(compensation) (WTA). these two approaches imply different
presumptions about the distribution of property rights and can differ
substantially
¦ In many contexts, methodological limitations necessitate the use of WTP
rather than WTA
The Meaning of Value and Use of Economic Valuation
Major Recommendations
• Policymakers should use economic valuation as a
means of evaluating the trade-offs involved in
environmental policy choices; an assessment of benefits
and costs should be part of the information set available
to policymakers in choosing among alternatives
• Economic valuation of changes in ecosystem services
should be based on the comprehensive definition
embodied in the TEV framework, including both use and
nonuse values
• The valuation exercise should be framed properly. In
particular, it should value the changes in ecosystem
good or services attributable to a policy change
23
-------
Aquatic and Related Terrestrial Ecosystems
• The phrase "aquatic and related terrestrial ecosystems"
recognizes the impossibility of analyzing aquatic systems absent
consideration of the linkages to adjacent terrestrial environments
• There have only been a few attempts to develop explicit maps of
the linkage between aquatic ecosystem structure/function and
value. There are, however, a multitude of efforts to separately
identify ecosystem functions, goods, services, values, and/or
other elements in the linkage
• From an ecological perspective, the value of specific ecosystem
functions/services is entirely relative. The spatial and temporal
scales of analysis are critical determinants of potential value
• There remains a need for a significant amount of research in the
ongoing effort to codify the linkage between ecosystem structure
and function and the provision of goods and services for
subsequent valuation
• A comprehensive identification of all functions and derived
services may never be achieved; nevertheless, comprehensive
information is not generally necessary to inform management
decisions
Aquatic and Related Terrestrial Ecosystems
Major Recommendations
• Aquatic ecosystems generally have some capacity to
provide consumable resources, habitat for plants and
animals, regulation of the environment, and support for
nonconsumptive uses, but considerable work remains to
be done in documentation of the potential of various
aquatic ecosystems for contribution in each of these
broad areas
• Because delivery of ecosystem goods and services
occurs in both space and time, investigation of the spatial ^
and temporal thresholds of significance for various
ecosystem services is necessary to inform valuation
efforts
• Natural systems are dynamic and frequently exhibit
nonlinear behavior, and caution should be used in
extrapolation of measurements in both space and time.
Methods are needed to assess and articulate this
uncertainty as part of system valuations
24
-------
Methods of Nonmarket Valuation
• Although a variety of nonmarket valuation approaches are currently
available, no single method can be considered best at all times and for
all types of aquatic ecosystem applications
• Revealed-preference methods can be applied only to a limited number
of ecosystem services. However, both the range and the number of
services that can potentially be valued are increasing with the
development of new methods
• Stated-preference methods can be more widely applied, and certain
values can be estimated only through the application of such techniques
¦ However, the credibility of estimated values for ecosystem services derived
from stated-preference methods has often been criticized \
• Benefit transfers and replacement cost and cost of treatment methods
are increasingly being used in environmental valuation, although their
application to aquatic ecosystem services is still limited and potentially
problematic
• Only a limited number of ecosystem services have been valued to date,
and effective treatment of aquatic ecosystem services in benefit-cost
analyses requires that more services be valued
Methods of Nonmarket Valuation
Major Recommendations
• Specific attention should be given to funding research at the "cutting
edge" of the valuation field, such as dynamic production function
approaches, general equilibrium modeling of integrated ecological-
economic systems, conjoint analysis, and combined stated-
preference and revealed-preference methods
• Specific attention should be given to funding research on improved
valuation study designs and validity tests for stated-preference
methods applied to determine the nonuse values associated with
aquatic and related terrestrial ecosystem services
• Benefit transfers should be considered a "second-best" method of
ecosystem services valuation and should be used with caution and
only if appropriate guidelines are followed
• The replacement cost method and estimates of the cost of treatment
are not valid approaches to determining benefits and should not be
employed to value aquatic ecosystem services. In the absence of
any information on benefits, and under strict guidelines, treatment
costs could help determine cost-effective policy action
25
-------
Case Studies and Lessons Learned
• Chapter 5 provides a series of case studies of the integration of
ecology and economics necessary for valuing the services of aquatic
and related terrestrial ecosystems
¦ First reviewed are situations in which the focus is on valuing a single
ecosystem service. Even when the goal of a valuation exercise is focused
on a single ecosystem service, a workable understanding of the functioning
of large parts or possibly the entire ecosystem may be required
¦ Attempts to value multiple ecosystem services are reviewed next. Since
ecosystems produce a range of services, and these services are frequently
closely connected, it is often hard to discuss valuation of a single service in
isolation. In addition, valuing multiple ecosystem services typically
multiplies the difficulty of evaluation
¦ Last to be reviewed are analyses that attempt to encompass all services
produced by an ecosystem. Such efforts will typically face large gaps in
understanding and information in both ecology and economics
• Chapter 5 also includes an extensive discussion of various implications
and lessons learned from the case studies that are reviewed. For
some policy questions, enough is known about ecosystem service
valuation to help in decision-making. For others, knowledge and
information may not yet be sufficient to estimate the value of
ecosystem services with enough precision to answer policy-relevant
questions
Case Studies and Lessons Learned
Major Recommendations
• Estimates of ecosystem value need to be placed in
context; assumptions about conditions in ecosystems
outside the target ecosystem and assumptions about
human behavior and institutions should be clearly
specified
• Concerted efforts should be made to overcome existing
institutional barriers that prevent ready and effective
collaboration among ecologists and economists regarding
the valuation of ecosystem services. Furthermore,
existing and future interdisciplinary programs aimed at
integrated environmental analysis should be encouraged
and supported
26
-------
Judgment, Uncertainty, and Valuation
• The valuation of aquatic ecosystem services inevitably involves
investigator judgments and some amount of uncertainty. Although
unavoidable, uncertainty and the need to exercise professional
judgment are not debilitating to ecosystem valuation
¦ It is also important that the sources of uncertainty be acknowledged
minimized, and accounted for in ways that ensure that a study's results
and related decisions regarding ecosystem valuation are not
systematically biased and do not convey a false sense of precision
• There are several cases in which investigators must use professional
judgment in ecosystem valuation regarding how to frame a valuation
study, how to address the methodological judgments that must be
made during the study, and how to use peer review to identify and
evaluate these judgments
¦ However, when such judgments are made it is important to explain why
they are needed and to indicate the alternative ways in which judgment
could have been exercised
• Just as there are different types of uncertainty in ecosystem valuation,
there are also different ways and decision criteria that an analyst can
use to allow for (and reduce) uncertainty in the support of
environmental decision-making
Judgment, Uncertainty, and Valuation
Major Recommendations
• If the good or service being valued is unique and not easily
substitutable with other goods or services, then the decision to
use WTP or WTA are likely to result in very different valuation
estimates
¦ In such cases, the committee cannot reasonably recommend that the
analyst report both sets of estimates in a form of sensitivity analysis because
this may effectively double the work; rather, the analyst should document
carefully the ultimate choice made and clearly state that the answer would
probably have been higher or lower had the alternative measure been
selected and used
• Ecosystem valuation studies should undergo external review by
peers and stakeholders early in their development when there
remains a legitimate opportunity for revision of the study's key
judgments
• Analysts should establish a range for the major sources of
uncertainty in an ecosystem valuation study whenever possible
• Under conditions of uncertainty, irreversibility, and learning, there
should be a clear preference for environmental policy measures
that are flexible and minimize the commitment of fixed capital or
that can be implemented on a small scale on a pilot or trial basis
27
-------
Ecosystem Valuation:
Synthesis And Future Directions
• Chapter 7 seeks to synthesize the current knowledge
regarding ecosystem valuation in a way that will be useful
to resource managers and policymakers as they
incorporate the value of ecosystem services into their
decisions, and includes the following:
¦ A synthesis of the report's general premises (10 total)
¦ A synthesis of the report's major conclusions \
¦ Guidelines and a checklist for conducting ecosystem services
valuation
¦ Overarching recommendations for conducting ecosystem valuation
¦ Overarching research needs, which imply recommendations
regarding future research funding
28
-------
Report in Brief • November 2004
Valuing Ecosystem Services
Toward Better Environmental
Decision-Making
Lake Mendota, Wisconsin. Photo courtesy Wisconsin Department of Natural Resources
Until the economic value of ecosystem
goods and services is acknowledged in
environmental decision-making, they will
implicitly be assigned a value of zero in cost-
benefit analyses, and policy choices will be
biased against conservation. The National
Research Council report, Valuing Ecosystem
Sendees: Toward Better Environmental Decision-
Making, identifies methods for assigning
economic value to ecosystem services—even
intangible ones—and calls for greater
collaboration between ecologists and
economists in such efforts.
The millions of miles of rivers, streams,
coastline, and acres of estuaries, wetlands,
lakes, and reservoirs throughout the United
States host a vast array of aquatic ecosystems that
provide many benefits to humans. These ecosystems
produce not only goods such as lumber and fish, but
they also provide a number of important functions or
services that play crucial roles in supporting human,
animal, and plant populations. These sendees include
nutrient recycling, habitat for plants and animals, flood
control, and water supply (see Box 1).
Human activities often compete with ecosystem
survival. For example, should a wetland be drained
for suburban housing? Although the economic value
of the new houses may be known, it is not as easy to
quantify the value the lost ecosystem services of the
wetland that would affect plant and animal life, alter
storm runoff patterns, and interfere with water
reclamation, among other impacts. Likewise, the
decision to build a dam to meet drinking water and
electricity needs could have dramatic consequences
011 downstream ecosystems.
In order to appropriately assess environmental
policy alternatives and the decisions that follow, it is
essential to consider not only the value of the human
activity, but also to consider the value of the ecosystem
service that could be compromised. Despite a growing
recognition of the importance of ecosystem sendees,
their value is often overlooked in decision-making, and,
to date, that value has not been well quantified.
Valuation Should Measure Trade-Offs
The Catskills/Delaware watershed provides 90
percent of the drinking water for the New York City
metropolitan area. Historically, the watershed has
produced high quality water with little contamination,
but increased housing developments, septic systems,
and agriculture caused water quality to deteriorate.
By 1996, New York City had two choices: build a
water filtration system at an estimated cost of up to
$6 billion or protect its major watershed.
When possible in environmental decision-making,
policymakers should use economic valuation as a way
Box 1. Examples of Sendees from Various
Aquatic Ecosystems
Wetlands transform inputs (nutrients, energy)
into valuable outputs (fish, crustaceans, and
mollusks).
Floodplains along rivers and coasts provide
flood protection, water reclamation, pollution
abatement, underground water recharge, and
recreation.
Mountain watersheds provide water supply,
recreation (e.g., hiking, camping, and fishing).
THE NATIONAL ACADEMIES
Advisers to the Nation on Science, Engineering, and Medicine
29
-------
to quantify the trade-offs in a policy choice. In order
to protect the Catskills watershed, measures were
taken to help limit further development, improve
sewage systems, and reduce the impact of agriculture
by using less fertilizers and building up riparian zones
along river banks at a total projected investment of
about $1 to $1.5 billion. New York City water
managers chose to protect the watershed.
Link Economic and Ecological Models
In the Hadcjia-Jama'arc floodplain in Northern
Nigeria, economists and hydrologists worked together
to estimate both upstream benefits and downstream
consequences of several proposed dam and water
diversion projects. A 1998 study showed that the
benefit of the project was $3 million in irrigation and
potable water, but that downstream floodplain losses
would result in about $23 million dollars in costs; an
estimated net loss of $20 million. A study in 2001
found that a one meter drop in groundwater would
result in an estimated $1.2 million loss in dry season
agriculture and a $4.8 million loss in domestic water
consumption for rural households.
Economists already produce estimates of value
for environmental decision-making. However, the
strength of their analysis depends in large part on how
well the underlying ecology of an ecosystem is
understood and measured. Ecologists are challenged
because ecosystems are complex, dynamic, variable,
interconnected, and nonlinear, and because our
understanding of the services they provide and how
they are affected by human actions are imperfect and
difficult to quantify.
In an analysis, it is important to ensure that the
ecosystem is well understood and also that the study
is designed so that output from ecological models can
be used as input to the economic models so that the
two can be linked effectively. The example of the
Nigerian floodplain also illustrates the importance of
measuring expected changes in the ecosystem for a
given ecological impact. Other changes that could
be measured include stream flow, water temperature,
and changes in the plant life and fish of the floodplain.
Consider All Ways Ecosystems are Valued
Clean drinking water, food production, and
recreation are all services of a lake ecosystem, but it
is not easy to measure each one separately or to
resolve conflicting views on which is more or less
important to a management decision. Many
economists use the Total Economic Valuation (TEV)
Framework to incorporate the multiple ways that
individuals or groups could value an ecosystem—most
of which have no market or commercial basis (see
Figure 1). Elements of the framework include:
• Use and Nonuse Values: Although different
TEV frameworks are used to assess value, most
ECOSYSTEM GOODS
& SERVICES
Nonuse values
e.g., existence, species preservation,
biodiversity, cultural heritage
Figure 1. The figure shows the multiple types of values from ecosystem goods and services
that are considered within a total economic valuation (TEV) framework.
30
-------
of them include both "use" and ""nonusc" values.
For example, an oil spill on a popular beach that
prevents people from using it represents lost use
value. Alternatively, if the oil spill did not disrupt
beach use, but damaged plant and animal life
offshore, this would represent a lost nonuse
value. Use values can be further divided into
consumptive uses (goods, water supply) and
nonconsumptive uses (recreation, habitat support,
flood control).
• Willingness to Pay and Willingness to Accept:
If the quality of a freshwater lake were improved
to enhance sportfishing, the economic measure
of the value of such an improvement to a
recreational angler would be measured by his
willingness to pay for such a change. If however
the quality of a lake was worsened from its current
level, then the economic measure to a recreational
angler would be his willingness to accept
compensation for the damage, or the minimum
amount of money the angler would accept as
compensation.
Quantify Ecological Impacts
How can a dollar amount be applied to ecosystem
changes? There are several economic methods that
can be used to place a value on ecosystem services
(see Box 2). These methods base values on various
aspects of consumer and producer behaviors, and
draw on stated or revealed individual preferences.
In the Great Lakes, policymakers conducted a
complex analysis to decide whether and how to
control the sea lamprey, an invasive species that preys
on the native lake trout, sturgeon, salmon, and other
large fish. One study polled 2,000 Michigan anglers
to estimate the value to them of a higher catch rate at
various fishing sites, taking into consideration distance
and travel costs to those sites. The study showed
that even a 10% increased catch rate would have a
value of about $3.3 million to fisherman. This value
was compared against the cost of various methods
to control the sea lampreys, for example using a
lampricide treatment, so that an appropriate decision
could be made.
Specific attention should be paid to pursuing
research at the "cutting edge" of the valuation field to
support this type of analysis. Because they are time
consuming, project-specific valuations have sometimes
been replaced by the benefits transfer method, which
assesses value based on an existing study of a similar
ecosystem. However, benefit transfer methods should
be considered second best to careful analysis of the
specific ecosystem in question.
Incorporating Judgment and Uncertainty
Perhaps the most important choice in any
ecosystem valuation study is how the initial question
is framed. In the Catskills/Delaware watershed,
policymakers made the critical decision early on that
it was not necessary to value all the services of the
watershed, but instead to focus only on water quality.
Other judgments may be necessary in framing an
issue, for example the choice between using the
Box 2. Assigning a Dollar Value:
Nonmarket Valuation Methods
Following are some of the most common methods
that are used to measure the economic value of
ecosystems services.
Household Production Function Methods model
consumer behavior based on the assumption that
ecosystem services can be substitutes for or
complementary to a marketed commodity. Travel-
cost models infer the value of an ecosystem
according to the travel time and costs needed to
visit it. Averting behavior models quantify what
people would spend to avoid a negative impact on
health, for example installing a filter if water quality
is poor. Hedonic methods analyze how
characteristics, including environmental quality,
alter how much people would pay for something.
Production Function Methods model the
behavior of producers and their response to
changes in environmental quality that influence
production. These methods have been applied to
explore the habitat-fishery, water quality-fishery
linkages, and erosion control and storm protection.
Stated-Preference Methods are commonly used
to measure the value people place on a particular
environmental item. Examples include how much
people would pay annually to obtain swimmable,
fishable, and drinkable freshwater, or to protect
Pooling Revealed- and Stated-Preference
Methods uses combined data from different
valuation methods to estimate a single model of
preferences.
Benefit Transfer Methods estimate the value an
ecosystem based on existing studies of a roughly
similar ecosystem.
31
-------
concept of willingness to pay or willingness to accept
in an analysis.
Uncertainty can arise at many steps in an
analysis. For ecosystem valuation, one of the biggest
sources of uncertainty is the lack of probabilistic
information about the likely magnitudes of some
variables. Other sources of uncertainty arise from
models or parameters used. Economic factors can
introduce uncertainty as well. For example, how does
the degree of visible cleanliness or the degree of
development and crowding affect the value of a
popular recreational watersite?
Although uncertainty and judgment are inevitable,
they are not debilitating to ecosystem valuation and
do not undermine the validity of the analysis. It is
only necessary to provide a clear explanation of how
judgments were made and how uncertainties were
accounted for.
Overarching Recommendations
When faced with environmental policy decisions
that seek to balance human activity and conservation,
the process of valuing ecosystem services can inform
the policy debate and lead to better decision-making.
The report makes the following recommendations for
how policymakers should conduct ecosystem
valuations:
• Seek to evaluate trade-offs: where possible, value
should be measured in a way that makes analysis
of trade-offs possible. If the benefits and costs
of an environmental policy are evaluated, then
the benefits and costs associated with the changes
in an ecosystem service must be evaluated as well.
• Frame the valuation appropriately: Measure
changes in ecosystem services, rather than the
value of an entire ecosystem.
• Delineate all sources of value from the ecosystem
and determine whether they are captured in the
valuation.
• Quantify ecological impacts where possible
beyond a simple listing and qualitative description
of affected ecosystem services.
• Make sure that economic and ecological models
are appropriately linked. The output from
ecological modeling must be in a form that can
be used as an input to economic analysis.
• Seek to value the goods and services most
important to a particular policy decision.
• Base economic valuation of ecosystem changes
on the total economic value framework. Include
both use and nonuse values.
• Consider all relevant impacts and stakeholders in
the scope of the valuation.
• Scrutinize any extrapolations made across space
(from one ecosystem to another), time (from
present to future impacts), and scale (from small
to large changes) to avoid extrapolation errors.
Committee on Assessing and Valuing the Services of Aquatic and Related Terrestrial Ecosystems:
Geoffrey M. Heal (Chair), Columbia University, New York; Edward B. Barbier, University of Wyoming,
Laramie; Kevin J. Boyle, University of Maine, Orono; Alan P. Covich, University of Georgia, Athens; Steven
P. Gloss, Grand Canyon Monitoring and Research Center, U.S. Geological Survey, Flagstaff, Arizona; Carlton
H. Hershner, Virginia Institute of Marine Science, Gloucester Point; John P. Hoehn, Michigan State University,
East Lansing; Catherine M. Pringle, University of Georgia, Athens; Stephen Polasky, University of Minnesota, St.
Paul; Kathleen Segerson, University of Connecticut, Storrs; Kristin Shrader-Frechette, University of Notre
Dame, Notre Dame, Indiana; Mark C. Gibson (Study Director) and Ellen A. De Guzman (Research Associate),
Water Science and Technology Board.
This brief was prepared by the National Research Council based on the committee's report. For
more information, contact the Water Sciences and Technology Board at 202-334-3422. Valuing Ecosystem
Services: Toward Better Environmental Decision-Making is available from the National Academies Press,
500 Fifth Street, NW, Washington, DC 20001; 800-624-6242 or 202-334-3313 (in the Washington area);
www.nap.edu.
Permission granted to reproduce this brief in its entirety with no additions or alterations.
Copyright 2004 The National Academy of Sciences
32
-------
Summary of the Q&A Discussion Following Session I
Bob Leeworthy (NOAA)
Classifying what he had to say as more of a comment than a question, Dr. Leeworthy
stated that "in leading many exercises in NOAA in actual management policy
applications," he and his colleagues have found that in their dealings with communities,
if they don't address "market economic impacts on local sales, income, and
employment," they are "shown the door" and are "considered to be irrelevant." He said
that he thinks economists need to be careful not to focus just on the net economic values
"that we as economists all agree to, but which everyone else would think are irrelevant."
Mark Gibson (National Academy of Sciences)
Mr. Gibson responded that the committee at first was trying to get... the economists,
and ecologists and the environmental philosopher "on the same sheet of music." Further,
he explained that what was presented was "a very short, quick snapshot of the work" and
he hoped that a closer inspection of the report would yield more information relevant to
the work being done by Dr. Leeworthy and his colleagues at NOAA.
Ann Watkins (U.S. EPA, Office of Air and Radiation)
Addressing her question specifically to Angela Nugent but opening it to other comments,
as well, Ms. Watkins said, "I noticed that you mentioned GPRA (the Government
Performance and Results Act) as one of the things that you have considered as you
looked at the questions that we have to answer, and you also looked at PART" (the
Program Assessment Rating Tool), both of which are components of OMB's (Office of
Management and Budget) analysis of our different programs. She said that she knew of
"several programs [that] have been zeroed out because they can't provide a measure of
value that is sufficient for OMB's standards under this PART analysis."
Angela Nugent (U.S. EPA, Science Advisory Board Staff Office)
Dr. Nugent responded that "the GPRA piece is yet to be done by the committee, but it's
part of the grand plan." She went on to explain that "as we design our survey of what the
Agency is struggling with, I think a necessary part of that is dealing with this program
assessment review tool of OMB and seeing how it has been applied to programs whose
primary thrust is eco-protection and how the Agency can actually strengthen its science
base to make that case." She went on to assert that certain groups within EPA have
already begun developmental programs to "strengthen the science" or identify how it can
be strengthened. Stating that "all these things that we are now treating as separate threads
obviously need to be woven together," Dr. Nugent identified the "ultimate revision of the
steering committee" as a "move to a situation where all of these kinds of analyses,
regional, national, park level. . . would support each other."
33
-------
Ed Bender (U.S. EPA, Office of the Administrator)
Dr. Bender opened by stating, "It's very important to value things, because we don't
protect them unless we value them." He continued, "However, in ecological risk
assessment, one of the fundamental gaps is that most of the assessment endpoints deal
with individual species—not really with what ecology is about." Dr. Bender wondered if
any of the panelists "had noticed that kind of problem and had any thoughts about how
economists might be able to help us look at the more complex and higher order
interactions that I know you're trying to look at as you look at ecosystem goods and
services."
Mark Gibson
Mr. Gibson said he would love to have a committee member help address that comment,
and began by saying, "There are key studies, I believe in Chapter 4, that talked about
invasive species and trying to evaluate . . ."
Ed Bender
(interrupting) "They're an organism. I'm talking about the interaction of organisms with
each other as well as with their environment, or habitat loss, or some of the other things
that we say are so important, yet we don't really have much information—those are not
really addressed in ecological risk assessments."
Geoffrey Heal (Columbia University)
Identifying himself as the Committee Chair, Dr. Heal stated, "I'm not certain that we
really address exactly the issue you're dealing with, but what we've done in the report is
to look at the valuation of ecosystem services—those services that come from the
operation of the ecosystem as a whole, and it relates to the services provided by the
ecosystem to the existing structure, for instance the physical and chemical . . . and certain
regulatory functions. To the extent that relationships between individual species or the
existence of particular species affects the services or improves the services that come out
of an ecosystem, then I guess the result you could lay out can, in some instances, attach
value to the existence or the interaction between the individual species. It's not the task
of the report, really, but whether we construe this, it will place a value on a particular
species, other than maybe its charismatic value, because that's a non-use value. But to
the extent that species don't obviously have a straight existence value because of their
charismatic characteristics, I don't think we really address the issue how you would value
individual species. I guess the perspective we would take is that species are part of what
makes an ecosystem function, and if you would pull a species out of an ecosystem—
particularly pull a keystone species out of an ecosystem, for example—the services
provided by the system can collapse. So, there's the implicit value in the species because
of that.
Nicole Owens (U.S. EPA, NCEE)
Dr. Owens offered "one quick response to that: One of the things we talk about in the
strategic plan is how they can use that kind of information to communicate functioning of
ecosystems to the public, and also use that to address uncertainty and how you might
describe how uncertain some of our estimates of the changes in ecosystems might be to
34
-------
the public. She added, "That should come in handy whether you're developing either
surveys or focus groups—to use that information you may have on one species to try and
convey to the public something about the functioning of the whole ecosystem."
AlMcGartland (U.S. EPA, NCEE)
"I have some advice for Dr. Leeworthy: At EPA sometimes we can shame the decision-
makers into listening about benefits—after all, decision-makers should be interested in
improving the welfare of society. I often refer back to the GDP accounts—a lot of the
welfare improvements that come from environmental improvements don't get captured in
the GDP accounts, and that's actually, I think, a good hook into benefit analysis."
"My question really is that I struggle with benefits and ecosystem stuff—it seems this
whole spatial dimension is a big problem: ecologists like to do these very localized
things, and of course national regulations require national benefits. I ask the panel and
Geoff and maybe others later to address the question: Is that a show-stopper or is there
hope on the horizon for dealing with that?"
Mark Gibson
Mr. Gibson's general response was that the issue is, in fact, a concern of ecologists and it
is addressed in his Chapter 3, along with "focused conclusions and recommendations in
that regard." He concluded by saying he would not characterize it as a "show-stopper"
but that it was a difficult issue for the committee to tackle and they went as far as they
could in developing conclusions and recommendations to that effect.
Liz Strange (Stratus Consulting, Inc.)
In response to Dr. Bender's question, Dr. Strange stated that she and her colleagues did
some work in the last few years where they looked "specifically at the eco-risk
assessment framework at EPA and tried to think in terms of ecosystem services, the
goods and services provided by ecosystems, as potential assessment endpoints." She
went on to say the she thinks "that's one of the ways to get at what you're talking about
because, of course, those goods and services depend upon ecological structures and
functions—in some cases depend on individual species or communities." Dr. Strange
said she believes that on the Global Change Research Program website there is a copy of
that framework, which "essentially was integrating the eco-risk framework of EPA with
the natural resource damage assessment approach that focuses on ecological. . .
services."
As an example of "another attempt to try and get at those things and integrate those
things," Dr. Strange also mentioned that she previously worked with one of the members
of Mr. Gibson's committee, A1 Kovitch, on "an EPA/NSF-funded project looking at
ecological integrity and what are some of the endpoints that you can use to present to the
public information about what we mean by ecological health or ecosystem servicesShe
closed by adding that "there was an evaluation study associated with that research."
35
-------
Angela Nugent
Dr. Nugent referred back to the question about spatial scales and said that the issue came
up when they did an example exercise on the CAFO (Concentrated Animal Feeding
Operations) analysis. She said some folks on the committee were strong proponents of
"having case studies be part of the benefits assessment supporting the rule, either as
stand-alone case studies or something that could be used to test and validate the national
model. . ." Dr. Nugent continued by saying that there is a general sense, especially at the
region level, that "there's a tremendous opportunity there to build on this local
experience, and maybe there are some leads on the empirical side that will help us answer
the question you asked." She said she thinks people on the SAB Committee are going to
look more in depth at this question of spatial scale—and also temporal scale, the duration
of a study and what assumptions are made about change over time.
END OF SESSION I Q&A
36
-------
Valuation of Ecological Benefits: Improving the Science
Behind Policy Decisions
PROCEEDINGS OF
SESSION II: CLEANING OUR COASTAL WATERS:
EXAMINATIONS OF THE BENEFITS OF IMPROVED WATER QUALITY
A WORKSHOP SPONSORED BY THE U.S. ENVIRONMENTAL PROTECTION
AGENCY'S NATIONAL CENTER FOR ENVIRONMENTAL ECONOMICS (NCEE)
AND NATIONAL CENTER FOR ENVIRONMENTAL RESEARCH (NCER)
October 26-27, 2004
Wyndham Washington Hotel
Washington, DC
Prepared by Alpha-Gamma Technologies, Inc.
4700 Falls of Neuse Road, Suite 350, Raleigh, NC 27609
-------
ACKNOWLEDGEMENTS
This report has been prepared by Alpha-Gamma Technologies, Inc. with funding from
the National Center for Environmental Economics (NCEE). Alpha-Gamma wishes to
thank NCEE's Cynthia Morgan and the Project Officer, Ronald Wiley, for their guidance
and assistance throughout this project.
DISCLAIMER
These proceedings are being distributed in the interest of increasing public understanding
and knowledge of the issues discussed at the workshop and have been prepared
independently of the workshop. Although the proceedings have been funded in part by
the United States Environmental Protection Agency under Contract No. 68-W-01-055 to
Alpha-Gamma Technologies, Inc., the contents of this document may not necessarily
reflect the views of the Agency and no official endorsement should be inferred.
11
-------
TABLE OF CONTENTS
Session II: Cleaning Our Coastal Waters: Examinations of the Benefits of
Improved Water Quality
The Value of Improvements to California's Coastal Waters: Results from
a Stated-Preference Survey
Nicole Owens and Nathalie B. Simon, U.S. EPA, National Center for
Environmental Economics 1
The Recreational Benefits of Improvements in New England's Water
Quality: A Regional RUM Analysis
Erik C. Helm, U.S. EPA, Office of Water; George R. Parsons, University
of Delaware; Tim Bondelid, RTI International 57
Valuing Water Quality Changes Using a Bioeconomic Model of a Coastal
Recreational Fishery
Matt Massey and Steve Newbold, U.S. EPA, National Center for
Environmental Economics; Brad Gentner, National Marine Fisheries
Service 77
Discussant
Bob Leeworthy, National Oceanic and Atmospheric Administration 79
Discussant
Nancy Bockstael, University of Maryland 83
Summary of Q&A Discussion Following Session II 91
in
-------
Draft Document: Do Not Quote, Cite or Distribute
The Value of Improvements to California's Coastal Waters:
Results from a Stated-Preference Survey
by
Nicole Owens and Nathalie B. Simon
National Center for Environmental Economics
United States Environmental Protection Agency
October 2004
Disclaimer: The opinions expressed in this paper are entirely those of the authors
and do not necessarily represent those of the USEPA.
No official Agency endorsement should be inferred.
1
-------
The Value of Improvements to California's Coastal Waters:
Results from a Stated-Preference Survey
Nicole Owens and Nathalie B. Simon1
National Center for Environmental Economics
United States Environmental Protection Agency
I. Introduction
The United States Environmental Protection Agency's (EPA) Office of Water is
responsible for regulating and monitoring national water quality. In order to make sound policy
decisions, policy makers need information on the benefits, costs, and other effects of alternative
options for addressing environmental problems. In the case of policies affecting water quality,
estimates of the public's willingness to pay for improvements in fresh water quality generally
begin with estimates provided by Mitchell and Carson (1993). This study, however, does not
address salt water areas.
The coasts and estuaries comprise a substantial part of our national resource base; these
coastal areas are depended upon for the aesthetic, economic, ecosystem, and recreational
services they provide. For example, gross annual income from coastal commercial fisheries is
close to $2 billion (1998$). However, coastal areas are also the most highly developed and
populated areas in the nation. This narrow fringe-comprising 17% of the contiguous U.S. land
area-is home to more than 53% of the nation's population. Further, this coastal population is
increasing by 3,600 people per day, giving a projected total increase of 27 million people
between now and 2015 (NOAA, 1998).
As coastal population has increased, the environmental quality of some of these areas has
declined or is threatened. Serious water pollution problems persist and, as such, many future
water policies will likely focus on coastal areas. The lack of estimates of the benefits of
improvements to these areas makes designing effective policies particularly difficult.
The purpose of this study is to estimate willingness to pay for water quality
improvements in California's coastal waters. Currently, States, tribes, and other jurisdictions
measure water quality by determining if water bodies are clean enough to support basic uses,
such as swimming, fishing, and aquatic life support. Thus, this study will estimate willingness
^rior to her death in 2000, Elizabeth McClelland was heavily involved in the project.
The authors also wish to thank Kevin Boyle, Don Dillman, George Parsons, and V. Kerry Smith
for their reviews of various drafts of the survey.
2
-------
to pay for improvements coinciding with these uses.2 The remainder of the paper provides some
information on EPA's valuation of water quality improvements, detailed descriptions of the
survey development process, information on the final version of the survey, a discussion of the
supporting theoretical model, as well as preliminary results.
II. Valuing Changes in Water Quality
Up to this point, changes in surface water quality have been valued using a Mitchell and
Carson (1993) study carried out in 1983. Mitchell and Carson determined respondents'
willingness to pay to improve water quality from boatable (the lowest rung on the heirarchical
water quality ladder developed by Resources For the Future) to fishable (sport fishing only, no
concern about consumability); and from fishable to swimable.3 The water quality ladder defined
these uses in terms of conventional pollutants (dissolved oxygen, BOD, TSS, etc.). However,
this study is not appropriate for valuing changes in coastal water for a number of reasons. First,
these values were obtained for inland fresh waters only and cannot be used to value coastal water
quality improvements. Second, toxic substances and nutrients were excluded from the study
since the water quality ladder only concerns conventional pollutants.
In addition, the water quality ladder is now an out-of-date conceptual framework. EPA
provides water quality information in The National Water Quality Inventory Report to Congress
(305(b) report). These documents provide information on the Nation's water quality, identifies
widespread water quality problems of national significance, and describes various programs
implemented to restore and protect our waters. The 305(b) Report's use designations are now
more complex. Not only do they deal with issues concerning the consumability of fish and shell-
fish (often constrained by toxic substances), they also deal with the health of aquatic
environments (affected not only by conventional pollutants but also by toxics and nutrients).
This increase in complexity obviates the use of a hierarchical ladder because different uses are
affected by different combinations and concentrations of the three main types of contaminants in
a non-hierarchical manner. Some contaminants affect some uses while others affect other uses.
For example, the presence of pathogens, a conventional pollutant, in water restricts swimming
and shellfish consumption, but would have little or no effect on the health of the aquatic
2Although we will ultimately hope to develop specialized surveys that elicit residents'
willingness to pay for improvements for each of the uses in each coastal state and one survey
that elicits inland residents willingness to pay for improvements in coastal waters, this paper
describes the development of only one of the state specific versions, California. The combination
of these surveys will provide us with coastal residents' willingness to pay to improve home-state
coastal water and inland residents willingness to pay to improve coastal water. We will not
capture coastal residents' willingness to pay to improve out-of-state coastal water. These values
may be elicited through another but similar project.
3EPA has also funded a freshwater quality valuation survey. The survey, developed
under the lead of Kip Viscusi, received final clearance from the Office of Management and
Budget in 2004.
3
-------
environment.
Additionally, the National Water Pollution Control Assessment Model, under
development by Research Triangle Institute for EPA, determines the number of river and
shoreline miles, and estuary and lake square miles, that would meet the various use designations
given different concentrations of contaminants. This model is expected to evaluate the
prospective changes in water coastal quality that would be brought about by different regulations
or other initiatives. Our survey has been developed to provide meaningful estimates of the value
of improvements to coastal waters, given the structure of the model; hence, water quality
improvements are described in percentage terms and separate values are obtained for each use.
III. Development of Survey Instrument
The survey instrument was developed over the course of two years through a series of
focus groups and protocol interviews primarily conducted in coastal states. The instrument has
also evolved dramatically as a result of feedback received from a peer review panel consisting of
Kevin Boyle, Don Dillman, George Parsons, and V. Kerry Smith.
Six initial focus groups were held in four areas of the country - Edison, NJ (1); Santa
Monica, CA (2); Washington, DC area (2) and one in a completely non-coastal environment (St.
Louis, MO). The first focus group was held in Edison, followed by those in Saint Louis, Santa
Monica, and Washington, DC respectively. These focus groups were held primarily to learn
what qualities the public values in coastal water, whether or not respondents are familiar with
certain terms, and to test early versions of descriptive text and valuation questions.
A. Initial Design
During initial phases of the project, it was thought that the survey would have multiple
sets of valuation questions, treating coastal water and estuarine water separately. Hence, one
important purpose of the first two focus groups was to gauge participant's familiarity with the
term "estuary." Discussions pertaining to coastal water concentrated on participants' uses and
experiences and, as expected, participants had no difficulty answering questions concerning
coastal water. When the discussion focused on estuarine waters, however, it became readily
apparent that participants had little understanding of the term and that the survey would need to
provide respondents with some background information on this topic. Specifically, participants
in the first focus group were generally unfamiliar with the term. Only one participant knew that
an estuary is an area of transition between fresh and salt water. Similarly, none of the
participants in the second focus group could define the term "estuary" and very few had heard of
the term. During these groups, a more detailed explanation of estuaries, their uses and locations
was given to participants by the facilitator. After these explanations were provided, participants
noted that they were familiar with and/or had visited several estuaries including Tampa Bay,
Chesapeake Bay, and Puget Sound. Many participants correctly noted that there is no "bright
line" between coastal water and estuarine water and it is virtually impossible to improve one
type of water without improving the other. Hence, later versions of the survey combine the two
types of water and discuss them at the same time.
4
-------
B. Early Survey Draft
The first draft of the questionnaire was written in March 2000. Appendix 1 contains
portions of this initial survey. This draft questionnaire was structured in four parts: introduction,
willingness to pay for water quality improvements, use descriptions and allocation, and
demographic questions. We intended to distribute this questionnaire to a nationally
representative sample using a phone-mail-phone administration mode. We planned to develop
two versions of this questionnaire — one for coastal states and one for inland states — to elicit
willingness to pay values for national coastal water improvements. An allocation question would
allow us to attribute values for different uses.
The introduction began with a warm-up question asking respondents to select the three
most important environmental problems in their state from a list. This was followed by a
definition of coastal waters and descriptions of estuaries and near-shore waters, giving features,
natural uses, recreational uses, commercial uses and examples. Each description was then
followed by several questions designed to elicit the respondent's familiarity with each water
type.
The second section of the questionnaire began with a second ranking question in which
respondents were asked to select the three most important coastal environmental problems in
their home state. A description of water quality in the United States followed, including a brief
explanation of the government's rating system and the largest sources of coastal water pollution.
Water quality was described as "good" if the water supports each of three uses: swimming,
production of fish and shellfish safe for consumption, and diversity of aquatic life. The
questionnaire then provided the number of miles of shoreline and the area classified as estuaries
in the U.S., together with the percentage of total coastal waters classified as "good" and "not
good". The three sources of coastal water pollution were identified as agriculture, industries,
and households and short lists of examples of the types of pollution contributed by each source
were given (e.g., runoff of crop fertilizers and pesticides for agriculture and runoff of lawn
fertilizers and pesticides for households).
A coastal water improvement program was then introduced preceded by a brief statement
indicating that if nothing is done conditions can at best be expected to remain the same but may
worsen due to increases in population. The program was described in rather vague terms but was
couched as being led by "agencies in charge of coastal water quality, fish and wildlife." The
program would clean up half of the "not good" areas so that only 20% remain categorized as
such. A broad list of possible clean-up efforts that could be included as part of the program were
provided, including activities such as removing sources of pollution, planting water-side
vegetation to absorb run-off, etc. This program description was then followed by a single
referendum question in which respondents were asked to state whether or not they would vote
for or against the program if it costs their household $X per month in the form of increased
federal taxes.
The third section of the questionnaire zeroed in on the three uses and attempted to elicit
the respondent's preferences across the three uses. A description of each of the three uses and
the hazards of using coastal waters classified as not supporting each particular use followed. The
5
-------
necessity for different types of clean-up efforts to remedy problems affecting each of the three
uses was then explained in an attempt to educate respondents to the fact that, in some sense,
clean-up efforts are separable across the three uses. An allocation question in which respondents
were asked to allocate each dollar spent on coastal water clean-up across the three uses rounds
out the third section of the questionnaire. Our thought was that this question would ultimately be
used to attribute values for coastal water improvements by use.
The fourth and final section of the questionnaire contains standard demographic
questions and questions regarding the respondent's recreational activities.
C. Protocol Interviews
Several versions of the draft questionnaire were tested through a series of four sets of
protocol interviews in coastal states on the east coast - Tampa, Portland (Maine), Baltimore, and
Richmond. In each location, at least two versions of the survey instrument were tested in 18
completed interviews. Respondents were provided with a copy of a survey and asked to
complete it to the best of their ability. Once the respondent completed the paper version of the
questionnaire, an interviewer went through the questionnaire with the respondent to discuss
his/her responses and thoughts regarding the questions.
Experience in Tampa, Florida
The protocol interviews held in Tampa marked the first time potential respondents
reviewed the survey instruments. For this occasion, we developed two versions of the survey
instrument with the most marked changes occurring in the willingness to pay scenario and the
description of the clean-up efforts by use. One version established a permanent increase in
monthly federal taxes to pay for the proposed coastal water improvement program. The other
described the increase in monthly taxes as occurring over a five-year period and provided much
more detail (including examples) of how and why improving coastal waters for each use would
require different clean-up efforts.
Reactions to the survey instrument were varied. While some respondents found the
survey interesting, others found it tedious and difficult to follow. It was apparent after reviewing
all of the comments that many changes needed to be made. In addition to numerous simplifying
wording changes, we identified areas requiring major revision.
A flaw in the survey concerned the description of coastal water uses and the allocation
question. A number of participants were confused by the allocation exercise and failed to
complete it properly. Those participants that understood the exercise, in general, summarily
rejected the separability of clean-up efforts, believing instead that by allocating more funds to
creating a diverse aquatic environment, they would ultimately be improving coastal water
conditions for swimming and production of fish and shellfish safe for human consumption as
well. Our attempts at providing convincing information to the contrary failed. Those
respondents who received the version with the additional information still felt that they would
get "more bang for their buck" if they allocated the entire sum to creating a diverse aquatic
6
-------
environment, even though they purportedly read and understood the additional paragraphs.
Experience in Portland, Maine
Armed with the feedback from Tampa, we revised our survey instruments substantially.
We dramatically changed the formatting of the survey, making it more "user friendly."
Complicated skip patterns were replaced with arrows and new headings were introduced to help
set the questions apart from the information presented. While these changes certainly made the
survey instrument more visually appealing, the more important changes were to the willingness
to pay scenario and the allocation question.
After much discussion, we decided to abandon the referendum style question devised for
the Tampa interviews and substantially revise the allocation question. While we felt it was still
important to obtain values for coastal water improvements by use, we decided to attempt to elicit
these values directly rather than through the allocation question. This change would require
reordering the information presented in the questionnaire so that a discussion of the various uses
preceded the willingness to pay questions. In order to convince respondents of the contribution
of households to the degradation of coastal waters, we added the following statement:
Much of the pollution affecting estuaries and near-shore waters is caused by the every day living
habits of the American people. Although the amount each household adds to the problem of
coastal water pollution may seem small, together residential communities have a large negative
impact on coastal water quality.
We also decided that we should attempt to obtain values for local improvements,
compared to national improvements, for each use in coastal states. Rather than provide
respondents with general information regarding the condition of coastal waters in the U.S., we
revised the background information preceding our scenario to include a table showing the
percent of coastal waters as well as the number of miles of coastal waters that were rated as good
for each of the three uses. We then replaced our referendum style question with a series of three
scenarios, each describing a program that would improve coastal waters for a particular use.
Each scenario was accompanied by a table showing the current coastal water conditions by use
and the expected improvement brought about by the program in question. The row containing
the use for which conditions would be improved was shaded to draw attention to the change. A
double-bounded dichotomous choice question eliciting the respondent's willingness to pay for
the improvement through an increase in income taxes rounded out the scenario.
Since this new version of the survey instrument would allow the estimation of
willingness to pay values for percent changes in coastal water improvements by use, we no
longer needed to rely on the allocation question to obtain these values. Still, we decided to
include a different allocation question for use in coastal states to elicit respondents' preferences
for local versus national coastal water improvement programs. For respondents in these states
living within 100 miles of coastal waters, we devised a question in which respondents would be
asked to allocate funds for improving each of the three uses across local (affecting coastal
conditions within 100 miles of their home) and national programs.
7
-------
We developed two basic versions of the new questionnaire for testing in Portland, Maine.
The most marked difference in the two versions was the inclusion of a willingness to pay
question in one version in which the program improved coastal water for all three uses.
The reactions to our two survey instruments from the participants in Portland were
encouraging. The respondents reacted positively to our new layout and simplified wording,
reporting generally that the questions were easy to read and understand. In addition, respondents
were much more willing to accept that households were large contributors (if not the largest
contributors) to the degradation of the coastal environment.
The feedback on our new willingness to pay questions was equally positive.
Respondents found the table format outlining the "before" and "after"conditions easy to follow
and comprehend. Even those respondents who admitted that they did not believe it was possible
to improve conditions for only one use without affecting conditions for all three uses reported to
focus on the highlighted use when answering the willingness to pay questions.
The allocation question continued to be a problem for some respondents. While several
respondents did not understand the allocation exercise at all, failing to perform any allocation
whatsoever, others were not certain whether they were to allocate funds across national and local
programs for each use or allocate funds across uses separately for national programs and then
local programs.
Experience in Richmond, Virginia
In spite of the progress we made in the versions tested in Portland, Maine, we came away
with three major concerns. First, we were concerned that respondents were not considering the
magnitude of the improvements in the willingness to pay questions but rather were focusing
simply on the use that was being affected by the program. While we were not prepared to
abandon the question format yet, we knew we needed to test the willingness to pay questions
more carefully in the next round of protocol interviews. Second, we were concerned that
respondents in coastal states were responding to the willingness to pay questions as if the
programs were affecting local coastal water conditions rather than national coastal water
conditions. This too would need closer scrutiny in the next round of interviews. Finally, we
recognized that we needed to revise the allocation question if we hoped to get meaningful and
useful responses. In addition to formatting changes, we realized that we would need to change
our definition of "local." We realized that respondents generally had difficulty discerning which
coastal waters were within 100 miles of their homes and we recognized that it would be difficult
for us to determine which households in our sample lived within 100 miles of coastal waters.
We found in our discussions with respondents that it was easier for them to envision and discuss
state coastal water conditions than those in a smaller area.
After considerable reflection, we decided to develop state-specific versions of the
questionnaire that could be used to elicit willingness to pay values for improvements to state
coastal waters. Although we realized that developing and administering separate state-specific
versions of the questionnaire would considerably increase the costs of survey administration, we
8
-------
remained unconvinced that our allocation question could obtain equivalent "local" values. The
state-specific versions of the questionnaire would be similar to the "national" version with the
primary differences being that the scenarios would provide "before" and "after" conditions
specific to a state and that no allocation question would be asked.
We also began to consider a more flexible mode of administration: the internet. Several
survey firms currently offer the option of internet survey administration. These firms have
recruited panels of potential respondents (in exchange for internet access) from which they are
able to draw representative samples. This administration mode allows great potential for
tailoring surveys to specific categories of respondents. As information about these survey firms
spread, we became more and more intrigued with the idea of conducting a computer-based,
internet survey as opposed to a mail survey. This survey mode would allow us greater flexibility
in question presentation and would allow us to easily tailor survey instruments to particular
states.
We developed and tested our first state-specific version of the survey instrument in
Richmond, Virginia along side a national version of the survey instrument containing a number
of formatting changes. Again, the survey instrument was met with generally positive feedback.
Our fears regarding the focus on the use affected rather than on magnitude of the improvement in
our willingness to pay questions was confirmed, however. Respondents reported not paying
much attention to either the percent change or the number of miles affected by each program.
Rather, they reported being concerned primarily with the use enjoying the improvement. The
formatting changes to the allocation question in the national version of the questionnaire
improved the performance of the question.
Experience in Baltimore, Maryland
Because our willingness to pay question continued to meet with difficulties, we decided
to change our approach yet again. Rather than present a program that affects only one use and
ask a double-bounded dichotomous choice question to elicit willingness to pay, we decided to
employ a conjoint approach. We modified our survey instrument so that in each scenario we
present the respondent with two programs, each affecting the percent of water considered "good"
for each use by a different amount. The effects of the two programs and the monthly costs to the
household for each are shown in a table accompanying each scenario. The respondent is then
asked to choose between the two programs with the status quo (no program) also provided as an
option. By varying the percent of miles affected by each program as well as the uses affected,
we will arrive at a willingness to pay for a percent improvement for each use. Each respondent
will be asked to answer four questions of this sort.
For our protocol interviews in Baltimore, Maryland, we again developed both a national
and a state specific version of the questionnaire. The conjoint approach was used in both
versions. In general, this approach met with great success. Respondents seemed to focus on all
aspects of the program - the uses affected, the magnitude of the changes, baseline conditions,
and cost - before answering.
9
-------
D. Computerized Versions of the Survey
Following our protocol interviews in Baltimore, Maryland, we made minor wording
changes to the survey and then began the process of having the our "pen and paper"survey
computerized. The benefits of using this mode of administration are numerous. First, using a
computerized format for the survey simplifies the process for respondents in that confusing skip
patterns are eliminated. The respondent sees only those questions that pertain to him.
Computerized surveys also create the potential for greater use of colorful and more meaningful
graphics to enhance the survey. In addition, the threat of interviewer bias is eliminated. Finally,
the administration time is significantly reduced in that completed interviews are automatically
downloaded to a database, simplifying the data cleaning process and allowing quick turn-around.
We decided to have the pilot survey administered by Knowledge Networks, a California-
based survey firm, to a random sample of approximately 600 California residents via WebTV.
Knowledge Networks maintains a large, national panel of respondents recruited through a
random process. Potential respondents are contacted by mail and provided introductory
materials about the company, together with a small monetary incentive for reading the materials.
Recipients are then contacted by phone and invited to enroll in the panel, along with other
household members. Panel members are provided the WebTV hardware and a monthly
subscription to the service which provides internet access. In exchange, respondents are asked to
complete surveys via the internet on a regular basis. Knowledge Networks maintains that its
panel is fairly representative of the population.4
Knowledge Networks has a sizable panel enrolled in California enabling us to conduct a
pilot survey in that state in addition to a full scale survey should changes need to be made to the
survey following the pilot. In order to test the computerized version fully, we decided to
conduct protocol interviews with panel members. We began tailoring our state specific version
to California, with the most dramatic changes to the survey taking place in four different areas.
First, because Knowledge Networks collects a variety of demographic questions on a regular
basis from panel members and makes this information available to its clients, we were able to
dramatically shorten the demographic section of the survey. Second, our peer review panel
suggested that we add questions from established national surveys in order to both gauge the
representativeness of our sample and match our respondents with respondents to these larger
surveys. In response to this suggestion, we added questions from the Panel Study of Income
Dynamics and from the National Survey on Recreation and the Environment. Third, we added
more detailed information on the quality of California's coastal waters and added more
information on the quality of coastal water in other states. Fourth, many of the initial questions
were re-ordered in order to improve the flow of the survey.
Once the survey was computerized, we conducted protocol interviews with
approximately 16 of Knowledge Network's panel members. Each participant took the survey as
4More information about Knowledge Networks can be obtained from the company's
web site: www. knowl edgenetworks. com.
10
-------
though in their own home and then went through a detailed debriefing session. Respondents
took approximately 30 minutes to complete the survey and most said it was quite interesting. It
was clear that respondents were able to understand all of the information provided in the survey.
Minor changes were made to the survey as a result of the protocol interviews. These
included eliminating an initial series of questions that asked respondents' opinions concerning a
variety of state issues or problems. It was initially thought that this would be a good
introductory question for respondents, but most found it difficult. This, along with the fact that it
increased the length of the survey while not providing us with vital data, led us to remove this set
of questions. Another area of the survey that needed improvement concerned the information
provided about other coastal states as a comparison. Data included for North Carolina was found
to be incorrect and many respondents noted that it was surprising that the water quality in North
Carolina was so low. Further, we needed to adjust the placement of information for states that
do not report water quality information to avoid confusion. Initially, the way in which we
conveyed this information suggested that these states had no water rated as good.
IV. Description of Survey Instrument
The pilot survey took place in California using the survey instrument described in more
detail below. The survey instrument is specific to the state of California and can be used to
estimate willingness to pay for water quality improvements by three specific uses: swimming,
production of fish and shellfish safe for human consumption, and support of diverse aquatic life.5
The California survey instrument is described in more detail below. In general, the
questionnaire is comprised of four distinct parts: an introductory section, a section focusing
specifically on California's coastal waters, a section containing the choice questions, and finally
a section containing standard questions about labor market activity. A hard copy of the survey
is provided in Appendix 2.
A. Part 1: Introduction
The first section of the survey provides respondents with a definition of coastal waters
and a detailed description of their natural, commercial and recreational uses. Following a
5Once analysis of the pilot data is complete and we are convinced of the adequacy of the
questionnaire, we hope to develop parallel versions of the survey instrument for the remaining
20 coastal states as well as a version for inland states. The coastal state versions of the survey
will elicit resident's willingness to pay for coastal water improvements within the state. The
inland version of the survey will elicit willingness to pay for coastal water improvements
generally. While we will not be able to gauge willingness to pay of coastal state residents for
improvements outside of their state of residence from the surveys we plan to develop, we
anticipate that the information gathered from these surveys will provide potentially useful
information for benefits analysis all the same.
11
-------
welcome statement, and a general definition of coastal waters, the respondent is provided with
use information in a simple table (see Figure 1). This table is followed by a map highlighting all
of the coastal states in the 48 contiguous states in the U.S. (see Figure 2).
The respondents' familiarity with coastal waters is then gauged through a series of
questions about recent trips to coastal waters and water recreation activities. For those
respondents who report visiting coastal areas in the last 12 months, detailed information about
the number of days participating in each of the activities is collected, including the number of
days in California. A number of these questions are borrowed from the National Survey on
Recreation, allowing direct comparison of results. Similar information is collected for
freshwater recreation activities.
B. Part 2: California's Coastal Waters
This section delves into a respondent's familiarity with pollution sources as well as his
perception of California's coastal water quality. In addition, it defines and describes the three
use categories and the water quality rating system employed by the EPA.
This section begins by showing a map of California's coastline with various coastal water
areas specifically indicated on the map (see Figure 3). Respondents are then asked about the
location of their primary residence with respect to coastal waters and the location of other
properties the household might own. Length of residence in California is also requested.
Respondents are then provided with a list of potential environmental problems that could
affect coastal waters and are asked to rate the seriousness of each problem for the state of
California on a scale from 1 to 5. Problems included in this list range from animal waste runoff,
to discharges and overflows from sewage treatment plants, to beach erosion. The list of
problems includes industrial, agricultural, and household sources of coastal water pollution and
is provided to each respondent in a randomized fashion. Following the list of potential coastal
water problems, respondents are asked to indicate which source (industry, agriculture or
households) is the largest source of coastal water pollution in California in their view. They are
also asked to report whether they believe coastal water quality has improved or not in the last
five years.
The water quality rating system used by federal and state governments is then described
to the respondents and information is given on the ratings California's coastal waters have
received for the three defined uses of swimming, production of fish and shellfish that are safe for
human consumption, and support of diverse aquatic life (including fish, shellfish, plants,
mammals, birds, etc. that live near aquatic environments). Information on California's coastal
waters is provided in pie charts, an example of which is shown in Figure 4. The information
provided is taken directly from The National Water Quality Inventory Report to Congress
(305(b) report).
Comparisons of California's water quality by use with that of other coastal states is
provided in a series of three bar charts — one for each use— showing the ranking of states by
12
-------
water quality level. An example of the bar charts is shown in Figure 5.
The final question in this section asks respondents to indicate which of the three uses is
the most important to them.
C. Part 3: Choice Questions
The third part of the questionnaire is comprised of the choice questions. Respondents are
presented with a series of five questions in which they are asked to select between two programs
to improve coastal water quality. In each choice set, respondents are also able to select the status
quo, should they find neither of the two programs satisfactory. Each of the two programs has an
associated household tax increase to cover the cost of implementation.
Information regarding water quality across three use definitions (swimming, production
of fish and shellfish deemed safe for human consumption, and the support of diverse aquatic life)
under each program, including the status quo, is provided in tabular format together with the cost
to each household for each program. Color is used in the table to help respondents distinguish
between the three alternatives. The programs differ, not only in the level of household tax, but
the degree to which they improve water quality across the three use definitions. A sample
question is provided in Figure 6.
The questions are structured in such a way as to facilitate comparison between the
programs with at most two water quality attributes varying at different levels across the two new
programs being introduced. In some instances, however, respondents are asked to choose
between two programs that offer varying magnitudes of uniform changes across uses. Each of
every respondent's five responses will be treated as a separate observation.
D. Part 4: Demographic Information
The fourth and final section of the survey is comprised of demographic questions. The
series of demographic questions required in our survey instrument is curtailed due to the
availability of this information from Knowledge Networks. As noted above, Knowledge
Networks collects and routinely updates standard demographic information on each panel
member and makes this available to its clients. In so doing, burden on the panel members is
reduced and the length of the survey shortened.
V. Economic Model
In choice experiments such as ours, individuals are typically asked to choose from
alternatives with varying attributes from a choice set. In making their selections, respondents
weigh the importance of the different attributes and implicitly trade one characteristic for
another, selecting the alternative that provides them with the greatest utility. The probability of
choosing any specific alternative can then be modeled straightforwardly using standard random
utility models. These types of models have been used to ascertain the value of beaches (Parsons
13
-------
et al. 2000), water quality in freshwater lakes (Needelman and Kealy, 1995; Bockstael,
Hannemann and Kling, 1987), and woodland caribou habitat enhancement (Adamowicz et al.,
1998).
A. Basic Model
Consider the following representation of an individual's utility associated with program i:
Ul = fixI+sI (1)
where x, is a vector of explanatory variables, including program attributes, cost of the program
and other individual characteristics. Effects of unobserved variables are captured by eL a random
term distributed as iid extreme value (weibull). A decision maker will choose program i from his
choice set/if that alternative provides greater utility than the other two alternatives: U; > Ujfor
all j*i.
The probability that an individual chooses program i from set J is given by:
exp(tfx)
Pr(0 = y 7Z 7 (2)
2, exp(^)
where the numerator is the exponential of the utility associated with program i and the
denominator is the sum, over all programs in the choice set, of the exponential utility associated
with each possible program. These probabilities are then entered in a standard likelihood
function of the following form:
i=nn pri,r <3)
n=1 j=1
where 6in =1 (for ie J) if individual n selects program i and =0 otherwise. Parameters are
selected so as to maximize L.
One advantage of choice experiments such as ours relative to traditional contingent
valuation is that they allow researchers to infer the value of the specific attributes in addition to
situational changes. Random utility models do not allow direct estimation of the value of
particular attributes; rather, the researcher must estimate the probability that a specific alternative
will be selected and can then infer the value of the various characteristics using the estimation
results. Once estimated, the model results can be used to estimate welfare changes associated
with the improvement or decline of specific attributes.
Ultimately, we are interested in estimating the welfare changes associated with
14
-------
improvements in water quality for the three use definitions in California. The gain in consumer
surplus associated with an improvement in the quality of water for swimming, for instance, can
be calculated as the change in expected utility divided by the individual's marginal utility of
income given by
_ 1"E ^,exp(X,V) - ln£ "„exp(X,/;)
0
r tax
where Ptax is the marginal utility of income estimated in the logit model, is the vector of water
quality measures under the status quo, and X;* is the vector of water quality with improved
quality of waters for swimming.
VI. Data and Preliminary Results
The survey was fielded to 746 Knowledge Networks panel members in two waves, the
first (a pretest) on June 4, 2004 and the second on July 1, 2004. Data collection continued
through August 1, 2004. The pretest was fielded to 141 Knowledge Networks panel members.
In late June, we examined respondents' answers to survey questions and in addition to precoding
several open ended questions, determined that no changes needed to be made to the conjoint
design in the survey instrument. We received 606 completes, yielding a completion rate of
81%. Table 1 contains descriptive statistics for the full dataset.
15
-------
Table 1
Descriptive Statistics
n=606
Variable
Mean
Std Dev
Min
Max
Male
0.50
0.50
0
1
Age
43.16
15.54
18
96
Household size
2.66
1.41
1
10
Income
52681.00
40095
2500
187500
Black
0.05
0.22
0
1
Hispanic
0.28
0.45
0
1
Other minority
0.14
0.35
0
1
Children
0.30
0.46
0
1
Recreational swimmer in past 12 months
0.25
0.43
0
1
Recreational fisher in past 12 months
0.10
0.30
0
1
Recreational boater in past 12 months
0.16
0.37
0
1
Observed wildlife in past 12 moths
0.49
0.50
0
1
Eat seafood at least one time per month
0.60
0.49
0
1
Preliminary conditional logit model results are promising and consistent with
expectations (Table 2). Regarding the choice specific attributes, as the cost associated with the
programs increases, respondents are less likely to choose a program over status quo conditions.
In addition, as the miles good for swimming, fishing, and aquatic life support associated with the
programs presented to respondents increases, respondents are more likely to choose a program
over the status quo. The interpretation of the remaining variables in the regression is slightly
different as the variables represent individual specific attributes. As income increases
respondents are more likely to move away from the status quo, males are more likely to choose
the status quo. As age and household size increase, respondents are more likely to choose a
program. Only three of the included participation variables are significant - recreational fishers
and boaters, and those eating seafood are more likely to choose a program over the status quo.
As these results are extremely preliminary, we plan to continue exploring alternative
models, to estimate elasticities of probabilities with respect to program cost, and to develop
16
-------
willingness to pay estimates for improvements in the percent of miles good for each of the three
uses explored in the survey.
Table 2
Preliminary Model Results
Conditional Logit
n=606
Variable
Estimate
T-Value
Cost
-0.008***
-8.93
Miles good for swimming
q Q4***
4.54
Miles good for fishing
0.05***
5.53
Miles good for supporting aquatic life
Q 2 2***
12.63
Male
q 23***
2.60
Age
0.01***
5.55
Household size
0.10***
3.22
Income
-3.04 10-6***
-2.73
Black
0.26
1.21
Hispanic
-0.09
-0.91
Other minority
0.20
1.50
Recreational swimmer in past 12 months
-0.18*
-1.63
Recreational fisher in past 12 months
-0.01
-0.04
Recreational boater in past 12 months
-0.22*
-1.63
Observed wildlife in past 12 moths
-0.8
-0.89
Eat seafood at least one time per month
-0.17*
-1.81
*** significant at 1%
* significant at 10%
Log Likelihood -2402
17
-------
More Information on Coastal Waters:
Coastal waters
may have:
shallow waters, marshes, saridy beaches, mud arid sand
flats, rocky shores, oyster reefs, river deltas, tidal pools,
sea grass beds and swamps.
Natural uses
include:
food, shelter and breeding grounds for many fish,
shellfish, mammals and shorebirds.
Recreational
uses include:
boating, fishing, shell-fishing, swimming, snorkeling and
bird-watching.
Commercial
uses include:
ports and marinas supporting shipping and industrial
uses; breeding grounds for some commercial fish and
shellfish.
Examples of
coastal waters
are:
the water along Chesapeake Bay, Clearwater Beach
(Florida), Ocean City (Maryland), Venice Beach
(California), Galveston Bay, Puget Sound, San
Francisco Bay, Tampa Bay and lots of smaller bays and
inlets where fresh waters and saltwaters mix.
-------
Figure 3 Map of California
/ |
|
CALIFORNIA
^SACRAMEj;
Berkeley
San Francisco®@Qgitfand
San Francisco Bay
San Jose®
* ©
Monterey Bay ¦¦ Salinas
North Pacific simi Valley
Oceaii w ,
aLos Angeles
J ' ' % »
Irfine^-*1^ „v\
Santa Monica Bay f
San Diego'
a.
J
Figure 2: Map showing states with coastal waters in the contiguous states in the U.S.
WA
OR
ME
NH
NY
MA
CTri
f
CA
North
Pacific
Ocean
>
. S
L'N'n jiP S'
n
TX
MS
LA
NJ
MDde
VA
NC
SC North
m ga Atlantic
Ocean
FL
Gulf of
Mexico
19
-------
Figure 4: Sample Pie Chart Showing Coastal Water Quality of California Waters by Use
Good /
64% /
Production offish and shellfish that
are safe to eat
Not
Good
36%
20
-------
Figure 5:
Sample Bar Chart Comparing the Quality of California's Coastal Waters
with Other Coastal States
This chart shows, on average, how California's coastal waters
compare to other states for Production of Fish and Shellfish
that are safe to eat:
1W%
m
70H
SB
m
m
2#*
1GS
OH
Percent of Coastal Waters "Good" for Fish and Shellfish
Consumption
m m 94% *»
srft am „ -
55%
6n m 64% m m
69*
73S
76S
34%
0% PS 2* K
. rn , n
III I l l
111
«A Nh DE OR CT FL TJ NJ WA CA MS Rl SC AL GA ME NC NY LA MO VA
tjurrifaafs are a best repressnla'SDn ofttie manllQfiria infonnarlion arailaale from lire iriG?Ki(tual slates.
Continue
21
-------
Figure 6: Sample Choice Question
Your Three Choices and How They Would
Affect the Quality of California's Coastal Waters
Percent of California's Coastal Waters Rated as
"Good"
Present
Conditions
Program 1
(Conditions
after 3 years)
Program 2
(Conditions
after 3 years)
Swimming
42% of miles
are good
5% gain to
47% good
0% gain to
42% good
Fish and shellfish
safe for eating
64% of miles
are good
5% gain to
69% good
0% gain to
64% good
Habitat to support a
large number of
different kinds of fish,
birds, mammals and
plants
52% of miles
are good
6% gain to
57% good
0% gain to
52% good
Yearly Tax Change for
your household
(permanent tax)
No increase in
taxes
Your taxes
increase by $80
per year
Your taxes
increase by $40
per year
Which one of the options listed in the table above would you
choose?
Select one answer only
Present Conditions: No change in your taxes, and the percent of coastal
waters that is good for each purpose stays the same as it is now
Program 1 Your taxes increase $80 per year to getthe improvements
showr under this program
Program 2: Your taxes increase $40 per year to getthe improvements
s h own u n d e r th i s p ro g ra m
# Don't know
22
-------
References
Adamowicz, Wiktor, Peter Boxall, Michael Williams, and Jordan Louviere (1998), "Stated
Preference Approaches for Measuring Passive Use Values: Choice Experiments and
Contingent Valuation," American Journal of Agricultural Economics, 80(l):64-75.
Bockstael, Nancy E., W. Michael Hanemann, and Catherine L. Kling (1987), " Estimating the
Value of Water Quality Improvements in a Recreational Demand Framework," Water
Resources Research, 23(5): 951-960.
Mitchel, Robert and Richard Carson (1993), "The Value of Clean Water: The Public's
Willingness to Pay for Boatable, Fishable and Swimmable Quality Water," Water
Resources Research, 29.
National Oceanic and Atmospheric Administration (1998), "State of the Coast."
http://state-of-coast.noaa.gov/
Needelman, Michael S., and Mary Jo Kealy (1995), "Recreational Swimming Benefits of New
Hampshire Lake Water Quality Policies: An Application of a Repeated Discrete Choice
Model " Agricultural and Resource Economics Review, April: 78-87.
Parsons, George R., D. Matthew Massey, Ted Tomasi (2000), "Familiar and Favorite Sites in a
Random Utility Model of Beach Recreation," Marine Resource Economics, 14: 299-315.
Train, Kenneth E. (2000), "Mixed Logit Models for Recreation Demand," in Valuing Recreation
and the Environment, Joseph A. Herriges and Catherine L. Kling (eds.). Edward Elgar:
Northampton!, MA.
23
-------
Appendix 1: Portions of Initial Draft
Water Quality in the United States
The government rates water as either good or not good.
Water quality is good if the ocean shoreline or estuary
is a safe place to swim,
has fish and shellfish that are safe to eat, and
supports many kinds of plants, fish, and other aquatic life.
Water quality is not good if the ocean shoreline or estuary
is an unsafe place to swim due to pollution,
has fish and shellfish that are unsafe to eat due to pollution, or
supports only a small number of different kinds of plants, fish and other aquatic life.
Of our nation's more than 58,000 miles of ocean shoreline and 34,000 square miles of estuaries,
60 % are rated "good"
40% are rated"not good"
Much of the pollution affecting estuaries and near-shore waters is caused by the every day living habits of the American people.
Some of the largest sources of pollution include:
Agriculture /
Runoff of crop fertilizers and pesticides
Runoff of animal waste from fields and pastures
Overflows from animal waste holding areas
Industries
Overflows from sewage treatment plants
• Discharges from industrial processes
Households^
Runoff of automobile grease and oil
Runoff of lawn fertilizers and pesticides
Overflows from septic systems
Runoff of paints and chemicals
Seepage of household waste from landfills
24
-------
Existing fines on industry and taxes support current water quality levels, but in order to improve
the quality of the water, additional funds are needed. If nothing more is done, conditions can, at
best, be expected to remain the same but may worsen due to increased population.
Before the Program
After the Program
Suppose a program were proposed where the agencies in charge of coastal water quality, fish,
and wildlife were to clean half of the "not good" areas so that the percent "good" would be 80%.
The program would likely take three years before noticeable results could be seen.
Methods for clean-up depend upon the exact problem but would include things like:
removing sources of pollution
planting water-side vegetation to absorb run-off
controlling runoff and seepage from areas with pollution
protecting sensitive environmental areas.
5-1 Keeping in mind that your household would have less money each month to spend on other
things, would you vote for or against the program if the cost to your household would be a
permanent $100 per month in increased federal taxes (that is, $1200 per year). (Please check
one.)
25
-------
For O Against O Don't know
What is the maximum you would be willing to pay per month for this program?
Please take a few minutes to tell us why you voted the way you did.
-------
Ways to Improve Coastal Water
For those areas of the coast where the water quality is "not good," the clean-up efforts in the
program we talked about above will depend upon the type of water quality problem that exists,
and the importance that persons like yourself place on various uses.
For those areas of the coast that have water that is "not good," there are 3 specific ways that our
coastal waters could be improved:
Making water swimmable,
Making fish and shellfish safe to eat,
Creating a diverse environment.
Making coastal water swimmable
Making coastal water swimmable means getting rid of the types of
pollutants that can make people sick when they go swimming.
Sometimes direct contact with the polluted water can cause illnesses
such as stomach illnesses, earaches or infections.
6-1 Have you ever heard of coastal beaches being closed to swimmers because of polluted
waters? (Please check one.)
O Yes O No O Don't know
If no or don't know, please skip question 6-2. If yes, please answer question 6-2.
6-2 Has a beach you were visiting ever been closed to swimmers because of polluted waters?
(Please check one.)
O Yes O No O Don't know
Making fish and shellfish safe to eat
27
-------
Making fish and shellfish safe to eat means getting rid of the types of pollutants that build
up in the bodies of some fish and shellfish that can make people sick in the short and long
run. Eating raw, contaminated fish or shellfish can cause stomach illnesses. Eating large
amounts of contaminated fish or shellfish over a long period of time (even when cooked)
can cause other long-term serious health problems such as cancer and liver disease.
6-3 Have you ever heard about fish advisories that limit the amount of coastal fish or shellfish
that should be eaten because of polluted waters? (Please check one.)
O Yes O No O Don't know
6-4 Have you ever limited the amount of coastal fish or shellfish you've
eaten or refrained from eating coastal fish or shellfish as a result of a fish
advisory issued because of polluted waters?
(Please check one.)
O Yes O No O Don't know
Creating a diverse environment in the water
Creating a diverse environment means getting rid of the types of pollutants that keep
many plants, fish, and other life from living in water. Although some fish and plants can
live in polluted waters, cleaning up the waters will allow a greater number of different
types of fish and aquatic life to thrive.
6-5 Have you ever been to an estuary or near-shore area that is a "wildlife refuge," "protected
wetland," "bird sanctuary" or similar restricted access area? (Please check one.)
O Yes O No O Don't know
Clean-up effort
Cleaning up coastal waters for each of these purposes requires a different kind
of effort. While some efforts will affect more than one use, each of the uses
must be approached separately to affect a change for that use. This is because the types of
pollutants that make swimming unsafe are different from the types of pollutants that make fish
and shellfish unsafe to eat. These are different from the pollutants that keep the environment in
the water from being diverse.
28
-------
This means it is possible to improve conditions in coastal water so that it is swimmable but this
same water may still not be able to support a diverse environment. This same water may also still
not be good enough to support fish and shellfish that are safe for people to eat.
It is also possible to improve conditions in the coastal waters and estuaries so that the fish and
shellfish caught in these waters are safe to eat, without increasing the kinds of fish and aquatic
life that are able to survive in the waters. These same waters may still not be safe for humans to
swim in even though the fish caught in these waters are safe to eat.
Rating of Uses
7-1
we have
clean-
percent
Please take a moment to think about the three ways of improving coastal water
discussed. In your opinion, how much of each dollar spent on coastal water
up should go to each of the three improvement categories? (Please write
in box.)
Improvement Category
Percent of $1 spent on
clean-up
Making coastal water swimmable
%
Making the fish and shellfish that
live in coastal water safe to eat
%
Creating a diverse environment in
coastal water
%
Total (should add to 100%)
100 %
29
-------
APPENDIX 2: QUESTIONNAIRE
California Survey
Thank you for agreeing to help us by completing this survey. This survey asks for your opinions about
coastal waters in California. Your opinions and those of others completing this survey are very
important and may be used to help prioritize programs that may affect your local area. There are no
right or wrong answers; we are simply interested in your opinions and your experiences.
OMB Approval No: 2090-0024
Approval Expires 01/31/2005
According to the Paperwork Reduction Act of 1995, an agency may not conduct or sponsor, and a person is not
required to respond to, a collection of information unless it displays a valid OMB control number. The valid
OMB control number for this information collection is xx-xx. The time required to complete this information
collection is estimated to average between 20 and 30 minutes.
30
-------
We'll like lo skirl h\ ui\ inu yon some information ahi>nl eotisitil \\tilers tinil tiskinu tihoul your
experiences IK eotisitil \\ tilers we metin llie shtillow still \\ tilers wilhin l\\o miles of shorelines of
oeetins. htiys. setis. or mills incluilinu lhe tiretis where lYeshwtiler ri\ers mix with stillwtiler
The next screen eonitiins more inlbrniiilion tihoul eotisitil wtilers
More Information on Coastal Waters:
Coastal waters may have:
Natural uses include:
Recreational uses include:
Commercial uses include:
Examples of coastal waters are:
shallow waters, marshes, sandy beaches, mud and sand flats,
rocky shores, oyster reefs, river deltas, tidal pools, sea grass
beds and swamps.
food, shelter and breeding grounds for many fish, shellfish,
mammals and shorebirds.
boating, fishing, shell-fishing, swimming, snorkeling and bird-
watching.
ports and marinas supporting shipping and industrial uses;
breeding grounds for some commercial fish and shellfish.
the water along Chesapeake Bay, Clearwater Beach (Florida),
Ocean City (Maryland), Venice Beach (California), Galveston
Bay, Puget Sound, San Francisco Bay, Tampa Bay and lots of
smaller bays and inlets where freshwaters and saltwaters mix.
31
-------
Of the 48 contiguous states in the US, 21 have coastal waters. These states are shown in yellow on the map.
Q1 In the past 12 months, have you visited any coastal waters for recreation or pleasure in one or
more of the 21 coastal states shown on the map? (select one answer only)
~ Yes (skip to Q3)
~ No
~ Don't know (skip to Q5)
Q2 Have you ever visited any coastal waters in any of the 21 coastal states shown on the map? (select
one answer only)
~ Yes
~ No (skip to Q5)
~ Don't know (skip to Q5)
Q3 Does your household own a boat that is used primarily on coastal waters? (select one answer only)
~ Yes
~ No (skip to Q5)
~ Don't know (skip to Q5)
32
-------
Q4 For which activity do you use your boat the most on coastal waters? (select one answer only)
~ Recreational fishing
~ Recreational boating
~ Commercial fishing
~ Chartered boat rides
~ Chartered fishing trips
~ Other (please specify )
~ Don't know
Q5 How often do you eat seafood? (select one answer only)
~ More than 3 times per week
~ 2-3 times per week
~ 1 time per week
~ 2-3 times per month
~ 1 time per month
~ Less than once per month
~ Never (skip to Q8)
~ Don't know (skip to Q8)
Q6 About how much money per month do you spend on seafood that you
personally eat? (select one answer only)
~ Less than $5
~ Between $5 and $9.99
~ Between $10 and $19.99
~ Between $20 and $29.99
~ Between $30 and $39.99
~ Between $40 and $49.99
~ More than $50
~ Don't know
Q7 Does any of the seafood you eat come from California waters? (select one
answer only)
~ Yes
~ No
~ Don't know
Q8 Have you ever heard about fish advisories that limit the amount of
coastal fish or shellfish from California that one should eat because of
pollution? (select one answer only)
~ Yes
~ No
~ Don't know
33
-------
Q9 If Q1 = No or Don't know, skip to instructions before Q9h
[For the following activities in Q9, if 0 days or "Don't recall" is selected, skip to the next activity. If 0 days or
"Don't recall" is selected for trips in California, skip to the next activity.
The next few questions are about your coastal water recreation activities over the last year. During the last 12 months, on how
many different days did you personally participate in each of the following activities? (select one answer from each row in the
grid) (Randomize order)
Number of Different Days in the Last 12 Months
Don't
More recall
0 1-2 3-5 6-10 11-20 than 20 number of
days days days days days days days
a. Fish in coastal waters? (up to 2 miles
from shore) ~ ~ ~ ~ ~ ~ ~
ai If a >0 then ask:
How many of these days were single-
day trips in California? ~ ~ ~ ~ ~ ~ ~
If single day trips in California >0 then
ask:
Thinking about your most recent
single-day fishing trip to coastal
water in California, what was the
name of the coastal fishing site you
visited on this most recent trip?
Name
What is the name of the city or town
closest to (Name)?
City/T own
About how many miles is (Name)
from your home?
Miles
About how long did it take you to get
from your home to (Name)?
Hours Minutes
34
-------
Number of Different Days in the Last 12 Months
Don't
More recall
0 1-2 3-5 6-10 11-20 than 20 number of
days days days days days days days
(if ai>0, then ask) Did you eat any of
the fish you caught on this trip?
~ Yes
~ No, didn't eat any fish
~ No, didn't catch any fish
~ Don't know
b. deep-sea fish (more than 2 miles from
shore)? ~ ~ ~ ~ ~ ~ ~
If b>0 then ask:
How many of these days were single-
day trips in California? ~ ~ ~ ~ ~ ~ ~
c. boat or sail on coastal waters? ~ ~ ~ ~ ~ ~ ~
If c>0 then ask:
How many of these days were single-
day trips in California? ~ ~ ~ ~ ~ ~ ~
If single day trips in California >0 then
ask:
Thinking about your most recent
single-day boating trip to coastal
water in California, what was the
name of the coastal boating site you
visited on this most recent trip?
Name
What is the name of the city or town
closest to (Name)?
City/T own
About how many miles is (Name)
from your home?
Miles
About how long did it take you to get
from your home to (Name)?
35
-------
Number of Different Days in the Last 12 Months
Don't
More recall
0 1-2 3-5 6-10 11-20 than 20 number of
days days days days days days days
Hours Minutes
visit a beach on coastal waters for any
outdoor recreation activities?
If d>0 then ask:
How many of these days were single-
day trips in California?
swim in coastal waters?
If e>0 then ask:
How many of these days were single-
day trips in California?
If single day trips in California >0 then
ask:
Thinking about your most recent
single-day swimming trip to coastal
water in California, what was the name
of the coastal swimming site you
visited on this most recent trip?
Name
What is the name of the city or town
closest to (Name)?
City/T own
About how many miles is (Name)
from your home?
Miles
About how long did it take you to get
from your home to (Name)?
Hours Minutes
observe wildlife near coastal waters?
If £>0 then ask:
How many of these days were single
n n n n n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n n n n n
36
-------
Number of Different Days in the Last 12 Months
Don't
More recall
0 1-2 3-5 6-10 11-20 than 20 number of
days days days days days days days
day trips in California?
37
-------
Q9g [ask only if Q1 = 1] Thinking about the number of days you spent participating in each of the
coastal water activities we asked about, would you say that this was a typical recreational year for
you? (select one answer only)
~ Yes (skip to instructions before Q9h)
n No
~ Don't know (skip to instructions before Q9h)
Briefly explain why the past 12 months were not a typical recreational year for you?
This next set of questions asks about freshwater recreation activities. By "freshwater" we mean waters in inland lakes, ponds, rivers,
streams, etc., excluding areas where freshwaters and saltwaters mix.
During the last 12 months, on how many different days did you personally participate in each of the
following activities? (select one answer from each row in the grid) (Randomize order)
More Don't
than recall
0 1-2 3-5 6-10 11-20 20 number
days days days days days days of days
h. fish in a freshwater lake, pond, river or
stream? ~~~~~~ ~
i. boat or sail on a freshwater lake, pond,
river or stream? ~~~~~~ ~
j. visit a beach on a freshwater body for any
outdoor recreation activities? ~~~~~~ ~
k. swim in a freshwater lake, pond, river or
stream? ~~~~~~ ~
1. observe wildlife near a freshwater lake,
pond, river, or stream? ~~~~~~ ~
38
-------
Now we would like to ask you about coastal waters in California, Here is a map showing the California
coast.
OREGON ll IDAHO
Q10 Is your primary residence located within 10 miles of coastal waters? (select one answer
only)
~ Yes
~ No
~ Don't know
Qll Aside from your primary residence, does your household own any property in California within
10 miles of coastal waters? (select one answer only)
~ Yes
~ No (skiptoQ13)
~ Don't know (skiptoQ13)
39
-------
Q12 What other type of coastal property does your household own? (select aU that apply)
~ Residential, single family home
~ Residential, condominium — one unit
~ Residential, condominium — multiple units
~ Residential, apartment building
~ Commercial
~ Don't know
Q13 How long have you lived in California? (select one answer only)
~ Less than 1 year
~ 1-5 years
~ 6-10 years
~ 11 -20 years
~ Over 20 years
~ Don't know
Q14 The next question is about problems that may be affecting coastal waters in California. Please
rate the seriousness of each problem by selecting a number from 1 to 5, with 1 meaning "not at
all serious" and 5 meaning "very serious." (select one answer from each row in the grid)
(Randomize order)
How would you rate the seriousness of
each of the following problems in
California in terms of its impact on coastal
waters?
pesticide and fertilizer runoff from farm areas
discharges and overflows from sewage
treatment plants
discharges from oil refineries and other
industrial waste
seepage of waste from landfills
storm water runoff from roads and highways
pollution from commercial shipping (including
oil and chemical spills)
pollution from recreational boats (including oil
and gasoline spills and debris)
litter and other debris
animal waste runoff from farms and ranches
beach erosion
Not at all
Serious
1
2
3
4
Very
Serious
5
Don't
know
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
40
-------
other (please specify
) n n n n n
n
Q15 Most coastal water pollution comes from one or more of the following sources. Which one of
these do you believe is the largest source of coastal water pollution in California? (select one
answer only) (Randomize order)
~ Agriculture sources including runoff of crop fertilizers and pesticides, runoff of animal
waste from fields and pastures, and overflows from animal waste storage areas.
~ Industry sources including overflows from sewage treatment plants, discharges from
industrial processes, absorption of waste into the soil at landfills, accidents, and spills.
~ Household sources including runoff of automobile grease and oil, runoff of lawn fertilizers
and pesticides, overflows from septic systems, runoff of paints and chemicals, and
absorption of waste into the soil at landfills.
~ Don't know
Q16 Now, we would like to ask you about coastal waters in California.
Would you say that in the last five years California's coastal waters have gotten cleaner, stayed
the same, or gotten dirtier? (select one answer only)
~ Gotten cleaner in the last five years
~ Stayed the same in the last five years
~ Gotten dirtier in the last five years
~ Don't know
Q17 Which one of the following is your main source of information on the condition of California's
coastal waters? (select one answer only)
~ Newspapers
~ Magazines
~ Television broadcast news
~ Internet
~ Personal experience
~ Friends and family
41
-------
The federal government and states use information on pollution concentrations to rate the quality
of coastal waters for different uses.
Coastal water is rated as "good" or "not good" based on its ability to support the following three uses:
recreational swimming
the production of fish and shellfish that are safe for people to eat
the ability to support a large number of different kinds of fish, birds, mammals and plants.
The following describes what it means for water to be "good" for each use.
• Recreational swimming:
If water is "good" for recreational swimming it means that it is free from the types of pollutants
that make people sick (stomach illnesses, earaches, rashes or infections, and in rare cases long-
term health effects) when they go swimming. In other words, if water is rated "good" for
swimming, people can swim in the water without risk of illness.
• Fish and shellfish safe for eating:
If water is rated "good" for fish and shellfish consumption it means that the fish are free from
contamination that can make people sick. Some types of pollutants build up in the bodies of
some fish and shellfish and can cause stomach illnesses and other health problems in people.
• Large number of different kinds of fish, birds, mammals and plants:
If water is rated "good" for supporting large numbers of different kinds of life, it means that the
water is free from the types of pollutants that keep many fish, birds, mammals and plants from
living in water. In other words, "good" water allows a greater number of different kinds of fish
and aquatic life to thrive.
For each of the uses above, water is considered "not good" if it does not support the use all of
the time because of pollution.
42
-------
The pie charts below show the percent of California coastal waters that, on average, is "good"
and "not good" for each of the three uses.
Swimming
Production of fish and shellfish that are safe
to eat
Supports a large number of different kinds of
fish, birds, mammals and plants
Q18 For which of the three uses we just described is water quality the most important to you? (select
one answer only)
~ Recreational swimming
~ Fish and shellfish safe for eating
~ Large number of different kinds of fish, birds, mammals, and plants
~ Don't know
43
-------
This chart shows, on average, how the water quality of California's coastal waters compare to the water quality of other states for Swimming:
Percent of Coastal Waters Where Water Quality is "Good" for
Swimming
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
87% 87%
92% 94%
QCo, 070/ 98% 98% 99% 100% 100% 100% 100% 100% 100% 100%
y o /o /o — — — — —
69% 70%
0/ 73%
42% 42%
CA DE M
A LA FL NC MS SC NY Rl CT VA NJ WA GA TX ME MD NH OR AL
Numbers are a best representation of the monitoring information available from tne individual states.
44
-------
This chart shows, on average, how the water quality of California's coastal waters compare to the water quality of other states for the Production of Fish and Shellfish that
are safe to eat:
Percent of Coastal Waters Where Water Quality is"Good" for
Fish and Shellfish Consumption
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
87% 87%
91o/o 93% 94%
97%
55%
61%
63% 64% 64% 65% 67% 6J!°/o
73% 76%
34%
0% 0% 2%
5%
MA NH DE OR CT FL TX NJ WA CA MS Rl SC AL GA ME NC NY LA MD VA
Numbers are a best representation of the monitoring information available from the individual states.
45
-------
This chart shows, on average, how the water quality of California's coastal waters compare to the water quality of other states for supporting large numbers of different
kinds of fish, birds, mammals and plants:
Percent of Coastal Waters Where Water Quality is"Good" for
Supporting Aquatic Life
100%
90%
80%
70%
60%
50%
40%
30%
20%
10% 3%
0% ¦=»
85% 87% 88%
92% 92%
99% 100% 100% 100%
33%
39%
48%
52%
55%
60%
66% 69%
73%
8%
JZL
DE LA WA MD Rl CA MA CT SC NJ FL VA NC GA MS TX NY NH ME AL OR
(no
Numbers are a best representation ofthe monitoring information available from the individual states. data)
46
-------
Now we would like to know whether or not you would support a program that increases the percent of
California's coastal waters that are "good" for swimming, eating fish and shellfish, and supporting a large
number of different kinds of wildlife.
Currently, taxes on households, industries, and agriculture as well as fines on agriculture and industry
pay for the programs that support current water quality. If nothing more is done, the quality of coastal
waters will remain about the same.
To improve the quality of the water, new programs will be needed as well as new funds to pay for them.
On the next several screens, we will give you information on programs that improve California's coastal waters.
You will be asked to compare two programs at a time with the present conditions and to select which program,
if any, you prefer.
The table on the next screen shows the percent of coastal waters that will improve under each of two new
programs and the taxes required from each household to fund the new programs.
As you make your choice, please keep in mind the following:
Even though each program improves coastal waters in different ways, both would take three
years before the improved water returns to "good."
Neither program would improve the quality of freshwater lakes and rivers or coastal waters in
other states where swimming and fishing may take place or where a healthy aquatic
environment may exist.
Selecting a program means that your household would have less money to spend on other
things.
It is already possible in some places in California to swim in and eat the fish from the same
coastal waters. These same waters in some cases may also support a healthy aquatic
environment.
47
-------
Your Three Choices and Mow Tlicy W ould AITed (lie
Quality of California's Coastal Waters
Percent of California's Coastal Waters Killed ns "Cood"
Present
Conditions
Program 1
(Conditions after
3 years)
Program 2
(Conditions after
3 years)
Swimming
42% of miles are
good
_% gain to
% good
_% gain to
% good
Fish and shellfish safe for eating
64% of miles are
good
_% gain to
% good
_% gain to
% good
Habitat to support a large number
of different kinds of fish, birds,
mammals and plants
52% of miles are
good
_% gain to
% good
_% gain to
% good
Yearly Tax Change for your
household (permanent tax)
No Increase in
Taxes
Your taxes increase
by
$ per year
Your taxes increase
by
$ per year
48
-------
Q19 Which one of the options listed in the table above would you choose? (select one answer only)
~ Present Conditions: No change in your taxes, and the percent of coastal water that is "good"
for each purpose stays the same as it is now
~ Program 1: Your taxes increase by $ [fill with program 1 amount] per year to get the
improvements shown under this program
~ Program 2: Your taxes increase by $ [fill with program 2 amount] per year to get the
improvements shown under this program
~ Don't know
For those that choose the Present Conditions (Q19==l):
19 A. You chose Present Conditions over the two programs offered. Which of the following reasons BEST
describes why you made this choice? {select one answer only)
1. The improvements were not large enough for the money.
2. I am satisfied with the way things are now.
3. I am opposed to higher taxes.
4. I do not believe the programs will work as stated.
5. I do not have enough information to make a good decision.
6. I do not trust the government to run the programs well.
7. Someone else should pay for the improvements.
8. Other (Please specify )
49
-------
For those that choose a program (Q19==2 OR 3):
19b. Which of the following reasons BEST describes why you chose this program? {select one answer only)
1. The program I selected was less expensive than the other but still provided some improvements.
2. The program I selected was more expensive than the other but provided larger improvements in areas I care
about.
3. The program I selected provided larger improvements than the other in areas I care about.
4. I am most concerned about improvements for swimming and picked the program with the largest
improvement in this area.
5. I am most concerned about seafood consumption and picked the program with the largest improvement in
this area.
6. I am most concerned about wildlife habitat and picked the program with the largest improvement in this
area.
7. I was indifferent between the programs but wanted to pick something.
8. Other (Please specify )
The screen before the next choice questions should read:
Now consider two different programs - programs 3 and 4. As before, the table on the next screen shows
the percent of coastal waters that will improve under each new program and the taxes required from
each household to fund the new programs.
As you make your choice, please keep in mind the following:
Even though each program improves coastal waters in different ways, both would take three
years before the improved water returns to "good."
Neither program would improve the quality of freshwater lakes and rivers or coastal waters in
other states where swimming and fishing may take place or where a healthy aquatic
environment may exist.
Selecting a program means that your household would have less money to spend on other
things.
It is already possible in some places in California to swim in and eat the fish from the same
coastal waters. These same waters in some cases may also support a healthy aquatic
environment.
50
-------
The screen before the third choice questions should read:
Now consider two different programs - programs 5 and 6. As before, the table on the next screen shows
the percent of coastal waters that will improve under each new program and the taxes required from
each household to fund the new programs.
As you make your choice, please keep in mind the following:
Even though each program improves coastal waters in different ways, both would take three
years before the improved water returns to "good."
Neither program would improve the quality of freshwater lakes and rivers or coastal waters in
other states where swimming and fishing may take place or where a healthy aquatic
environment may exist.
Selecting a program means that your household would have less money to spend on other
things.
It is already possible in some places in California to swim in and eat the fish from the same
coastal waters. These same waters in some cases may also support a healthy aquatic
environment.
The screen before the fourth choice questions should read:
Now consider two different programs - programs 7 and 8. As before, the table on the next screen shows
the percent of coastal waters that will improve under each new program and the taxes required from
each household to fund the new programs.
As you make your choice, please keep in mind the following:
Even though each program improves coastal waters in different ways, both would take three
years before the improved water returns to "good."
Neither program would improve the quality of freshwater lakes and rivers or coastal waters in
other states where swimming and fishing may take place or where a healthy aquatic
environment may exist.
Selecting a program means that your household would have less money to spend on other
things.
It is already possible in some places in California to swim in and eat the fish from the same
coastal waters. These same waters in some cases may also support a healthy aquatic
environment.
51
-------
The screen before the last choice question should read:
Now consider two final programs - programs 9 and 10. As before, the table on the next screen shows the
percent of coastal waters that will improve under each program and the taxes required from each
household to fund the new programs.
As you make your choice, please keep in mind the following:
Even though each program improves coastal waters in different ways, both would take three
years before the improved water returns to "good."
Neither program would improve the quality of freshwater lakes and rivers or coastal waters in
other states where swimming and fishing may take place or where a healthy aquatic
environment may exist.
Selecting a program means that your household would have less money to spend on other
things.
It is already possible in some places in California to swim in and eat the fish from the same
coastal waters. These same waters in some cases may also support a healthy aquatic
environment.
See attached excel spreadsheet for tax and percent changes for all versions.
Q20a In the last few questions we asked you to consider different programs that would improve coastal
water quality. Did you think the improvements would take place in a specific part of California?
(select one answer only)
~ Yes (please let us know where you thought the improvements would take place )
~ No (skip to Q2la)
~ Don't Know (skiptoQ21a)
Q20b Why did you think the improvements would take place here?
Q20c Would you have answered differently if the improvements were to take place somewhere else in
California? (select one answer only)
n
Yes
n
No
n
Don'
52
-------
(If Ql="yes")
Q20d Given that you have limited time and resources, if you could not enjoy coastal water recreational
activities at the location of your choice, would you look for another location or would you engage in
other activities?
~ Look for another location
~ Do other activities (e.g., swim at neighborhood pool or fresh water lake, fish in freshwater stream or
river, play tennis, go shopping, etc.)
Q21a In the questions that asked you to consider different programs that would improve coastal water
quality suppose that we told you that all improvements in swimming would take place in "bays,"
"estuaries," or "inlets" rather than in California's ocean waters directly. Do you think you would
have answered these questions differently? (select one answer only)
~ Yes
~ No (skip to Q22)
~ Don't Know (skip to Q22)
Q21b Please take a moment to tell us why?
Q21c When we asked you to choose between different programs for improving the water quality of California's
coastal waters, was there anything about the questions or descriptions that seemed confusing?
~ Yes-What was confusing?
n No
Q21d When we asked you to choose between different programs for improving the water quality of California's
coastal waters, did the programs and their impacts seem believable?
~ Yes
~ No-^Why not?
Q21el. Of the following issues, which do you consider the most important?
~ Pollution of drinking water
~ Pollution of rivers, lakes and reservoirs
~ Contamination of soil
~ Air pollution
~ The loss of natural habitat for wildlife
~ Coastal Water pollution
~ Extinction of plant and animal species
53
-------
~ Urban sprawl and loss of open spaces
[Only show items that were not selected in previous questions]
Q21e2-Q21e7. Of the remaining issues, which do you consider the most important?
~ Pollution of drinking water
~ Pollution of rivers, lakes and reservoirs
~ Contamination of soil
~ Air pollution
~ The loss of natural habitat for wildlife
~ Coastal Water pollution
~ Extinction of plant and animal species
~ Urban sprawl and loss of open spaces
We would now like to learn 21 little hit more ahoul you and your household. This last set of questions is
lor background purposes only. W e would like to remind vou (lint nil i 11 Toi'in;ilion vou provide w ill he
confidential. and your name will not he associated with any responses in this survey.
Q22 Are you a member of an environmental, conservation or outdoor sporting organization? (select
one answer only)
~ Yes
~ No
~ Don't know
54
-------
Q40 How many people in your household contributed to your income in 2003?
Number of people
55
-------
Standard Knowledge Networks Questions
Do you have any comments on the survey in general?
Thank You!! We appreciate your help with this important study.
Please feel free to share any comments you have about this survey or the topic of water quality.
56
-------
The Recreational Benefits of Improvements in New England's Water Quality: A Regional RUM
Analysis
George R. Parsons
COLLEGE OF MARINE STUDIES AND DEPARTMENT OF ECONOMICS
UNIVERSITY OF DELAWARE, NEWARK, DE 19716
Erik C. Helm
U.S. EPA, OFFICE OF SCIENCE AND TECHNOLOGY
1200 PENNSYLVANIA AVE., WASHINGTON D C. 20460 (4303T)
Tim Bondelid
RTI
RESEARCH TRIANGLE PARK, NC 27709
This study was funded by the U.S. Environmental Protection Agency's Office of Policy,
Economics, and Innovation through Cooperative Agreement CR82486-01-02
57
-------
1. Introduction
The purpose of this paper is to measure the economic benefits to recreation from
improved water quality in six northeastern states. The states include Maine, New Hampshire,
Vermont, Massachusetts, Rhode Island, and Connecticut. All lakes, rivers, and coasts (oceans
and bays) in the region are included in the analysis. The benefits are measured using separate
random utility maximization (RUM) models for fishing, boating, swimming, and viewing. All
models are for day-trip recreation which accounts for approximately 77% of all water based
recreation trips in the region. The models are estimated using data from the 1994 National
Survey of Recreation and the Environment (NSRE) and water quality modeling simulations
based on the National Water Pollution Control Assessment Model, Version 1.1 (NWPCAM1.1)
(RTI, 2000).
We consider three welfare scenarios in our analysis. The first two are hypothetical. They
assume that all water bodies in the region attain some minimum level of quality. We consider a
moderate and then a high level of quality defined by levels of biological oxygen demand,
dissolved oxygen, total suspended solids, and fecal coliforms. The third scenario considers a
simulation of the actual improvement realized under the Clean Water Act through 1994.
Our paper is organized into 4 sections. Section 2 lays out the RUM models. Section 3
discusses our application and the data. Section 4 presents the parameter estimates and welfare
results. Section 5 restates some of the important caveats in our analysis.
2. The Model
We estimate separate models for fishing, boating, swimming, and viewing. Each is
estimated in two stages: participation and site choice. The participation model considers the total
number of trips a person makes over the season. Site choice considers the site chosen for the last
trip taken. A site is a lake, segment of a river, or segment of a coastline. The two models are
linked using an approach suggested by Bockstael, Hanemann, and Kling (1987) and latter
adapted by Hausman, Leonard, and McFadden (1995).
It is easiest to describe the model beginning with site choice. An individual is assumed to
visit one of S possible recreation sites on a given day. Let i = 1,..., S denote a site. Each site i
gives a person utility Ui. This site utility depends on the cost of reaching the site and the
characteristics of the site
(1) U, = tcJtc +xjx +£,
where lcI is the trip cost of reaching site i, x, is a vector of characteristics of site i, and et is a
random term. The /is are parameters to be estimated. The vector x, includes characteristics of the
sites that matter to individuals when making site choice - water quality, access and so forth.
A person is assumed to visit the site that gives the highest utility. That utility is called the
person's trip utility and is defined as
(2) V = Max{Uh U2,....,US}.
58
-------
Substituting equation (1) into (2) gives
(3) V = Max{tcj ptc +xjx+ e„ lc2 fJ>,c + x2fix + e2,...., lcjlc + xjx + z,.\
Now consider a change a water quality at one or more sites. Assume that x, represents site
characteristics at site i without an improvement in water quality and assume that x*; represents
site characteristics at site i with an improvement. Only the element pertaining to water quality in
x, has changed between the two states of the world. For some sites there may be no change.
Without the change in water quality a person's trip utility is Vshown in equation (3).
With the change in water quality and assuming the change only takes place at sites 1 and 2, trip
utility is
(4) 1* = Max {tc, filc + x", fix + e„ lc2 fJ>,c + x2fix + e2,...., lcjlc + xjx + s..}
The change in utility due the water quality improvement is
(5) Aw = V-V .
If a person visits site k without the improvement in water quality, but chooses to visit site
1 now that it is cleaner, trip utility increases by Aw = U*l - Uk. If the person visited site 1 without
the water improvement and continues to visit site 1 with the improvement, trip utility increases
by Aw = U*1 -11The person makes the same trip but enjoys cleaner water. If the person visited
site k without the improvement and continues to visit site k after the improvement there is no
change in welfare. Perhaps sites 1 and 2 are located far from the person's home or have other
features the person dislikes. Finally, if there is a relative change in water quality at sites 1 and 2,
the person may shift from one site to the other and have a change in welfare. For example, a shift
from site 1 to 2 would give an increase of lf2- U,. In any case all of these pathways to utility
change are captured in equation (5) in Aw.
The change in trip utility is converted to money terms by dividing Aw. by the negative of
the coefficient on trip cost. In the RUM model -[J>tc is a measure of the marginal utility of
income. It tells us how much an individual's site utility would increase if trip cost were to
decline for that trip. The increase in welfare due to an improvement in water quality at sites 1
and 2 is
(6) cs = Aw/- fitc.
In application, we use an expected value for Aw. because its actual value is random and
unknown. To see this substitute equations (3) and (4) into equation (5). Assume the parameters (J>
are known or estimated. Since each site utility has a random component eh Aw . and cs must also
be random. For this reason, the statistical expected values of V and V are used in application.
The expected increase in welfare due to a water quality improvement is
(7) cs={E(r)-E(V)\l-fl„
59
-------
where E denotes an expected value over the site utilities and will depend on the distribution of
the errors terms in each site utility. Equation (7) gives a per trip value for the change in water
quality.
The site choice model is usually estimated using some form of a multinomial logit model.
We use a simple logit model in our application. A person's probability of visiting site k on a
given choice occasion in a simple logit model is
(8) piik) = exp(/q ptc + xk fij / £exp(fc, ptc + x, fij .
This form applies for any site and implies the following log-likelihood function
(9) AOT = nnrf/lnM')
where dj = 1 if individual j visited site i and dj = 0 if not. The pr(i) in equation (9) takes the form
shown in equation (8). This function gives the likelihood of observing the patterns of visits
actually observed a dataset. The parameters (J> are chosen to maximize A(fi). These estimated
parameters, in turn, may be used to estimate per trip welfare shown in equation (7). In the simple
logit model expected trip utility takes the form
(10) E(V) = In £ expC/c, ptc + x, fij .
This is sometimes called the 'inclusive value'. The per trip value of a water quality improvement
then is
(11) cs = { In £ exp(/6', ptc + x*JJ- In £ exp(/c, ptc + x, fij J / - ptc
where x*; is with the improvement and x, is without.
Our participation decision models the number of trips an individual takes during a year.
The participation function takes the Poisson form
(12) pr( Rj = Tj) = e-AjPj I y\ In kj = au(T) +
where rf is the number of trips taken by person j during the season. I'; = E(V") / - //,6 is a
monetized utility index or consumer surplus for a recreation trip predicted using the parameter
estimates from the site choice model. The vector Zj is a set of individual characteristics for person
j believed to influence trip taking, like family size, age and so forth.
This is Hausman, Leonard, and McFadden's (1995) formulation of the participation
model. It is a simple adaptation of Bockstael, Hanemann, and Kling's (1987) model. The
adaptation is the monetization of the expected utility. Since this is a linear transformation of a
scalar, the models are the same. The transformation merely rescales the parameter estimate on
the index. Neither model is strictly utility theoretic.
Using an estimated participation model in a Poisson form, Hausman, Leonard, and
McFadden (1995) show that the annual change in welfare due an improvement in water quality
like that discussed above is
60
-------
(13) CS = (/*j - r"j) / cTu
where r"; and are predicted values of trips for person j from the participation model with and
without the change in water quality, and a u is the coefficient on /', = E(Vs) / - fi ,c in the same
model. See Parsons, Jakus, and Tomasi (1999) for more detail on the participation function.
3. Application and Data
Our application is to six northeastern states: Maine, New Hampshire, Vermont,
Massachusetts, Connecticut, and Rhode Island. All rivers, lakes, and coasts in the region are
included in the analysis. The data are from two sources. The trip and respondent characteristic
data are from the 1994 National Survey of Recreation and the Environment (NSRE94). The site
characteristic data were developed using NWPCAM1.1, a national water quality simulation
model that is built around the RF1 river/stream network database (EPA's Reach File 1).
In the NSRE94 individuals throughout the United States were contacted at random by
phone and asked to report the total number of day and overnight trips taken separately for
viewing, boating, fishing, and swimming at domestic water-based recreation sites over the past
twelve months. See Appendix A for the survey questions defining recreation uses. Each person
was also asked to report the site visited on the last trip for each type of recreation and to report
the location of his or her hometown. As usual demographic data were gathered for each
respondent. This included income, age, job status, family size, and other characteristics. Our
sample includes all individuals surveyed from the six northeastern states. Our sample size is 632.
Table 1 presents descriptive statistics over the sample population. Our analysis is for day trips
only. The participation rates and average number of trips for each type of recreation are
Recreation
Use
Percent of the Sample Taking at Least
One Day Trip to a Water-based
Recreation Site Over the Past 12 Months
(n = 632)
Average Number of Day
Trips Taken by People
Taking at Least One Trip
During the Year
Viewing
25%
7.08
Boating
14
8.80
Fishing
12
10.06
Swimming
24
10.05
These rates are from the general population and exclude overnight trips. About 77% of all trips
were day-trips. Our analysis accounts for most of the day-trips taken to sites in the region — less
than 3% of the "last trips" in the NSRE94 to the six states were taken by residents outside the
region. The average distances traveled on a day trip in our sample and average distance to all
sites in the choice set are
61
-------
Recreation
Use
Average Distance
Traveled
On Day Trips (miles)
Average Distance to all Sites
in the Choice Set
(miles)
Viewing
72
104
Boating
61
104
Fishing
50
104
Swimming
54
104
The maximum distance to a site in the choice set is 200 miles. Again, these are day trips
only and ignore trips taken by persons outside the region.
The site characteristic data were constructed using NWPCAM 1.1 and the EPA's RF1
database. There are 20,925 rivers, 2,975 lakes, and 1,231 coasts in the data set. A site on a river
is defined as a stretch of river from one confluence to another without a major tributary, lake, or
population center intervening. If a major tributary, lake, or population center is passed, a new
site is defined. A coastal site is defined as the coastal line along a bay or ocean between the
mouth of a major river or beginning of a new municipality and the mouth of another major river
or beginning of a new municipality. The lake data set is all major lakes and ponds in the region.
A single lake, no matter how large, is never divided into more than one site.
Site-specific water quality data were estimated using NWPCAM 1.1 (RTI, 2000). In this
model, place-specific pollutant loadings from both point and nonpoint sources across the nation
are linked and routed through the RF1 surface water network. The model incorporates a
hydrodynamic and water quality modeling algorithm that allows it to estimate instream pollutant
concentration throughout the network for dissolved oxygen (DO), biological oxygen demand
(BOD), total suspended solids (TSS) and fecal coliform bacteria (FCB).
In our application we estimate separate models for each recreation type. Because of the
large number of sites in each person's choice set, we estimate the model using a random draw of
sites. Each person's choice set includes his or her actually chosen site plus 36 other randomly
drawn sites. Each choice set for estimation is composed of 12 rivers, 12 lakes, and 12 coasts. See
Parsons and Kealy (1992) for more on estimation with randomly drawn choice sets.
Each model considers four basic attributes for site utility in equation (1): trip cost,
resource type, choice set size, and water quality. Trip cost is the sum of travel and time cost
(14) tc = (.35+ rtdist) + (income / 2040) * (rtdist / 40)
where rtdist is round trip distance and income is annual income. Round trip dist is the linear
distance between each site and a person's hometown. Travel cost is assumed to be 35 cents per
mile. The opportunity cost of an hour is approximated using annual income divided by 2040
which is the typical number of hours worked in a year. The average travel speed is assumed to be
40 miles per hour.
Resource type is a set of dummy variables distinguishing river, lake and coastal sites.
Choice set size is a control variable to account for the fact that even though each person has the
62
-------
same number of alternatives in the choice set in estimation (36 sites), in reality some will have
far more than others. Persons with larger choice sets, all else constant, are more likely to take a
trip.
Water quality is defined as low, medium, or high. This is an index based on the levels of
biological oxygen demand, total suspended solids, dissolved oxygen, and fecal coliform. The cut
offs for high and medium are
Biological
Oxygen
Demand (mg/L)
Total Suspended
Solids (mg/L)
Dissolved
Oxygen
(% saturation)
Fecal Coliforms
(MPN/lOOmL)
High
Water
Quality
<1.5
<10
>83
<200
Medium
Water
Quality
<4
<100
>.45
<2,000
All four object measures must be below (or above in the case of dissolved oxygen) the cutoffs
shown before the site is classified as having that quality level. If any single characteristic falls
short of its cut off for medium quality, the site is classified as low quality. Sites with low water
quality have no plant or animal life and often have visible signs of pollution (trash, oil). Site with
medium water quality have some game fishing and usually few visible signs of pollution. Sites
with high water quality are suitable for extensive human contact, have the highest natural
aesthetic, and support high quality sport fisheries.
The water quality data are based on NPWCAM1.1 pollutant loading data and water
quality modeling results (for mid-1990's conditions). Coastal water quality is based on the
predicted water quality at the mouths of nearby rivers. In some instances, watershed averages are
used when data were missing from the simulation results. The distribution of water quality
across sites is
Percent of all
Rivers
Percent of all
Lakes
Percent of all
Coasts
High Quality
49.9%
28.5%
30.8%
Medium Quality
36.4
59.9
37.7
Low Quality
13.7
11.6
31.5
Site utility takes the following form in our application
(15) If", = pmtctct + [f'/iv, + pnccstl + (rjiwq, + + In (size) + e™
63
-------
where i denotes a site and m denotes a recreation use (m = viewing, boating, fishing, or
swimming). The choice set size variable is estimated with its coefficient set equal to one since it
is entered as a weighting factor only. This gives us 20 parameters to estimate in the site choice
model — 5 parameters in each of the four models. More complex versions of the model which
included site size, separate measures for each objective water quality measure in our index, and
an intermediate step in water quality between our high and medium gave rise to models that
failed to converge and in the isolated cases where convergence was achieved gave results that
ran strongly against our priors.
Four participation models, one for each recreation use, were estimated separately in
Poisson form and included the attributes shown in Table 1. The expected utility index (Tj / - /?J
in these regressions was constructed from the relevant site choice stage. All participants and
nonparticipants were included in each regression. Attempts to estimate the model by full
information maximum likelihood, once again, lead to convergence problems. The results shown
here are based on sequential estimation.
We consider three welfare scenarios using our model. The first two assume water quality
at all sites attains some minimum level. The first assumes water quality attains at least a medium
level as defined above at all sites in the region. Under this scenario 13.7% of all rivers, 11.6% of
all lakes, and 31.5% of all coasts realize water quality improvements. The second assumes water
quality attains a high level of quality at all sites. This is a significant improvement in water
quality in the region affecting 50.1% of all rivers, 71.5% of all lakes, and 69.2% of all coasts
over the six northeastern states.
The last scenario considers the water quality we are likely to have realized in 1994 in the
absence of the Clean Water Act and assuming no state, local, or judicial controls were otherwise
established. In this scenario we assume water quality improves from a hypothetical 'no-CWA'
state of the world to current conditions. This is approximately the recreational benefits realized
due to the existence of the Clean Water Act in 1994. The 'no-CWA' conditions were estimated
using the same simulation model used to estimate current conditions. Pollutant loadings were
adjusted in that model to reflect loads likely to have been attained in the absence of the Clean
Water Act. To get an idea of how the CWA simulation is changing water quality in the model,
consider the following table. The table reports the value of the ratio
Number of sites at quality level wq with the improvement
Number of sites at quality level wq without the improvement
where wq = low, medium or high. The table shows the degree of shift from lower to higher
quality sites.
River Ratio
Lake Ratio
Coast Ratio
high
1.25
1.44
1.10
medium
1.01
.94
1.17
low
.56
.07
.79
64
-------
The next section presents the parameter estimates and welfare results for each of these scenarios.
4. Parameter Estimates and Welfare Results
The parameter estimates for the site choice model are shown in Table 2. For the most
part, the signs are as expected. The coefficient on trip cost is negative and significant in all four
models. Recall that this variable is used as the marginal utility of income and is important in
converting measures of utility change into dollars. The coefficients on the resource type
dummies suggest that coasts, all else constant, are the most important resource for recreation use.
Recall that lake is the excluded category so the resource type coefficients are interpreted relative
to lakes. The coefficient on coast is highest for viewing and lowest for fishing. The coefficient
on river suggests that river sites, all else constant, are the lowest valued among the three resource
types except for boating. There are a number of large rivers in the region where boating is quite
popular. This, no doubt, accounts for the result on boating. The negative river coefficient for
swimming is largest capturing the infrequent use of rivers in this activity.
The coefficient on middle WQ is positive and significant in two of the four models -
fishing and swimming. This implies that moving from low to middle level water quality imparts
benefits mostly to fishing and swimming uses. Between these two recreation types the utility is
increased most for fishing. Boating also has a positive but insignificant coefficient on middle
WQ. Viewing has a negative and insignificant coefficient. Modest improvements in water
quality appear to yield little or no increase in utility for these recreation uses.
The coefficient on high WQ is positive and significant in all four models as one would
expect. The coefficients also show that high water quality gives higher utility than middle water
quality. Again, going from low to high water quality, the utility increase is greatest for fishing
and swimming. However, the coefficients on viewing and boating imply utility increases for
these recreation uses as well. It is interesting to note that for fishing most of the increase in
utility comes from moving from low to middle water quality. For viewing and boating almost all
of the utility increase comes from moving from middle to high water quality.
The results of the Poisson models are shown in Table 3. The coefficient on the monetized
utility index (expected utility or inclusive value from the site choice stage divided by the
negative of the 13 coefficient on trip cost) is positive in all four regressions. This coefficient
gives us some idea of how responsive participation in each recreation use will be to
improvements in water quality. Viewing and fishing participation are the most responsive to
improvements. Swimming is somewhat less responsive and boating shows little if any
responsiveness.
Income has a positive effect on viewing, boating, and fishing participation and a negative
affect on swimming. Urbanities have lower participation rates in all uses, all else constant, but
the effect is insignificant in the viewing model. As one ages the probability of participating in all
four recreation uses decreases. Retired folks have a higher probability of participating in boating
and fishing and a lower probability in viewing and swimming. Men have higher probabilities of
participating in boating and fishing, and women in viewing and swimming. Education level
increases ones probability of participating in all uses except for fishing where it has a negative
and significant affect on participation. Unemployed also increases the probability for all uses
except fishing but the coefficient is insignificant. Being a student increases the likelihood that
you will participate in viewing and swimming. Being a homemaker increases your likelihood for
65
-------
swimming only. Larger families have higher probabilities for viewing and boating. Having more
leisure hours increases one's probability of participating in all uses but fishing. And finally,
owning a boating dramatically increases one's probability of boating and fishing and to a lesser
extent viewing. We excluded boat ownership from the swimming model.
Now we turn to annual benefit estimates for water quality improvements. The annual
average per person benefits over all resource types for our three scenarios are as follow
Viewing
Boating
Fishing
Swimming
All sites
improve to
middle WQ
$.04
$3.14
$5.44
All sites
improve to
high WQ
$31.45
$8.25
$8.26
$70.47
Improvements
due to Clean
Water Act
(CWA)
$.47
$.62
$2.40
$5.59
These averages include participants and nonparticipants and are computed using equation (13).1
The first two scenarios use current conditions as the baseline. The CWA scenario uses pre-CWA
water quality as the baseline. Table 5 shows the same results for each scenario by recreation use
and separately for improvements to rivers only, lakes only, and coasts only.
For modest improvements in water quality (to middle WQ) almost all of the benefits go
to fishing and swimming. The annual fishing benefit is about $3 per person. The annual
swimming benefit is about $5. Again, this includes participants and nonparticipants. Table 5
shows a negative benefit for viewing due to the negative coefficient on middle WQ in the view
model. In the table above, I have simply recorded no benefit for viewing. Table 5 also shows that
most of the swimming benefit is coming from cleaning up the coast, and most of the fishing
benefit is coming from the clean up of coasts and lakes.
For significant improvements in water quality (to high WQ), all four recreation uses
realize benefits and the benefits are must larger. Swimming and viewing are the highest at $70
and $31 per person. Boating and fishing are about $8 per person. For fishing 38% this benefit is
realized in moving from low to middle quality, and 62% is realized in moving from middle to
high quality. For swimming the same incremental benefits are 8% and 92%. And, as noted
earlier for viewing and boating, nearly all of the benefit is realized in the second increment. Most
the benefits are coming through a clean up of the coastlines.
xPer trip values using equation (11) are also provided Table 4. Since annual values are
typically of more interest for policy we focus our discuss on these.
66
-------
For improvements due the Clean Water Act, all recreation uses realize benefits.
Swimming and fishing are the largest at $6 and $2 per person. Viewing and boating are positive
but less than $1 per person. In this case the source of most of the benefits are the rivers and lakes
where the CWA has had it largest effect.
Table 6 shows aggregate benefits for each scenario. These are calculated by multiplying
the mean per person benefit for each state by its population in 1994 over the age of 16. All
numbers are in 1994 dollars. Summarizing Table 6, we have
All sites Attain
medium WQ
All sites attain
high WQ
Due to the Clean
Water Act
Total Benefit in
Millions of 1994
Dollars
$77
$1,295
$99
Distribution of Total
Benefits by
recreation use:
Viewing
0%
26%
5%
Boating
0%
7%
7%
Fishing
36%
7%
26%
Swimming
63%
60%
61%
The aggregate benefits to the region range from $ 77 million for improvements to medium water
quality to $1.3 billion for improvements to high water quality. Again the benefits go mostly to
swimming and fishing for a medium clean-up. The benefits go mostly to swimming and viewing
for improvements to high water quality. Overnight trips, non-recreation use, and nonuse values
are excluded from these numbers.
The aggregate benefits due to the Clean Water Act in 1994 dollars are $99 million. These
estimates assume the controls set by the Act are not in place and are not replaced by any state,
local or judicial controls. The estimates are based on RTFs simulation model. The benefits go
mostly to fishing and swimming.
5. Caveats
While our models give plausible results for broad changes in water quality across the region,
several caveats in the research are worth repeating.
0 Using finer measures of water quality in the RUM model persistently led to
complications in the econometrics, usually a model that failed to converge. By finer
measures, we mean using the objective water quality variables separately and having an
intermediate step between high and medium quality.
67
-------
0 More complex specifications (nested and mixed logit models), models with more site
characteristics, and estimating by full information maximum likelihood, also created
problems with convergence and implausible parameter estimates.
0 There is no coastal water quality simulation model per se. The RTI model essentially
uses water quality estimated from the mouths of rivers near coastal sites. And, the
coastlines are by far the most aggregated sites. Since coastlines were the source of many
of the benefits, caution is warranted.
0 Water quality data at a site level were not available for many lakes in our data set. For
these lakes we used a watershed average water quality.
0 Our baseline pre-CWA water quality levels assume no local, state, or judicial action in
the absence of the CWA. This is an extreme position that leads to some overstatement of
the benefits attributed to the CWA.
0 Our benefits measures exclude overnight trips, non-recreation uses of the water, some
smaller water bodies, and nonuse value. This leads to some understatement of the
benefits for each scenario.
68
-------
References
Bockstael, N. E., W. M. Hanemann, and C. Kling (1987). "Estimating the Value of Water
Quality Improvements in a Recreational Demand Framework," Water Resource Research, 23(5):
951-960.
Hausman, J.A., Leonard, G.K., and Mcfadden, D. (1995). "A Utility-Consistent Combined
Discrete Choice and Count Data Model: Assessing Recreational Use Losses Due to Natural
Resource Damage," Journal of Public Economics, 56, 1-30.
Parsons, G. R., P. Jakus, and T. Tomasi (1999). "A Comparison of Welfare Estimates from Four
Models for Linking Seasonal Recreation Trips to Multinomial Logit Models of Choice," Journal
of Environmental Economics and Management, 38: 143-59.
Parsons, G. R. and M. J. Kealy (1992). "Randomly Drawn Opportunity Sets in a Random Utility
Model of Lake Recreation," Land Economics, 68, 93-106.
Parsons, G. R., D. M. Massey, and G.R. Parsons (2000). "Familiar and Favorite Sites in a
Random Utility Model of Beach Recreation," Marine Resource Economics 14, pp. 299-314.
Research Triangle Institute (RTI). 2000. National Water Pollution Control Assessment Model
(NWPCAM) Version 1.1. Final report prepared for the U.S. Environmental Protection Agency,
Office of Policy, Economics and Innovation, Washington, DC.
69
-------
Table 1: Descriptive Statistics
Description
Sample Mean
Income
Annual income
$56,574
Urban
Urban dummy (=1 if live in an urban area)
.18
Age
Age
43
Retired
Retirement dummy (=1 if retired)
.18
Gender
Gender (=1 if male)
.42
Education
Level of education (scale 1-5)
4.4
Unemployed
Unemployment dummy (=1 if unemployed)
.13
Student
Student dummy (=1 if full time student)
.10
Homemaker
Homemaker dummy (=1 if homemaker)
.22
Family Size
Number of people in family at home
2.9
Leisure Hours
Leisure hours per week
21.8
70
-------
Table 2: Random Utility Model of Site Choice
Viewing
Boating
Fishing
Swimming
Price
-.042
-.062
-.055
-.030
(33.1)
(24.8)
(21.1)
(36.0)
River
-.090
.716
-.689
-5.489
(5.6)
(3.9)
(4.4)
(7.7)
Coast
4.59
3.54
1.865
3.69
(37.7)
(24.1)
(11.2)
(44.2)
High WQ
.421
.496
.912
.881
(2.56)
(1.73)
(3.16)
(6.7)
Middle WQ
-.136
.016
.898
.325
(1.4)
(0.1)
(4.94)
(3.6)
Log-Likelihood
-.335
-5.13
-4.91
-13.44
71
-------
Table 3: Poisson Participation Model
Viewing
Boating
Fishing
Swimming
LOG INCLUSIVE/^
0.0064
0.0012
0.0051
0.0017
(13.498)
(1.226)
(5.772)
(5.006)
INCOME
0.98E-06
0.31E-05
0.77E-05
-0.24E-05
(1.459)
(3.664)
(10.942)
(-3.873)
URBAN
-0.0134
-0.8017
-0.8967
-0.1391
(-0.222)
(-6.269)
(-8.321)
(-2.291)
AGE
-0.0125
-0.0230
-0.0212
-0.0145
(-6.249)
(-7.541)
(-8.570)
(-7.248)
RETIRED
-0.5016
1.1352
0.5027
-0.8049
(-3.663)
(5.069)
(3.576)
(-6.642)
GENDER
-0.4950
0.1895
1.2663
-0.2319
(-9.869)
(2.568)
(15.552)
(-4.724)
EDUCATION
0.1314
0.0406
-0.1668
0.2254
(9.000)
(1.848)
(-8.981)
(16.362)
UNEMPLOYMENT
-0.6402
-0.6904
0.0602
-0.2889
(-4.955)
(-3.356)
(0.534)
(-2.586)
STUDENT
0.6564
-0.3914
-0.5052
0.2846
(10.351)
(-3.220)
(-4.611)
(4.351)
HOMEMAKER
-0.6104
-1.0769
0.0310
0.3243
(-8.014)
(-6.087)
(0.243)
(5.616)
FAMILY SIZE
-0.0407
-0.1185
0.0642
0.1240
(-2.673)
(-4.285)
(3.516)
(11.243)
LEISURE HOURS
0.0081
0.0040
-0.0037
0.0079
(8.444)
(3.353)
(-2.339)
(8.889)
BOAT OWNED
0.5422
2.7759
1.7806
--
(4.400)
(34.360)
(26.152)
Log-Likelihood
-3708
-1176
-2229
-3878
72
-------
Table 4: Mean Per Trip Benefits Per Person (1994$)
Viewing Boating Fishing Swimming
Due to Clean Water Act:
All Sites
$0.22
$0.49
$1.45
$1.69
Rivers Only
0.10
0.28
0.45
0.01
Lakes Only
0.13
0.12
0.58
0.72
Coasts Only
-0.03
0.07
0.38
0.93
Sites Attain Middle WQ:
All Sites
-0.48
0.03
1.67
1.48
Rivers Only
-0.01
0.003
0.13
0.0006
Lakes Only
-0.05
0.007
0.87
0.31
Coasts Only
-0.41
0.02
0.70
1.19
Sites Attain High WQ:
All Sites
9.75
5.99
3.87
19.43
Rivers Only
0.82
1.82
0.76
0.03
Lakes Only
2.17
1.63
1.10
5.96
Coasts Only
7.41
3.07
2.19
15.29
-------
Viewing
Boating
Fishing
Swimming
Due to Clean Water Act:
All Sites
$0.47
$0.62
$2.40
$5.59
Rivers Only
0.21
0.36
0.72
0.03
Lakes Only
0.38
0.17
0.95
2.58
Coasts Only
-0.13
0.07
0.65
3.04
Sites Attain Middle WQ:
All Sites
-1.61
0.04
3.14
5.44
Rivers Only
-0.02
0.003
0.54
0.002
Lakes Only
-0.15
0.01
1.29
1.06
Coasts Only
-1.43
0.03
1.34
4.41
Sites Attain High WQ:
All Sites
31.45
8.25
8.26
70.47
Rivers Only
2.25
2.51
1.86
0.11
Lakes Only
6.21
2.39
1.73
21.20
Coasts Only
24.67
4.01
4.39
55.50
74
-------
Viewing
Boating
Fishing
Swimming
Total
Due to Clean Water Act:
All Sites
$5,120
$6,893
$26,298
$61,085
$99,333
Rivers Only
2.336
3.990
7.921
0.292
14.539
Lakes Only
4.198
1.850
10.431
28.271
44.749
Coasts Only
-1.410
0.744
7.077
33.214
39.625
Sites Attain Middle WQ:
All Sites
-17.614
0.418
34.340
59.490
76.634
Rivers Only
-0.242
0.035
5.903
0.019
5.715
Lakes Only
-1.650
0.092
14.151
11.639
24.233
Coasts Only
-15.691
0.290
14.688
48.220
47.506
Sites Attain High WQ:
All Sites
344.015
90.268
90.318
770.725
1295.326
Rivers Only
24.558
27.499
20.312
1.152
73.520
Lakes Only
67.958
26.128
18.939
231.838
344.863
Coasts Only
269.766
43.815
48.028
606.958
968.567
75
-------
Appendix A
Survey Questions Defining Four Recreation Uses
Boating
Did you leave from your home to take any trips or outings where the
primary purpose was to go boating in the last 12 months? Boating inludes
trips to go motorboating, sailing, windsurfing, canoeing or kayaking,
rowing, tubing or other floating. Please do not include trips taken for
any other primary purpose such as swimming, fishing, or to just be near
water.
Fishing
Did you leave from your home to take any trips or outings where the
primary purpose was to go fishing in the last 12 months? Please do not
include trips taken for any other primary purpose such as swimming,
boating, or to just be near water.
Swimming
Did you leave from your home to take any trips or outings where the
primary purpose was to go swimming outdoors in something other than a pool
in the last 12 months? Please do not include trips taken for any other
primary purpose such as fishing, boating, or to just be near water.
Viewing
Did you leave from your home to take any trips or outings where the
primary purpose was to visit a beach or waterside in the last 12 months?
Please do not include trips taken for any other primary purpose such as
fishing, boating, or swimming. Please include trips for example, your
picnics, nature study outings, and vacations, where you purposely chose to
be by the water.
76
-------
Valuing Water Quality Changes Using a Bioeconomic Model
of a Coastal Recreational Fishery
AUTHORS:
Matt Massey & Steve Newbold
U.S. EPA - National Center for Environmental Economics
Washington, DC
Brad Gentner
U.S. NOAA - National Marine Fisheries Service
Silver Spring, MD
ABSTRACT:
Most previous studies on the effects of water quality on recreational fishing have
focused on a single element in the chain of effects that connect water quality changes to
the welfare of anglers. Due to a scarcity of detailed water quality data, most of these
studies have also been forced to examine water quality effects across large aggregated
areas. The result is a large number of studies that are difficult to combine to evaluate
specific water quality policies in a comprehensive manner. This paper describes a
bioeconomic model of a coastal recreational fishery that combines standard models of
fish population dynamics, angler catch, and recreation site choice. We use a structural
modeling approach that allows us to combine a variety of data sources and provides
more flexibility for evaluating various water quality policies than most previous
valuation models.
First, we develop a population model that describes the influence of water quality on
overall fish abundance through the effects of dissolved oxygen (DO) on the survivorship
of young juvenile fish. The population model is based on data on survival,
reproduction, and the effects of DO on juvenile fish from the fisheries science literature
and government reports. The model is calibrated using average historic recreational
harvest levels in and out of the study area and historic commercial harvest levels for the
entire fishery.
Second, we estimate a catch model that describes the influence of fish abundance and
water quality on anglers' average catch rates. The catch model is estimated using a
combination of three data sources. First, we use monthly data on water quality
conditions from 23 water quality monitoring stations distributed throughout Maryland
four coastal Bays. Next, we incorporate catch data from a sample of anglers who fished
for summer flounder. Each fisherman reported their date and location fished, catch,
fishing methods, and some personal characteristics. Lastly, we include a measure of
fish abundance from fishery-independent data collected in bottom trawl surveys, all in
77
-------
Maryland's coastal bays in 2002. The disaggregated nature of this data allows us to
estimate spatially and temporally varying catch rates.
Third, we estimate a recreation demand model that describes the welfare effects and
changes in trip demand from changes in catch rates. The recreation demand model is
based on data from a stated choice survey of anglers who fish for summer flounder on
the Atlantic coast. In the survey, respondents were asked to choose between two
flounder fishing trips of varying quality (catch, regulations, cost, etc.) and a "do
something else" option. Using this model we estimate the value of several changes in
water quality typically valued in the literature and changes in participation rates.
Next, we integrate the population, catch, and recreation demand models to create a
bioeconomic model that accounts for the feedback on the fish population through
changes in the overall harvest pressure in the recreational fishery on the fish stock. The
bioeconomic model is used to estimate the aggregate benefits to recreational anglers
from several illustrative scenarios of changes in water quality. Results indicate that
improving water quality throughout the range of the species could lead to substantial
increases in the fish population and associated benefits to recreational anglers from
increased catch rates. Water quality improvements confined to Maryland's coastal bays
alone would have much smaller impacts. Because DO appears to only weakly affect the
"catchability" of summer flounder (i.e., the average angler catch conditional on fish
abundance); the largest effects predicted by the model come from the long run impact of
DO on fish abundance through its affect on juvenile survival. This finding suggests that
studies that simply include DO measures as site characteristics (and as a proxy for short
run catchability effects) may be missing the major (long term) effects of DO on fish and
fishermen. Important areas for improved data collection and model development are
also discussed.
78
-------
Valuation of Ecological Benefits: Improving the Science Behind Policy Decisions
Workshop Sponsored by the U.S. EPA
Wyndham Washington Hotel
Washington, DC
October 26-27, 2004
Session II: Cleaning Our Coastal Waters: Examinations of the Benefits of
Improved Water Quality
Discussant: Bob Leeworthy, Leader Coastal and Ocean Resource Economics Program,
Special Projects, National Ocean Service, National Oceanic and Atmospheric
Administration
The Value of Improvements to California's Coastal Waters: Results From a Stated-
Preference Survey, by Nicole Owens and Nathalie Simon
Comments:
1. The procedures followed in designing the survey questionnaire and sample design
were very good. In reviewing the paper, I was quite surprised that the same
procedures were used that we at NOAA are currently using in designing a survey
to value the coral reefs in Hawaii. The use of focus groups, protocol interviews
(one-on-one interviews with debriefings), a large-scale pre-test, and final survey
with peer review used seems to have become a standard model. And, I think the
model is a good one. It appears much was learned in the process and significant
changes in both questionnaire and sample design as a result.
2. The decision to switch from a contingent valuation approach to a state-preference
approach appears to be a good decision. This approach seems better fitted to the
problem and the approach seems to have strong scientific backing.
3. The use of an Internet Panel (Knowledge Networks) also appears to be a good
decision. The use of Knowledge Networks Internet Panels has not been fully
endorsed to date for policy/management application by the U.S. Office of
Management and Budget (OMB). Several papers to be presented at this
workshop demonstrate that Knowledge Network's Internet Panels can provide
"representative samples" for a variety of applications, especially for
environmental/ecological benefit estimation. So again, I think the sample design
used has scientific support. As a note, we at NOAA are also planning to use
Knowledge Network's Internet Panel for our study to estimate the economic value
of Hawaii's Coral Reefs. We will seek nothing less than approval to apply to
policy/management. Again, I believe the science presented at this workshop
supports that decision.
4. I have some problems with a couple of the survey questions. Specifically,
Questions 3 and 4.
Q3. Does your household own a boat that is used primarily on coastal waters?
Q4. For which activity do you use your boat the most on coastal waters?
79
-------
recreational fishing
recreational boating
commercial fishing
charter boat rides
charter fishing trips
other (please specify)
I am not sure how this information can be used. Using the qualifier in Q3 that the
boat had to be primarily used on coastal waters doesn't make sense to me. I think
a great deal of information is lost for a variable, which might be an important
explanatory variable.
If I use my boat 51% of the time in freshwater and 49% of the time in coastal
waters, this question says it is not important to know that I use my boat 49% of
the time in coastal waters. Many studies have identified boat ownership as an
important variable in explaining use and use value. Conditioning use to primarily
used is potentially losing important information.
A similar problem exists with Q4. Conditioning the identification of activity to
the most use looses potentially important information. From 1987 to 1992, we
conducted the Public Area Recreation Visitors Survey (PARVS) at 50 coastal
sites from Maine to Washington. What we learned was that for coastal sites with
multiple attributes, people engaged in multiple activities. And, at very few sites
did a majority of users indicate that there was one activity that was the "main
reason for visiting the site" or the "main activity" they participated in during the
visit to the site. I would have changed to check all that apply and possibly
followed this up with estimates of the number of days of each activity over the
past 12 months in coastal waters. You could use responses here to identify the
most important use based on relative days of use.
5. A similar problem exists for Q9. Why limit information on trips to single day
trips? Are multiple-day trips of no value? In the Florida Keys, very few trips are
day trips. And almost all trips are multiple activity trips with no activity being
either the "main reason for the trip" or the "main activity" on the trip. Coastal
water quality is critical to the Florida Keys. In 1995-96, we estimated over 2.5
million visitors spent over 13 million days in outdoor recreation activity. I think
you are missing a significant amount of activity by limiting your analysis to day
trips. I would agree that, because you are limiting the current application to
residents of California, day trips will be a relatively high percent of trips, but I
still don't understand the logic of dropping this portion of total activity dependent
on coastal water quality.
6. It was never made clear why the trip questions were being asked. Will there be an
attempt to do a revealed preference set of trip models using the random utility
model approach? If so, was there any thought to designing a revealed preference
80
-------
approach consistent with the stated-preference approach i.e., a joint estimation of
RP and SP data?
Adamowicz et al (1994) combined revealed and stated preference methods for
valuing environmental amenities. In 1996, the Association of Environmental and
Resource Economists (AERE) workshop was devoted to this topic. Experts from
the marketing and transportation fields were invited to share their experiences
with combining revealed and stated preference data. Among the lessons learned
was that combining revealed and stated preference data yielded better predictions
of demand for a good or service across many types of goods and services. Of
course we don't ever actually observe consumer's surplus, so one makes the
inference that if we are predicting demand for a good or service better, we are
estimating consumer's surplus better. I didn't see any mention of this in the
paper.
7. The empirical results presented in the current paper are labeled as preliminary, so
I didn't take them too seriously. However, I did have trouble with some of the
interpretations of the conditional logit model presented in Table 2 on page 16.
Only one paragraph on page 15 is presented with explanations.
I was, at first, a little confused, but the interpretation seems to be that, for
individual attributes, a positive coefficient means these factors (e.g. age, male,
black and household size) increase the probability that a person will choose the
status quo and a negative coefficient means these factors (e.g. income, Hispanic,
eating seafood, and participation in recreation activities) increase the probability
that a person will choose one of the programs (i.e., moves away from the status
quo). I think a little more explanation would help here.
8. I never saw the dollar amounts used. How many values? What was the range of
values? And, How were the range of values determined?
The Recreational Benefits of Improvements in New England's Water Quality: A
Regional RUM Analysis, by George Parsons, Erik C. Helm and Tim Bondelid
Comments:
1. The analysis uses data from the 1994-95 National Survey on Recreation and the
Environment (NSRE). I am and have been the Co-leader of NSRE since the early
1990's. It is stated in the opening paragraph of the paper "All models are for day-
trip recreation which account for approximately 77% of all water based recreation
trips in the region" (region being the Northeast region).
This estimate of day-trips accounting for 77% of all water based recreation trips
in the region is not correct. First, the trips were conditioned on an activity being
the primary purpose of the trip. As discussed above, NOAA's work through
PARVS revealed people are often not willing to say that their trips to coastal areas
were based on any one activity being either the main reason for the visit or the
main activity on their visit. So, the trip data obtained in NSRE 1994-95 was only
a sub-set of the total number of trips. Second, even though NSRE 1994-95
included both day-trips and multiple-day trips, modeling was limited to day-trips.
On this latter sub-setting, our profession seems to find multiple-day trips to be
81
-------
difficult to implement with the random utility model or that modeling multiple-
day trips requires a separate model. Multiple-day trips are a large proportion of
total use in the coastal areas. We need to model multiple-day trips that are not
conditioned on one activity being either the main reason for the trip or the main
activity or we are not accounting for much of the use. Above, in my comments on
the previous paper, I said that in the Florida Keys multiple-day trips, with no one
activity being either the main reason for the trip or the main activity on the trip,
account for almost all trips. Coastal water quality is important in the Florida Keys
and we need models to deal with these issues. This is less a criticism of this paper
and more a challenge offered to our profession.
2. Paper Caveats. " Using finer measures of water quality in the RUM model
persistently led to complications in the econometrics, usually a model that failed
to converge". I think this is probably related to caveat number 3 that"there is no
coastal water quality simulation model per se". At NOAA, we currently have a
project on estimating the value related to water quality changes in Southern
California. In this project, we have ambient water quality measures for each
beach on each day. Water quality is statistically significant in all models
estimated, including full information maximum likelihood estimations. Matching
up better water quality data to NSRE data is something for future research.
Valuing Water Quality Changes Using a Bioeconomic Model of a Coastal Recreation
Fishery, by Matt Massey, Steve Newbold and Brad Gentner.
Comments:
1. Overall this is a very impressive effort. The underlying model seems sound.
However, I don't think the actual application matches up with the model
presented in equations 1-4. A bioeconomic model that doesn't explicitly model
total effort and the institutional structure underlying the human system isn't a real
bioeconomic model. The interplay of the biological system and the human
system are fundamental. The use of calibration to account for the human system
is quite clever, but it leaves me with not much confidence in the result.
If the fishery management situation in place can be described as a common
property resource with an open-access fishery, then I think we would predict that
there would be "no benefits" realized from water quality improvements. The
commercial and/or recreational fishermen would dissipate any benefits. I saw no
discussion of the current institutional arrangement in the fishery selected for
application of the model.
2. As far as I could determine, equation 3 (trip demand) was never estimated.
Instead, a stated-preference model was implemented that only partially accounts
for changes in trip demand. Again, a bioeconomic model that doesn't explicitly
model total fishing effort is not much of a bioeconomic model.
82
-------
Comments on Session II
Cleaning Our Coastal Waters:
Examination of the Benefits of Improved Water Quality
Nancy Bockstael
INTRODUCTION
The term "ecological benefits" often conjures up images of obscure and indirect
pathways through which ecosystems affect humans. These are pathways that are difficult
to define, and for which related behavior is difficult to observe. Yet water-based
recreation, the target of much past non-market valuation activities, remains an important
pathway through which ecosystem health affects humans. As such it deserves continued
study.
The three preliminary analyses in this session are quite different in the commodity valued
and the type of data relied on, but all three focus on measuring the benefits of water
quality through traditional recreation pathways and all three use a random utility model
framework to model choice and estimate welfare measures.
In what follows I will try to point out what I think to be some vulnerabilities in the
current preliminary versions of these analyses. While my comments may seem diffuse,
I'll attempt to organize them around two general themes:
• How is the environmental quality variable measured and how is it incorporated
into an underlying model of individual preference revelation?
• Is the choice behavior underlying the use of the random utility model made clear
and are welfare measures consistent with this model?
DEFINITION AND MEASUREMENT OF THE ENVIRONMENTAL QUALITY
VARIABLE
Owens and Simon,' The Value of Improvements to California's Coastal Waters:
Results from a Stated-Preference Survey'
The first of the three papers, the one by Owens and Simon, seeks a means of valuing
water quality improvements in coastal waters. Contrary to what is implied in their
introduction, there have been many previous attempts to do this for specific estuaries or
other limited geographical extents, mostly using revealed preference data. But there is no
systematic treatment of benefits from coastal water quality improvement that can be
transferred to other areas and used to evaluate EPA's water quality policies at the national
level. My sense is that a systematic, transferable type of analysis is the ultimate goal of
Owens and Simon's work. Although this particular study targets the coastline of
California only, this coastline represents a large share of the coastal waters of the US,
making the geographical extent of this study quite a bit larger than most salt water
recreational studies.
83
-------
The intent is to estimate the benefits from coastal improvements through a stated
preference exercise. Most of the paper describes the careful pre-testing process through
which a well-designed survey has emerged. I have no particular expertise in survey
design and will leave the critique of this survey to others, but one aspect of the survey
troubles me. I am unclear as to environmental commodity being valued. The commodity
might appear to be quite specific; it is an increase in the miles of California coastline that
are rated "good" in terms of being safe for swimming, producing fish and shellfish that
are safe for eating, or supporting habitat for "large numbers of fish, birds, mammals and
plants."1
While there are clearly ambiguities in the last definition, the really troubling aspect of the
valuation question seems to be that it does not specify where along California's extensive
coastline these improvements would take place. This information would not necessarily
be very important if the authors sought to reveal respondents non-use values for
improvements in the health of ecological resources. And indeed this may be what is being
sought in questions about increases in ratings of habitat. But the questions related to
safety of fish consumption and ratings for swimming would certainly appear to relate to
use. This interpretation is further supported by the large number of use-related questions
asked in the survey, which suggests both to the survey respondent and to the rest of us
that recreational use values are of interest to them. Yet how can a respondent give a
credible use value answer to a question framed with no locational information. This is
directly contradictory to the premise of travel cost models that use behavior in the face of
varying travel costs to reveal demand curves whose estimation gives us consumer surplus
answers. In that model, the distance to a recreational site represents a cost that cannot be
counted in the surplus measure. The presence of travel costs plays a major role in the
model and accounts for much of the resulting variation across people in valuation
measures.
The authors ask follow-up questions about whether the respondent thought the
improvements would take place in a specific part of California - and if so, where. This
information may shed some light on what respondents were thinking when giving their
answers and may even give the authors a way to untangle the problem. It is not so
important that use and non-use values be estimated separately, but it is important that a
cogent story of what is being valued can be told. For example, if people tended to
respond that they did not think about where the improvements would take place, I'm not
sure I would know how to interpret their bids. And if they responded that they thought
these improvements would occur many miles from their home, then I would wonder if we
were missing a portion of benefits attributable to use. Finally, if they assumed
improvements would occur close to home, are we left with no measure of the benefits of
1 The added miles are represented in terms of "percent increase" but the authors use this term loosely, no
doubt in an attempt to make the question understandable. But it can easily convey the wrong idea to
respondents. The authors appear, for example, to label a change from 40% to 50% of the coastline as a
10% increase. Perhaps wording this as an increase of 10 percentage points would be both more accurate
and still understandable.
84
-------
cleanup for non-locals? Any effort that can be made to resolve this ambiguity in the
location of clean-up will be well worth it.
Parsons, Helm, and Bondelid,' The Recreational Benefits of Improvements in New
England's Water Quality: A Regional RUM Analysis'
A second paper in this session, the one by Parsons, Helm, and Bondelid, is a heroic
attempt to use existing information from past surveys to value water quality
improvements in New England. I use the term "heroic" because this is a very difficult
thing to do, and yet the returns from doing it well could be enormous. The potential
contribution of this paper is in developing a means to use the data from the 1994 National
Survey of Recreation and the Environment and water quality simulations from the
National Water Pollution Control Assessment Model of RTI to generate systematic and
comparable benefit estimates of water quality improvements for water resources
throughout the U.S.
In this paper, recreationists are viewed as choosing among recreation sites represented by
all lakes and stretches of riverfront and coastline within 200 miles of their home. Four
water quality measures - biological oxygen demand, dissolved oxygen, total suspended
solids, and fecal coliforms - are generated by the simulation model for each of the over
25,000+ sites within New England. Assuming such simulated measures are accurate,2
we are still left with the question: by what means do recreationists perceive these water
quality measures? Recreational modelers have long been concerned about possible
discrepancies between the dimensions of water quality that can be measured objectively
and those that people can perceive or learn about. Some signals may well connect
(however loosely) the objective and perceived measures, but the form the connection
takes must be thought through carefully. In this paper, the four objective measures are
converted into one variable that takes on only three levels. This does not necessarily help
the correspondence between objective measures and perception, since the thresholds
chosen may have little to do with how people perceive water quality differences. The use
of this one "tri-nary" variable is brought into further question, since it is considered
equally applicable to swimming, fishing, viewing and boating decisions.3
Economists are continually reminded by statisticians and econometricians4 that
correlation is not causation. Put another way, unless we are fairly certain we have
controlled for unobserved heterogeneity in our data, our econometric results may be
reflecting the effect of omitted variables that are highly correlated with the variables of
interest in our models. The paper by Helm, Parsons and Bondelid would seem
especially vulnerable to this accusation. The alternative sites vary only in terms of travel
2 We are told that the measures for coastline are likely inaccurate because the measures are extrapolated
from the nearest river mouth. This will be especially troublesome for swimming which does not tend to
occur in such areas and will usually underestimate water quality because pollutants from rivers will not yet
have been diluted by ocean currents.
3 This is in direct contrast to the Owens and Simon paper that attempts to convince people that different
environmental quality criteria matter for water to be ranked safe for swimming, safe for producing edible
seafood, and good for fish, bird, mammal and plant habitat.
4 By this I mean the work often referred to as quasi-controlled experiments, matching, or exploiting
regression discontinuities.
85
-------
cost, the set of two dummy variables signaling whether the site is a lake, river or
coastline, and the set of two dummy variables signaling the level of the 'tri-nary' water
quality variable. One distance related variable and two categorical measures are hardly
sufficient in describing the multiple differences in the vast number of lake, river and
coastline recreational sites within 200 miles of any individual5. Given the size of New
England, a very large proportion of the 25,000+ sites will be within 200 miles of most
individuals. But the geographic size of New England is misleading, as characteristics of
sites that can be expected to matter to people vary dramatically over its range, even
holding the category (river, coast or lake) constant. Water temperatures and local
amenities, to name only two considerations, will be drastically different over sites. How
can we possibly interpret with any confidence the coefficients associated with the simple
categorical water quality measures when so much is left out of the model?
The failure of attempts to model recreational decisions in the more logical nested
framework, as well as the failure of more complete site descriptors to generate usable
results, suggests a certain instability. It also suggests the likelihood that the model is
missing something important. It would be especially illustrative if the water quality
levels could be mapped. This might reveal the types of omitted variables (especially those
that tend to vary spatially, such as water temperature, fish species, etc.) that need to be
controlled for in making sense of this problem. With out a careful consideration of what
is being left out of this model, we can have no confidence that the significant coefficients
are reflecting any response to water quality variation at all.
Massey and Newbold,' Valuing Water Quality Changes Using a Bioeconomic Model
of a Coastal Recreational Fishery'
The third paper, by Massey and Newbold, uses contingent rather than revealed behavior
and draws on an already existing study rather than a new data undertaking. The
particular appeal of this paper is its attention to the pathways by which changes in water
quality affect recreational fishermen. In this sense, it is a particularly appropriate paper
for a workshop on the Valuation of Ecological Benefits.
In this paper, the water quality variable of interest from the perspective of policy is
dissolved oxygen (DO). In their conceptual model, the authors consider how water
quality affects stock abundance (given population dynamics) and catchability at any site
(given that fish may migrate to avoid areas with low dissolved oxygen). They also allow
for the fact that water quality might directly affect site desirability. By giving careful
attention to the biological modeling, interesting non-linearities and thresholds are induced
so that dissolved oxygen measures affect recreationists' decisions in realistic ways. The
result that DO appears only to affect fishermen through its affect on stock abundance has
interesting implications. If this is true, then effects will only be realized in the long run
and attempts to pick up such effects with a simple behavioral model including some
simple measure of current DO linearly will miss the point. I do not have the expertise to
5 Given the size of New England, all 25,000+ sites could be within 200 miles of some individuals and most
individuals will have an enormous number of site alternatives defined this way.
86
-------
comment on the quality of the biological modeling, but the spirit of this research seems to
be just exactly what we need.
EXTRACTING WELFARE MEASURES FROM RANDOM UTILITY MODELS
The random utility model has become the workhorse of environmental valuation. These
types of choice models are the rule rather than the exception in revealed preference as
well as stated preference data analysis. The random utility model can be a very plausible
model of behavior, since individuals often choose among discrete alternatives. In its
simplest form, both estimation and welfare measurement are easy to accomplish,
reducing barriers to its use. In fact, the random utility model has become so ubiquitous in
the literature that I wonder whether it is not often treated too cavalierly.
All three papers in this session use a random utility framework, although one paper
models stated preference responses, another contingent behavior, and a third revealed
behavior. There is a tendency in these papers (as well as others in the literature) to
reduce the underlying theoretical model's discussion to a boiler-plate presentation ending
in the usual formula for calculating welfare measures from estimated coefficients.
Paying little attention to the details of getting welfare measures from these models would
be OK, if it were not for the fact that we know welfare measures in these models are
sensitive to the details of the problem. Herriges and Kling recently compared welfare
measures derived from a random utility model applied to exactly the same data but
incorporating different functional forms and different variants of a measure of
environmental quality. They also compared welfare measures across different linked
models, again with different functional forms at the two stages. The results are quite
startlingly different and suggest that the devil is definitely in the details.
With this is mind, let us quickly review the models estimated in the three papers. Owens
and Simon estimate the parameters of a conditional indirect utility function based on
stated choices among hypothetical programs that would improve different amounts of
coastline in exchange for different tax payments. With parameter estimates in hand, the
authors indicate that future welfare measurement will be based on the formula:
In exp(X//?) - In exp(X/^)
j l
( ) Ptax
where the X's are explanatory variables from the random utility model, the J3's are
estimated parameters, ptax 1S the coefficient on the tax variable, and X" and X' are the
values of the explanatory variables given the status quo level of water quality and those
resulting from projected improvements. The formula and definitions of the X's are
difficult to square with the earlier definitions of the j subscripts which were defined as
indexes of different programs (water quality improvements and public expenditures?).
What is the summation really over?
87
-------
An additional problem arises in the use of ptax, the coefficient on the tax (i.e. cost)
variable in the model. The authors refer to this as the marginal utility of income. The
well-known formula for compensating variation in the context of a random utility model
is really given by:
In ^ exp[v(y - pj, q- )] - In £ exp[v(y - pj, q}0)]
CV = —L L
Py
where v is the conditional indirect utility function (conditional on the discrete choice
made), qf and q/ are the initial and subsequent levels of environmental quality in
alternative j, y is income and pj is the cost of alternative j. This is the definition of CV
only if the errors are additive and Type I extreme value and the income minus price term
enters linearly into the conditional indirect utility function. If the income minus price
term does not enter linearly, then the welfare measure has no closed form solution
(although it is possible, but difficult, to iteratively solve this problem.) Most important,
the nature of the underlying random utility model is such that income minus price appears
in the model as one term. And it is only because of this feature of the random utility
model that the coefficient on price (or cost or tax payment) has the interpretation of the
marginal utility of income. This fact accounts for its prominent place in the CV formula.
In their preliminary data analysis, the authors estimate a model with both price and
income included separately. Space does not allow a complete discussion of the
implications of this, but at the very least this compromises the interpretation of any
measure such as the above since we have to ask: if the coefficient on price is (minus) the
marginal utility of income, how do we interpret the coefficient on income? It may well
be true that we expect substantively different behavior from different types of people and
those types may be well proxied by income. But such a story requires telling and the
source of differences in response needs to be made clear. If the different responses are
truly due to income effects, then the simple linear form of the random utility model is
inappropriate since it implies constant marginal utility of income over the range of the
choices. We can't have the story both ways and some reconciliation is necessary.
Admittedly the authors' analysis was purely preliminary and done without time for
thought. Subsequent analysis will no doubt give much more careful treatment to the
underlying theory, the underlying behavioral model, the role of income and the resulting
welfare measures.
Interestingly, Massey and Newbold also include income separately from price in their
model but interpret the coefficient on price rather than the coefficient on income as the
marginal utility of income. Here again the underlying theoretical model is a boiler-plate
presentation and not particularly relevant to the problem at hand. Massey and Newbold
use a repeated nested logit model so as to be able to treat the responses to four different
contingent behavior experiments in a consistent framework. Parameters are assumed to
vary randomly across respondents (as in Train's mixed logit models) but to remain
constant over multiple responses of the same individual. Since Herriges and Phaneuf
88
-------
have investigated this type of model with an error components interpretation, the authors
may wish to compare their approach to that of these other authors.
While the idea of applying a repeated nested logit model to the contingent behavior data
is an intriguing one, the authors realize that the data really will not support this
interpretation. The repeated nested logit was developed as an internally consistent means
of capturing the participation as well as the site choice dimension of the recreation
decision. However, the contingent behavior experiment, which does allow the
respondent to "opt out" of the choice experiment by taking no trip, is fundamentally
different from a recreationist's day-to-day decision about whether to take a trip or not.
Treating the contingent behavior responses as if they mimicked people's day-to-day
recreation decisions is misleading. The contingent experiment contains an implicit
assumption that the individual would be free to pursue recreation on every choice
occasion. No information about weather, work obligations, etc. are explicitly or
implicitly introduced.
More attention to the underlying theoretical model is given by Parsons, Helm and
Bondelid who base their linked model on an approach originally suggested by Hausman,
Leonard and McFadden. In this model a random utility model is first estimated and then
a price index of sorts is calculated using the standard log sum formula divided by the
coefficient on price. This pseudo-price index is included in a count model (Poisson
model) of number of trips. It is now well-known that none of the linked models are
internally consistent - they do not derive from a consistent theoretic model of utility
maximizing behavior. Since the few models that are internally consistent have other
drawbacks (e.g. inflexibility of functional form choice and difficulty of estimation of the
Kuhn Tucker model), this is not necessarily a bad choice of approaches as it might
approximate behavior if not be exactly utility theoretic. While linked models of this sort
have some appeal, deriving the welfare measure from the participation decision rather
than the random utility portion of the model is perhaps less appealing. As Smith and
Herriges and Kling have shown, the pseudo-price index is not a price and can not be
treated as such in the count demand function. Therefore, a consumer surplus measure
that must necessarily be based on such an interpretation of the price index is
questionable.
CONCLUSIONS
All three papers have the potential to contribute to the literature. Each has a particular
strength and each is a good beginning in analyzing benefits from water quality
improvements. In finishing these analyses the authors have several areas in which they
could make their papers stronger. One has to do with improving the link between
recreationists' water quality perceptions on the one hand and objective measures that
policy changes are likely to affect on the other. Clearly this link induces more
vulnerability in revealed than stated preference analyses, but even in the context of
contingent valuation or behavior, one needs to be sure that the policy variable links
clearly to a commodity the respondents are understanding and bidding for.
89
-------
So much attention needs to be paid to the careful acquisition of the considerable data
needed to accomplish these studies that the last steps of the benefit measurement
sometimes are taken for granted. But in the end, benefit measures depend on the details,
and a cavalier treatment of the random utility model will lead to indefensible welfare
measures. The fact that - unlike price changes, for example - welfare consequences can
never be observed even after the fact places a heavy burden on welfare economists to get
the underlying story right.
References:
Hausman, J. G. Leonard, and D. McFadden. 1995. "A Utility-Consistent, Combined
Discrete Choice and Count Model: Assessing Recreational Use Losses Due to Natural
Resource Damage". Journal of Public Economics, Vol 56, pp 1-30.
Herriges, J. and C. Kling. 1999. "Corner-Solution Models of Recreation Choice." In J.
Herriges and C. Kling (eds.), Valuing Recreation and the Environment, Chapter 6, pp
163-197. Edward Elgar, Northampton, MA.
Herriges, J., and D. Phaneuf. 2002. "Inducing Patterns of Correlation and Substitution in
Repeated Logit Models of Recreation Demand." American Journal of Agricultural
Economics, Vol 84 (4), pg 1076-1090.
90
-------
Summary of the Q&A Discussion Following Session II
Scott Swinton (Michigan State University)
Directing his question to Matt Massey, Dr. Swinton referred to the structure of the
bioeconomic model that was used, which, "like any system model, has boundaries." He
wondered, "If we were to try and open the boundaries a little bit further and understand
what drives the driving variable of your model, which was dissolved oxygen, I'm curious
about how changes in agricultural management can change water quality rather than just
saying 'Okay, suppose we have bad water quality, then what does it cost us?'" Dr.
Swinton closed by asking, "How would you extend that?" and "What affects dissolved
oxygen and water quality?"
Matt Massey (U.S. EPA, NCEE)
Dr. Massey acknowledged that the model currently "doesn't deal with that at all—we just
sort of assume water quality conditions and assume policy somewhere else affects them
and makes those changes, and we just run the changes through." He continued, "It's
conceivable we could add another step that would model agricultural use and residential
and commercial development that would allow us to simulate changes in dissolved
oxygen and runoff and those types of things." Admitting that "it would be a great thing
to do," Dr. Massey went on to explained that they "had a terrible time getting the data
together" just to do what they did. He said that with the recreation data they used, they
had to go with stated preference data rather than with the revealed preference data they
actually preferred because they just couldn't get people to cooperate in providing those
data. He clarified that Steve Newbold, one of his co-authors who was "the ecologist of
the group, had to throw his credentials around" to get much of the data that biologists
were reluctant to share. He closed by reiterating that the idea, though "possible" and
"interesting" is "kind of beyond the scope of what they're doing now."
Alan Krupnick (Resources for the Future)
Directing his comment to Nathalie Simon and Nicole Owens, Dr. Krupnick stated, "Some
of us who are working on stated preference techniques have two issues that come up with
your work, that we all wrestle with as well. One is the units of measurement for these
attributes." Citing specifically the unit of "percentage of miles that are changed," Dr.
Krupnick wondered whether, using information from focus groups or other observations
they have, the researchers "could comment on that particular measure—whether you tried
others—and how to communicate these things to people." He added that, "Some of
Nancy's [Bockstael] comments get to this as well, about sort of the location of where
these are and so on."
Dr. Krupnick went on to say, "The second part that struck me is I think that you have a
very abstract program for actually bringing about these changes, and I'm wondering if
people really were willing to accept that. It would be a good thing if they did, because
it's hard to come up with a program that is kind of transparent to people and doesn't get
91
-------
in the way of their responses and meet their protests. But I'm wondering if people really
did accept that because in all these programs you've got switches—you know, some help
swimming more than help aquatic health, and so on. I wonder if you get much push back
from people about: What are these programs? What are their components? How do we
know they'll work?"
Nathalie Simon (U.S. EPA, NCEE)
Dr. Simon addressed Dr. Krupnick's first comment by saying, "Basically, we did try
other units of measure. We started off looking at the number of miles, and then we also
tried in other versions of the survey both number of miles and percentage change." She
said they found that "people really were focusing on the percentage change and seemed
to like that better."
In addressing the second comment, Dr. Simon said she thought that for the most part
people were willing to accept the abstract program. She added that funding issues
seemed to be the more critical concern—"people didn't want higher taxes for any
reason." Dr. Simon closed by saying they are still working through that and still need to
clean up their data.
Nicole Owens (U.S. EPA, NCEE)
Dr. Owens said, "I think Nathalie is right—most people did seem to buy the scenario."
She expounded that in the survey and in their one-on-one interviews, they made an effort
to ensure that people understood that "sometimes it's different kinds of contaminants that
might be affecting one of these types of use." She concluded by saying that "it was easy
for people to see that it's possible to reduce or eliminate a contaminant that affects
whether or not they can swim" in a particular area or to conceive of some other single-
issue program, but they were not readily able to conceive of multi-concern programs,
such as those that also looked at fish population or some other ecological factors.
Spencer Banzhaf (Resources for the Future)
Dr. Banzhaf requested "more discussion related to Nancy's [Bockstael] comment about
putting income and cost in the model separately," and he stated his support for this
approach. He continued, "It seems to me there are two ways to think about what's going
on here," and he used a corollary example of modeling restaurant choice to make his
point. "We see richer people going to more-expensive restaurants and different kinds of
restaurants than we see poorer people. One way to model that is that rich and poor
people have different marginal utilities of income—rich people can afford to go to these
more-expensive types of restaurants. And so we'd model that in a certain way in the
structural model, by putting in that your income affects your restaurant choice. Another
possibility is that there's heterogeneity in tastes for different kinds of restaurants, and
there's something about class, as well as maybe race, or education, or other kinds of
internals we see—something about class and income that has certain kinds of people
sorting to different kinds of places. And that piece is just really a taste shifter." Dr.
Banzhaf closed by clarifying that he was not disagreeing with Nancy's broader point that
92
-------
"the devil is in the details and you have to pay attention to how you model it," but it
seems to him that the choice isn't quite so obvious.
Nancy Bockstael (University of Maryland)
Dr. Bockstael stated that had she had more time she "would have gone into this very
topic." She said she agrees completely that "income shows up as significant in a lot of
these models, in the sense that it's really proxying for education and preferences or
something like that. Moyer really has done some nice work on this, where he has viewed
it as that. He has treated people in different income ranges and allowed the coefficients
on various things to change in those ranges, so that you don't have the problem of
introducing income continuously in a model where it's not going to make any sense. You
know, you can't start with a utility theoretic model and then decide you're going to just
throw whatever in. You have to have a way of introducing that that makes sense, and I
think that the best way that it makes sense, from my perspective any way, is the way that
Moyer uses it—shifting the parameters discreetly but allowing marginal utility to be
constant over large ranges so that it's only a glitch at certain thresholds. But, I say it's
proxying something else."
Patricia Casano (General Electric Company)
Prefacing her comment by clarifying that she is "not an economist," Ms. Casano
addressed Nathalie and Nicole and said she was "struck by the results you put up
indicating that 25% of the survey respondents had used coastal waters for recreational
swimming; 10% had used coastal waters for fishing and that sort of thing." She said that
she wasn't questioning whether the numbers were right or not but stated that they
"seemed really low" to her and she was surprised by the indication that "less than a
majority, generally speaking, of the survey respondents used coastal waters for any of the
scenarios that you were looking at." Ms. Casano closed by asking, "Does that play into
your analysis of the results at all, and if so, how?"
Nathalie Simon
Dr. Simon responded by clarifying that "those variables were measuring recreation over
the last twelve months" only, not over a lifetime. However, she allowed that they still
might appear low to Ms. Casano. She continued, "We did include them in the initial
regression that we ran, so it does figure into that—it was part of that conditional logit
model. But, again, we're still exploring a number of different functional forms, and we
still have a lot of work to do in terms of our analysis."
Kerry Smith (North Carolina State University)
Dr. Smith expressed a multi-faceted concern with the issue raised earlier in Spencer
Banzhaf s comment. He said, "The first issue I want to raise is: What income? In the
second two papers, if I remember correctly, we had a repeated mixed logit and a standard
model. Well, when we repeat, that implies a certain number of choice occasions that are
93
-------
embedded in the no trip alternative. So, implicitly, we have to ask ourselves: What is the
relevant budget for each of those created choice occasions? And, I don't mean these
comments to be critical, because I've done the same thing—I don't know what to do. But
we've got a question of total income versus relevant income for the choice that you're
representing, and that's at least as important as how you introduce the income in the first
place, as well as assuming how many choices there are—and we just fabricate that,
typically."
Dr. Smith continued with his second point: "If we're going to say: Okay, income proxies
for something else about people, then we've got to begin to question: What the heck is
the travel cost coefficient? Because the way we can interpret that as the marginal utility
of income is based on a prior set of restrictions that we've already imposed to recover
that, so we're still in the scoop, I think."
"Of course the third issue that arises in these sorts of models is the link between time
horizon of choice and the implicit substitution assumptions we're making as we evaluate
those. This bears a little bit on what Bob [Leeworthy] was talking about: As we move to
other kinds of trips, we're going to get more and more into these kinds of issues. Now, I
don't think any of us has the answers, and we're not going to get the best model—the
question is judging how bad do we get, which is essentially, I think, what Nancy's
comment was: When we array all these models, how do we make a judgment about
what's important and what is not important for the use of the model?"
END OF SESSION II Q&A
94
-------
Valuation of Ecological Benefits: Improving the Science
Behind Policy Decisions
PROCEEDINGS OF
KEYNOTE ADDRESS: THE ECONOMICS OF ECOSYSTEM SERVICES
A WORKSHOP SPONSORED BY THE U.S. ENVIRONMENTAL PROTECTION
AGENCY'S NATIONAL CENTER FOR ENVIRONMENTAL ECONOMICS (NCEE)
AND NATIONAL CENTER FOR ENVIRONMENTAL RESEARCH (NCER)
October 26-27, 2004
Wyndham Washington Hotel
Washington, DC
Prepared by Alpha-Gamma Technologies, Inc.
4700 Falls of Neuse Road, Suite 350, Raleigh, NC 27609
-------
ACKNOWLEDGEMENTS
This report has been prepared by Alpha-Gamma Technologies, Inc. with funding from
the National Center for Environmental Economics (NCEE). Alpha-Gamma wishes to
thank NCEE's Cynthia Morgan and the Project Officer, Ronald Wiley, for their guidance
and assistance throughout this project.
DISCLAIMER
These proceedings are being distributed in the interest of increasing public understanding
and knowledge of the issues discussed at the workshop and have been prepared
independently of the workshop. Although the proceedings have been funded in part by
the United States Environmental Protection Agency under Contract No. 68-W-01-055 to
Alpha-Gamma Technologies, Inc., the contents of this document may not necessarily
reflect the views of the Agency and no official endorsement should be inferred.
11
-------
TABLE OF CONTENTS
Keynote Address: The Economics of Ecosystem Services
The Economics of Ecosystem Services
Geoffrey Heal, Columbia University 1
Summary of Q&A Discussion Following Keynote Address 10
111
-------
Summary of Keynote Address: The Economics of Ecosystem Services
Geoffrey Heal, Columbia University
Dr. Heal opened by saying, "I want to talk about what I think of as an emerging area,
which I've loosely called "The Economics of Ecosystem Services," not because that's a
really snappy title but because I couldn't think of anything better. It's an area that's
attracting an increasing amount of interest both in academia and in policy circles, and
you've seen some evidence of that already so far today." He referred particularly to three
governmental committees that are examining the general area of economics in
ecosystems, two internal to EPA and one external at the National Academy of Sciences,
about which Mark Gibson had already given a presentation (see Session I). Dr. Heal said
he has had the pleasure of serving on two of those three committees that are at work in
this area of increasing concern.
Reiterating that this is basically a relatively new area of focus, Dr. Heal stated that it
dates back, as far as he can see, "to 1997 with the publication of a book, edited by
Gretchen Daily in the biology department at Stanford, called Nature's Services: Societal
Dependence on Natural EcosystemsSaying that "interaction between economics and
ecologists goes back further than that," he cited the Journal of Ecological Economics and |
"the Beijer Institute in Sweden, which also works on the economics/ecology interface and
has been doing this since about 1990. However, neither the economics community nor
the Beijer group, of which I'm a part, really focused on the concept of ecosystem
services, and I'm going to argue during my talk that that concept of ecosystem services is
really a very important one and is a rather powerful organizing concept. The introduction
of the concept has really made a difference in the way we think about things."
Dr. Heal went on to quote the following lines spoken by Teddy Roosevelt nearly a
century ago: "The nation behaves well if it treats natural resources as assets which you
must turn over to the next generation increased, and not impaired, in value." He went on
to ask rhetorically whether this was "the first statement of the importance of strict
sustainability—by a Republican president no less?" (laughter) He continued, "It's clear
that Roosevelt, interestingly, was thinking about natural resources as assets, and in fact,
as a form of capital, and that's an issue I want to come back to." Dr. Heal stated that this
new field of ecological economics is "possibly even a new paradigm," although he uses
that term "with great caution because it's hugely over-used, in many ways."
He continued by enumerating the various components of "society's capital": physical
capital (buildings, computers, etc.); human capital; intellectual capital; social capital;
"and last, but not least, natural capital." Dr. Heal said these all represent assets that yield
a return to society, "and they're all assets in which we can make an investment."
Focusing on the concept of natural capital, Dr. Heal sought to identify its components.
He stated, "Certainly since Hotelling's 1931 paper on the Economics of Exhaustible
Resources we've known that mineral resources are a form of capital. What Hotelling did
1
-------
in that 1931 paper was essentially to take a capital theoretic approach to the management
of natural resources, though I guess "capital theoretic" wasn't a current phrase in 1931."
He went on to remind the audience that the famous Hotelling Rule that came out of that
paper, which says that "the rate of capital gain on a resource should equal the rate of
interest, is essentially an asset management rule—a rule for efficient management of
assets which, incidentally, was developed by those researchers way before any general
theory of efficient management of assets." Dr. Heal identified "environmental systems,
as a whole" as another more-intangible type of important natural capital asset, and he
gave the example of lakes and rivers that are used to generate hydropower. He cited
Sweden, in particular, which gets "about 75% of its electric power from hydroelectricity,
so the Swedish system of lakes and rivers is a massive public utility and represent a large
fraction of Sweden's natural assets in the public utility area."
Dr. Heal said, "Extending this line of argument more generally we can think of
ecosystems as assets, as part of our natural capital stock." He reiterated that all forms of
capital provide services—they provide a return, "and the return that natural capital
provides is the services of natural ecosystems." He explained, "Now, there are two
concepts coming together when I make that statement: there's the concept of natural
capital from economics, and there's the concept of ecosystem services, which basically
comes from ecology. Ecologists, I guess, developed this concept of ecosystem services
as a way of characterizing how ecosystems matter to society, what services ecosystems
provide to society." Dr. Heal went on to identify some typical classifications of the
nature of some of these services provided to society by ecosystems: climate stabilization,
pollination and other assistances to food production, waste decomposition, recreation, etc.
He added, "There's a review of these in the National Academy of Sciences' volume that
Mark (Gibson) was talking about earlier," and went on to summarize that "ecosystem
services are the return on natural capital, and natural capital essentially consists of
ecosystems. The economic value of natural capital is obviously the present value of the
ecosystem services it provides."
Stating that this idea could be taken in several different directions, Dr. Heal clarified,
"What I want to do for the bulk of my talk, actually, is talk about the National Academy
of Sciences report and how it develops some of these ideas, but let me first take a little
digression into the area of sustainability, which has been an area of interest to me for
quite a long time." Allowing that there are a number of different ways of defining
sustainability, he said that most of the definitions "revolve around the concept of natural
capital, so I'm just trying to indicate that the concept of natural capital has applications in
a variety of areas. One way of defining sustainability is to say that sustainable income is
the interest on capital stocks—all of the capital stocks taken together." Dr. Heal added,
"That's a Hicksian concept," and reminded the audience that Hicks defined income as
"the maximum amount you can spend today consistent with spending the same amount
indefinitely into the future." He pointed out that "There's a concept of sustainability
right there in that concept of income that Hicks developed back in the 1930's, but if you
think about what that means, it really means that income is interest on capital, broadly
defined."
2
-------
Exploring more definitions of sustainability that support both weak and strong versions of
sustainability, Dr. Heal pointed out one in which "the weak version of sustainability is
policy that keeps the total value of all capital constant—preferably increasing, but at least
constant. So, non-decreasing value of total capital stocks is what is sometimes described
as weak sustainability. Non-decreasing value for natural capital alone is what is
sometimes referred to as strong sustainability." He stated that he didn't wish to go into
the merits or demerits of the various definitions, but was "just trying to emphasize the
point that natural capital, whose value is the present discounted value of ecosystem
services, is a key concept in discussions of sustainability." He also added that "one of the
interesting consequences of keeping a non-decreasing total value for all capital stocks is
that it implies the present discounted value of future welfare is non-decreasing."
Returning to the issue of ecosystem services, Dr. Heal commented that ecosystem
services are frequently public goods (such as those he had mentioned previously: climate
stabilization, pollination, etc.). Furthermore, he stated that "a great majority of them are
non-market goods, so when it comes to valuing them, this raises some questions, but
questions that are fairly conventional in the field of environmental economics—questions
which are, in fact, the lifeblood of environmental economics." He pointed out one aspect
of ecosystem services which is "certainly rather distinctive, and that's that there is
frequently a considerable amount of uncertainty about the functional relationship between
the state of an ecosystem and the services that it provides."
Switching to a discussion of "the National Academy of Sciences report (the National
Research Council report) and how it addresses some of these things," Dr. Heal said that
the report starts off by "classifying the various ways in which ecosystems and ecosystem
services can have value." He described this as a "conventional classification into use and
non-use values, with a sub-classification of the use values into direct and indirect values"
and added that "there's a two-way classification which is central to the report. One is a
classification of the types of values that ecosystem services can have. The second,
obviously, is a classification of how you can go about valuing them." Emphasizing that
this is all fairly standard economics, he identified the optional ways to value them: "with
revealed-preference techniques, with stated-preference techniques, or with some
combination of the two." He added that in writing the report, he and the others spent
some time "trying to work out when one or the other is more appropriate and which of
the various techniques is more appropriate for which particular types of services." He
also stated that "the discussion of these issues in the report does address some of the
issues raised by the NOAA Blue Ribbon Commission on Contingent Valuation and some
of the critics of the CV approaches there. I don't think we have anything enormously
original to say about that, but I think there's quite a clear integration of the literature on
that within that section of the report."
Dr. Heal identified one of the key questions that they focused on in the report is "how the
services provided by an ecosystem (i.e., the services provided by natural capital) change
as the ecosystems are impacted by human activity." He presented the example of how
the extent of mangrove swamps and other types of coastal wetlands affect the
productivity of offshore fisheries and identified the pertinent questions as: "What exactly
3
-------
is the functional relationship there—how much will a change that we make in the extent
of coastal wetlands affect fisheries and on what sort of timescale?" As another example,
he brought up an issue that he has been involved with: New York City's decision to
conserve the Catskills watershed. The primary question they have dealt with here is:
"How does the extent of a watershed and the nature of the vegetation in that watershed
affect the watershed's ability to provide ecosystem services?" He identified the two
"critical ecosystem services" that most watersheds provide as water purification and
stabilization of stream flow and said, "If you're thinking about the conservation of a
threatened watershed because of the value of those services, then it's actually quite
important to have some understanding how different ways of using that watershed and
different levels of human impact on that watershed will affect the provision of those
services." Ideally, he said, you're looking for some kind of functional relationship
between the state of the watershed and the services it provides.
Dr. Heal went on to say that "we don't have to answer that type of question if all we want
to do is to value the current services of ecosystems, but if we want to value changes in
the services that result from extended human activity or from policy intervention, then we
do have to answer these sorts of questions about what's the nature of the link from the
physical characteristics of an ecosystem and the extent of the ecosystem and the human
intervention in the ecosystem through to the services that it provides." For emphasis, he
repeated, "If we want to value the change in natural capital which comes from the
destruction or the conservation of a system like a watershed or a wetland, then we have to
be able to answer those kinds of questions." He went on to state that "the biggest
challenge that we face here is linking changes in the bio-geo-chemical state of an
ecosystem to a change in the service flow," and he said that the NAS report pushes quite
hard for more of the integrated economic and ecological modeling that is required to
address this.
He continued, "What we really need here, ultimately, is what I might loosely call an
ecological/economic production function, which is a function that has ecological
variables as its domain and economic variables as its range. Basically, you would then
perform economic analysis on that production function—you want to be able to
differentiate that production function and find the marginal productivity of this type of
change in the vegetation, this type of change in the extent of the area, etc." Dr. Heal
stated that with this marginal productivity, you could then conduct policy analyses. He
went on to explain that ecologists characterize ecosystems in terms of their structure and
their functions, and he clarified structure as meaning "a description of the things that are
in it—the species, the number of each species, the structure of the soil, the climate, the
vegetation, and things like that." He clarified functions as "the flow of energy through the
ecosystem, the productivity of the ecosystem, and a range of variables like that" and he
said that "the ecosystem, acting through its structure and functions, produces ecosystem
goods and services, which are of importance to humans. As we said before, those
services have use and non-use values, consumptive and non-consumptive uses and so on.
Then, of course, human activities, in principle, have an impact on the structure of the
ecosystem and therefore affect the ecosystem services."
4
-------
"What we really want to be able to do is to go right through that system from the top
down to the bottom and say how a human action will affect the structure and function of
an ecosystem and, in turn, affect the extent of the goods and services provided, and
therefore the value of those goods and services provided. Then you can use that
calculation in a cost-benefit analysis to compare it with the alternatives available." Dr.
Heal admitted that this can be quite a complex thing to do, and it isn't easy to link the
economic and ecological models. He characterized ecological models as having "a habit
of being fairly complex," often involving non-linearities, thresholds, and irreversibilities.
He said that although these complications also existed in economic models, they seem to
be "more dominant and more central to the true characteristics" in ecological models.
To provide an example of "what you run into when you try to do this type of stuff," Dr.
Heal brought up the case of Lake Mendota, a lake beside the campus of the University of
Wisconsin-Madison, which he termed "the most widely studied lake in the world, by far."
He cited studies that looked at the eutrophication of the lake and estimated that 70% of
the fertilizer applied to the farmland surrounding the lake actually ends up in the lake.
The high level of phosphorous in the fertilizer causes the lake "to sort of switch states,
biologically speaking, and become eutrophied. There's a huge reservoir of phosphorous
in the sediment at the bottom of the lake, and under certain conditions this phosphorous is
released into the lake water, causing a sudden pulse in the water's phosphorous level."
He went on to explain that while the amount of phosphorous leaving the lake by means of
an outflowing stream is directly proportional to the concentration of phosphorous in the
lake, the inflow is more complicated. There's a basic rate of phosphorous inflow, which
is set by the rate of fertilizer use by the farmers on the adjacent land and the rate of
rainfall, but "once the concentration of phosphorous in the lake water reaches a certain
critical level, phosphorous is released from the sediment into the water and you get a
sudden increase in the rate of phosphorous inflow into the lake because of that. So, you
end up with a sort of S-shaped relationship between phosphorous concentration and
phosphorous inflow because of that pulse."
Dr. Heal went on to identify different equilibrium points along the relationship curve. In
particular, he pointed out a lower point, at which the lake was healthy and usable, and a
higher point, at which the lake was eutrophied. He pointed out that a sudden heavy
rainfall can "kick" the phosphorous concentration from the lower, normal equilibrium
value up to the high-concentration equilibrium value, where the lake is eutrophied, and it
can be very difficult to move the phosphorous concentration back once it has been
elevated in such a way. He concluded, "The point here is not to give you lectures on lake
ecology, but to illustrate the complexity of these ecological models and the complexity,
therefore, of the linked economic/ecological models, because the services that this
ecosystem provides depend on which of these equilibria we're at. At the lower level, it
can provide quite a high level of services; at the higher level, on the other hand, it
provides a much lower level of services. The relationship between the inputs to the
system and the ecosystem services it provides is actually given by a quite complex
dynamic process where what's happening today depends not only on the inputs today but
on a whole history of past inputs. This makes it quite difficult to write down the kind of
5
-------
production function I was describing before, and in ecology this kind of thing is quite
common."
Dr. Heal provided another example based on the responses of watersheds to oxides of
nitrogen, citing a study done on the Catskills by researchers at the Institute for Ecosystem
Studies. In this situation, the water bodies' natural ability to buffer the effects of the
deposition of oxides of nitrogen keeps the chemistry of the water at a steady state until
the buffering capacity is exhausted—then there is a sudden change in the chemistry of the
water, producing a relationship between the inputs and the outputs which is highly non-
linear and which also depends on the history of past inputs rather than just on current
levels of inputs. He summed up the situation by saying, "While I think we definitely
need to link the economic and ecological models, it's complicated and it's understandable
that it hasn't been very extensively done to date. There are a small number of good
examples, but that number really ideally should be much greater."
Dr. Heal commented that Chapter 5 of the report presents some of the examples he has
referred to and pulls together "a whole range of case studies which try to integrate
ecological and economic thinking in the valuation of ecosystem services." He further
clarified that the chapter begins with "some relatively simple cases involving a single
service provided by an ecosystem—the decision on policy issues is made on the basis of
a single service, usually something to do with water," for example drinking water, flood
control, and fisheries. Then the report goes on to look at more complicated examples
representative of ecosystems that provide many different services "all of which matter for
the policy decision, and therefore you have to worry about valuation of all of the services
and about the impact of human activity on the provision of all of the services." Dr. Heal
stated that there are some "really quite good case studies" that focus simply on single-
service situations, but when you progress to the more common multi-service situations
"there is regrettably a paucity of really well-worked case studies." He cited the Lake
Mendota example as one that "has been very well worked with some really effective
integrative studies."
Saying that the last topic he wanted to address was "the issue of uncertainty," to which an
entire chapter of the report is devoted, Dr. Heal said, "I think it's implicit in what I've
said so far that in any attempt to link economic and ecological modeling there will be a
significant level of uncertainty in the final output." Though this is always the case in the
statistical analysis of economic studies, he said it's "particularly pronounced" in the type
of situations being discussed. He stated that's one of the reasons why in the report they
emphasize the need for a sensitivity analysis, and they recommend "both conventional
sensitivity analysis and also Monte Carlo analysis when the data are sufficient and the
opportunities are available for that."
Dr. Heal said that in the report they also "talk at some length about option values, which
are very important in this context." This is because we're dealing with ecosystems in
which there are "potentially significant irreversible changes" due to human activity while
at the same time being uncertain about the consequences of the changes. However, over
time, we may learn something about these consequences. As Dr. Heal stated, "That's a
6
-------
classic situation for the existence of a quasi-option value." He explained, "Quasi-option
values (or just option values, for simplicity) are associated with situations where there's a
potentially irreversible change and you don't know the full consequences of that change
although you may learn about the consequences of that change over time. Then what the
theory of option values tells you is that there is a real merit, or advantage, to maintaining
a flexible stance and using the available time to learn more about the importance of the
system that you're thinking about conserving or changing." Dr. Heal said he and his
colleagues noticed that "there are actually no studies at all of the significance of option
values associated with avoiding irreversible changes in complex ecosystems." He added,
"Let me emphasize the issue again, just in case I didn't make this clear: When you're
looking at the costs and benefits of changing an ecosystem and making potentially
irreversible changes, then on the benefits side associated with conserving the system you
should enter a number which reflects this option value, a number that reflects the fact that
if you conserve the system you can, in the future, revisit the decision on whether to
damage it or not when you have better understanding of the consequences of that. That's
what we call the quasi-option value—that's what Arrow and Fisher first analyzed in the
QJE [The Quarterly Journal of Economics] paper back in the 1970's."
Dr. Heal went on to say that none of the studies cited in the report look at option values at
all, and he added that he is not aware of any attempts by researchers in the field to
compute option values for their ecological/economic studies. Stating again that he
believes this is an important area for empirical research, he said it has left "a big gap in
our understanding of some of the numerical issues in the conservation of these
ecosystems." He added that when you talk about this type of uncertainty, ecologists
always raise the issue of adaptive management, which means "managing an ecosystem, if
it's possible to control it in some sense, in such a way as to actually learn about its
behavior—in effect, experimenting to some degree with the ecosystem so as to get more
information about the parameters of the system and how it responds in various ways."
Dr. Heal noted that the issue of adaptive management is dealt with in the report and went
on to say that from an economic perspective it is interesting that "there's an interaction
between this ecologist's concept of adaptive management and the economist's concept of
option value." He clarified by stating that "one of the things that gives a flexible stance
an option value in the face of a potentially irreversible change is that if you postpone
making the change, you get a better estimate of the value of making or not making the
change. If you can actually experiment through adaptive management, you can
potentially increase the value that you get from learning in a situation like that, so there
can be an interesting positive interaction between option values and adaptive
management."
A concept that "comes up naturally when you're talking about uncertainty in this
context," and one of the issues that Dr. Heal and his colleagues "discussed at some length
in the report is the issue of the precautionary principle." He briefly reviewed the history
of this principle, which was advanced in the Rio Declaration in the early 1990's, became
commonplace in European legislation on environmental conservation, and is often cited
by NGO's as an argument for not making certain types of change. Dr. Heal said, "I have
to say that, potentially slightly controversially, the committee saw little value added in
7
-------
the precautionary principle. It seemed to us that many of the issues that motivate people
to talk about the precautionary principle and lead them to advocate the precautionary
principle are in fact actually adequately captured in concepts that economists already
have—namely the concepts of risk aversion and option value." Dr. Heal said, "If you
approach a decision from the perspective that society may be very risk averse, and
particularly may be very risk averse about making irreversible changes (and this is
captured in the concept of option value), then I think actually there's little value added by
using this so-called precautionary principle. Much of it is already there in the body of
economic thinking about decision-making under uncertainty, but we haven't done a
terribly good job of articulating that connection."
Dr. Heal returned at the end of his talk to why he believes it's interesting and useful to
think in terms of ecosystem services. He believes that the ecosystem services approach
gives researchers a better handle on understanding why the conservation of natural
environments matters from an economic perspective. He said there are currently "some
big shortcomings in the way we go about this." He acknowledged that "we're very good
at talking about why pollution is bad for people's health, and a lot of the ways in which
we pitch the conservation of our natural environments is in terms of the impact on human
health." He also stated that we know that people have a willingness to pay for
conservation, for example on the issues of wilderness areas and threatened species, but he
added that "we're not particularly good at actually articulating in detail to a skeptic why it
matters that we conserve the natural environment and why it matters that we conserve
threatened species."
Dr. Heal believes that "thinking in terms of natural capital, and in terms of natural capital
as providing ecosystem services, which are a return on that natural capital, can give us a
much better handle on explaining in detail why the conservation of the natural
environment works." He said that in his view, "One of the ultimate challenges in this
area is explaining why biodiversity conservation matters and why extinction matters.
Almost all environmental economists are personally concerned about the extinction of
species—it matters to most of you in this room that species go extinct." He raised the
question: "Is this purely a moral judgment?—Is this purely an aesthetic judgment?—Or
is there also a sense of an economic element in this as well?" He believes, "Thinking in
terms of ecosystem services does potentially give you a handle for analyzing your
concerns about extinction and about biodiversity loss in economic terms rather than in
moral or aesthetic terms." Dr. Heal was quick to add, "I'm not undervaluing moral or
aesthetic thinking at all, but it often doesn't have much impact on policy makers, I regret
to say. Economic thinking, on the other hand, tends to have much more impact on policy
makers."
In closing, Dr. Heal stated, "If you think about ecosystems providing services and about
the range of the services provided as a function of the biodiversity inherent in those
ecosystems and of certain species in those ecosystems being key to the way those
ecosystems operate, then you can get a different model of why it matters to conserve
species and to conserve biodiversity. That extra way of thinking—having that extra
8
-------
element in the economist's toolkit—I think is ultimately one of the most valuable
contributions of this type of approach to environmental economics."
"Thank you."
-------
Summary of the Q&A Discussion Following the Keynote Address
Nancy Bockstael (University of Maryland)
Dr. Bockstael commented, "Geoff, you had a slide in which you talked about the bio-geo-
chemical changes that might happen in the ecosystem and how they might affect
ecosystem services. There was quite a bit of detail on that slide, and then there was an
arrow that went around the corner, and it sort of stood for feedbacks. I think this arrow
hides a lot and raises an interesting question that EPA has asked me over and over again
in my work on land use, which is: If land use change affects ecosystems, what are the
feedbacks from the ecosystem back to human's decisions to change land use? My answer
is that there aren't obvious feedbacks that affect the demand for land use change. The
resulting ecosystem changes affect people through different pathways. The ways in
which we benefit from some of the improvements in these ecosystems or that we lose
from changes in the land use hit different people in different ways - through water
recreation or storm damage, for example. But there a logical feedback mechanism that
causes the land use change decisions naturally to adjust. We develop areas, we affect
stream health, we affect stream geomorphology—but none of that feeds back on the
demand for housing and the development decisions, so the public sector has a role here I
guess. I'm wondering if this issue of that feedback arrow and that the feedbacks aren't
clear came up in your discussions . . . and if they aren't clear, is that a pervasive thing—
and if so, are there any indications for what we do in this area? I know that's not a very
well-formed question, but. . ."
Geoffrey Heal (Columbia University)
Dr. Heal responded, "No, it seems a very good question, though it seems a very hard
question—and a very interesting question. I guess part of what you're saying there—and
I'm more re-phrasing the question than answering it, really—is that the impacts that you
and I and others have on ecosystems don't come back directly to us." Dr. Heal said that,
instead, our impacts "occur as external effects imposed on other people," who can
potentially be a long way away. He used the example of people in New York City who
"escape" on weekends to the Catskills and because of their activities while up there
"impose a negative external effect on people a couple hundred miles away. You don't
see it. So, one thing that comes out of this is that you need to choose the scale for
decision making very carefully. A lot of land use planning in the U.S. is carried on at a
very, very micro scale—and these data are too small to capture for many of these effects.
So, that's another reason why I'm integrating the valuation of watersheds into a valuation
of Chesapeake Bay, for example—it gives you a chance to operate on a scale big enough
that you capture some of these effects and you can bring them back into your position
paper." He offered the further example of another study he was involved with which
looked at pollutant accumulations in the Gulf of Mexico that came primarily from the
Mississippi River, representing a drainage basin of almost half the continental U.S.
He continued, "The arrow that you were talking about was really designed to indicate that
human activity impacts ecosystems . . . and that impact is not well understood. Even if
10
-------
you know, for example, how much development has occurred in an otherwise pristine
ecosystem, ecologists can't really tell us very much precisely about how that impacts the
ecosystem's watershed population, for example. There's no simple functional
relationship between the amount of pollutant in the Catskills and the quality of the water
that New York City gets."
Dr. Heal concluded by stating that part of the underlying problem here is that "we just
need a lot more research on how human transformation of ecosystems affects the services
they can provide—but this has been a very complicated relationship" which is often not
easy to see "because the cause and the effect are spatially quite separate."
Nancy Bockstael
Dr. Bockstael interjected, "I would add that sometimes they're temporally quite
separate."
Dr. Heal
Dr. Heal responded, "Yes, you're quite right. With species extinction, for example, there
are a lot of species around that the ecologists like to call the walking dead." He explained
that these are species whose populations are low and population/genetics modeling
indicates that they'll become extinct at some point "but they may be around another 50 or
100 years before the last one dies. So, there can be long time lags between the necessary
conditions for extinction being in place and the actual extinction occurring."
Marca Weinberg (U.S. Department of Agriculture, Economic Research Service)
Posing a follow-up question to Dr. Bockstael's, Dr. Weinberg commented, "What we
struggle with a lot is the linking question that you started with—how do you link policies
and outcomes?—in particular, the value of changes in the natural environment that are
initiated by the policies or that might be initiated by hypothetical policies." Saying that
she agreed that bioeconomic modeling and process modeling are really critical to
understanding these systems, she asked: "Since most policy is at the federal scale, how do
you design data collection efforts or modeling efforts to allow an assessment of the
benefits or costs of national-scale policies?"
Geoffrey Heal
Dr. Heal answered, "I'm not sure that the relevant policies are always at the federal level.
... To the extent that the relevant issues are land-use issues, they are often very locally
controlled, on a surprisingly small scale. That sometimes makes it harder rather than
easier because if you want to control the management of a watershed and it's a large
watershed, you may actually end up talking to 10 to 15 independent sovereign entities in
order to get their perspective on that." Dr. Heal continued by saying that in his view,
"that's actually a significant weakness in environmental protection. ... it would be
desirable to have land use decisions made on the basis of larger entities. Land use could
really be managed that way because there's a significant influence that way—there's the
potential for disaster, but there's also the potential for somewhat more straightforward
solutions . . ."
11
-------
Addressing the other issue that was raised concerning linking policies to outcomes, Dr.
Heal commented, "The point I was trying to make for part of my talk was that the link
between the physical nature of an ecosystem and the services it provides is very weak—
we don't understand that well. It's partly a question of collecting data, but it's also a
question of doing some modeling. There are actually quite a lot of data in the Heinz
Center that Tom Lovejoy runs ... a lot of data on the state of the U.S. ecosystems and the
way these have evolved over time. But, no one has tried to map that into statements
about services and to evaluate services to human communities. That remains to be done."
Marca Weinberg
Dr. Weinberg said she agreed, but she thinks "that's exactly the disconnect—the Heinz
data, by-and-large, is national scale and so it's not very helpful in informing decisions
that happen at the local scale. . . . We do have a lot of federal policies that affect resource
usage." She concluded by saying that she believes we could benefit from "some deep
thinking about how to develop models capable of informing those decisions."
Geoffrey Heal
Dr. Heal responded by saying some of that is being done in the area of non-point source
pollution, a major source of impacts on ecosystems.
Sasha Sud (Ontario Ministry of Natural Resources)
Mr. Sud said that he is presently working on a project looking at road development and
motors and how they affect ecosystem services. He stated, "Hypothetically, in trying to
value the impact of developing a road and seeing the impact on ecosystem services, I'd
say one of the services that we're looking at is water purification, and you're trying to
value the impact of how much less water purification takes place when you develop a
road over a wetland, for instance, and you disrupt the water cycle of that region. One of
the ways to value it would be to see how much less water gets purified—put a value to
it—and then value the ecosystem service based on the price of the water found in that
region." Saying that this price differs by region, he wondered what would be a good way
to put a value to an ecosystem service in a situation such as this.
Geoffrey Heal
Dr. Heal responded that "values of ecosystem services are invariably geographically
specific," offering the comparison of the value of water in the Sahara versus the value in
the Great Lakes area. . . ." He closed by saying, "If you can identify a relationship
between road construction and the nature of the watershed services, I think you will have
done very well. Of course, that's not an easy thing to do at all."
Avery Sen (National Oceanic and Atmospheric Administration)
Admitting that he is not an economist but is starting to work on the economics in social
sciences, Mr. Sen addressed Dr. Heal saying he enjoyed the presentation and that it seems
12
-------
"the conclusion that you draw, or what you observe, is that economics is slowly being
integrated into the rest of the natural sciences—physics, chemistry, and biology. As a
consequence, what I see is that there are going to be inherent limits to growth. There are
some people, I suppose, who won't like the idea that there are limits to growth imposed
by the physical world, and I'm wondering what obstacles you might see to your work"
and how those obstacles might be overcome.
Geoffrey Heal
Dr. Heal responded, "I guess some bits of environmental economics are about the extent
that we observe physical limits to growth. I think the standard economic response here is
that if you price [changes focus here]—well, there are different types of growth: there's
environmentally intensive growth and there's environmentally conservative growth. For
instance, there are different ways of generating energy—there are those ways that are
environmentally intensive and those that are not. Part of the problem we have at the
moment is that we just don't price environmental services right. You know, for the type
of growth that we have and the general type of economic activity we have, it's probably
excessively intensive in the use of the environment and excessively intensive in the
impact on the environment."
He concluded by saying, "I don't think that there are significant physical limits to growth
that we're about to bump up against—that is, that we have to bump up against and we're
about to bump up against—we don't have to. I think the reasons that we may possibly
bump up against them is not that there are real constraints on growth but that we're
simply not steering our growth in the right direction. . . . and we're not getting prices
right—we're not pricing environmental services appropriately. ... So, I don't see
physical limits as being a real issue. What I see [the need for] is thinking more smartly
and growing more smartly, just by considering environmental constraints."
Avery Sen
Mr. Sen clarified his position by saying, "I guess my question was more along the lines of
what perceptions might be to limits to growth and what effects the perceptions may have
as opposed to whether or not there will be limits."
Geoffrey Heal
Dr. Heal responded: "Well, a large fraction of the population receives no benefits
whatsoever from growth, as far as I can see." [laughter] He stated that his concern was
more with "getting people to start realizing that there might be limits to growth."
Clay Ogg (U.S. EPA, National Center for Environmental Economics)
Dr. Ogg commented, "One of the few successful T&DO's (time and displacement
optimization) of an agricultural wathershed, where they actually are claiming that they
reached their objectives, is in the Neuse watershed in North Carolina. They've basically
tracked what it is that accomplished their objective and it seems to be buffers and
following a nutrient recommendation, both of which would have been very relevant to the
13
-------
example that you used here." He went on to say, "The USD A, in its new Conservation
Security Program, pays farmers approximately 5% of their rent in exchange for adopting
a set of practices that are geared to reach a certain watershed objective or a certain set of
local objectives. It's a national program but it's keyed in to reaching what I think we
could call an ecosystem objective. What's happening is that their people are identifying a
very small set of changes that they ask farmers to make—very focused. . . . Your part of
the job is very, very difficult. The choices that agriculture has in terms of practical
options that they can actually carry out as part of the programs they have out there are
very limited. So, I think that having the whole team together is kind of critical given the
fact that, otherwise, the kinds of tasks you described here seem absolutely monumental. I
think we have to ground it in terms of: Here are our choices—let's try to figure out what
are the ecosystem benefits of carrying out what we can do here."
Geoffrey Heal
Dr. Heal replied, "I wasn't aware of that North Carolina case—it sounds interesting. It
all sounds pretty similar to what is being done in the Catskills watershed. They're
planning for . . . essential organic farming methods . . . and things like that. At some
point it will be interesting, when we get enough data on these things, to use these as a
study of what is the functional relationship between implementation measures like that
and the impact on the water quality and the stream flow and things of that sort."
(Unidentified, U.S. Forest Service)
Stating that the U.S. Forest Service routinely gets hit by lawsuits and challenges
regarding the studies they do and the documents they produce, he said he understands
very well the complexity of mapping ecological impacts or landscape impacts into
ecological outputs. He continued, "One of the things from your talk that I have a bit of a
quantitative issue with is that for us (the Forest Service), I don't think it's a question of
overvaluing economic outputs versus non-economic outputs—it's usually about jobs
versus ecosystem services, and jobs in some markets constitute a non-market value if you
look at it. My main question is: Given the complexity of some of these issues of trying to
go through the whole contingent evaluation process or perhaps a cost-benefit analysis
approach to decision making, there is an alternative model and that alternative model is
public participation in a broader circle of choice model. It can incorporate ecosystem
information into the social choice process. Given those two competing models, there
might be ways in which they could support each other, but there also might be ways in
which there apparently would be a conflict technological/technocratic models and social
choice models. Have you guys looked at those issues?"
Geoffrey Heal
Dr. Heal answered, "No, we specifically don't look at those issues in the report because
the mandate of the report was to look at the economic valuation of ecosystem services for
use in regulatory evaluation, cost-benefit studies, and things like that. I'm aware of what
you're saying—there are obviously two radically different alternative methods and . . ."
14
-------
The Questioner
Interrupting, the questioner said, "Given the complexity of what you're talking about,
especially when you're talking about project-level decision making and localizing the
decision making process, what sort of implications do you think it has for the
institutions?"
Geoffrey Heal
Dr. Heal continued, "I don't really see them as alternatives—I see them as operating
together. One of the things that studies of the type I was talking about here do is they
provide data for the political discussion and the political process. If there's a discussion
about conservation of a particular wetland, and an economic study suggests this is very
cost-effective or very cost-ineffective, then that study will impact the political discussion.
At certain levels [of the decision making process], the economic analysis is dominant, but
once you get to the very top of the decision making process it's almost always a social
choice political process. At that point, the economic variables are influential—it
becomes hard for politicians to argue against a very convincing economic case—but
they're not determinant."
Kerry Smith (North Carolina State University)
Dr. Smith referred to Dr. Heal's comments on "quasi options" and questioned whether
there actually had been a measure of quasi options. He said he felt the raw material was
there, and he cited Eric Helm's talk from Session II, which involved a retrospective study
which looked at scenarios designed to discover "what would happen if we did not take a
particular action and what was the string of benefits that were associated with that
action?" He said the reason he thinks that has the raw material to get at a quasi option
value is because one has to "pretend that that decision would be irreversible—you can't
reverse that outcome. That's one of the key elements—we want the additional
speculative value of what we learn as a consequence of not taking irreversible action."
He continued by asserting that in and of itself that ability to go back and to reconstruct
"what the past might have been . . . doesn't actually help very much in thinking about
future decisions"—it doesn't inform future choices. Dr. Smith then posed the question:
"Could we take the existing information that we've already developed in a range of
situations and design these scenarios in such a way that we could identify ... the
attributes of the uncertainty that's inherent in the decision process for which we have
already learned something? Then we could go back and say, "Okay, these look to be
important in that circumstance." He closed by saying, "That would not be a valuation
exercise when you're talking to any one person buy it would be a diagnostic evaluation
of decision processes—and it seems to me that's what you're calling for."
Geoffrey Heal
Dr. Heal responded, "Yes, that's a very interesting point—I'd have to think about that a
little bit, but that's a very suggestive idea." He said they'd have to go back and pry up
some data from 1972-1977, but judged that that would not be impossible.
15
-------
Kerry Smith
Dr. Smith continued, "Look at the information base at the time the decision was made,
and then compare that information base with what we learned as a consequence of the
activity and then design all the different attribute sets that would characterize the
uncertainty."
Geoffrey Heal
Dr. Heal responded, "You're right, it would provide us with data on the past choice, but
to me it would be interesting just to get some sense of the order of magnitude of these
option values. One of the things I find frustrating is not knowing whether these are
negligible or potentially quite significant, and I don't actually even have an intuition on
that myself. So, that type of exercise would be useful for that purpose."
END OF KEYNOTE Q&A
16
-------
Valuation of Ecological Benefits: Improving the Science
Behind Policy Decisions
PROCEEDINGS OF
SESSION III, PART 1: KEEPING WATER FRESH:
THE VALUE OF IMPROVED FRESH WATER QUALITY
A WORKSHOP SPONSORED BY THE U.S. ENVIRONMENTAL PROTECTION
AGENCY'S NATIONAL CENTER FOR ENVIRONMENTAL ECONOMICS (NCEE)
AND NATIONAL CENTER FOR ENVIRONMENTAL RESEARCH (NCER)
October 26-27, 2004
Wyndham Washington Hotel
Washington, DC
Prepared by Alpha-Gamma Technologies, Inc.
4700 Falls of Neuse Road, Suite 350, Raleigh, NC 27609
-------
ACKNOWLEDGEMENTS
This report has been prepared by Alpha-Gamma Technologies, Inc. with funding from
the National Center for Environmental Economics (NCEE). Alpha-Gamma wishes to
thank NCEE's Cynthia Morgan and the Project Officer, Ronald Wiley, for their guidance
and assistance throughout this project.
DISCLAIMER
These proceedings are being distributed in the interest of increasing public understanding
and knowledge of the issues discussed at the workshop and have been prepared
independently of the workshop. Although the proceedings have been funded in part by
the United States Environmental Protection Agency under Contract No. 68-W-01-055 to
Alpha-Gamma Technologies, Inc., the contents of this document may not necessarily
reflect the views of the Agency and no official endorsement should be inferred.
11
-------
TABLE OF CONTENTS
Session III, Part 1: Keeping Water Fresh: The Value of Improved Fresh Water
Quality
Recreation Demand Using Physical Measures of Water Quality
Kevin J. Egan, Joseph A. Herriges, Catherine L. Kling, and John A.
Downing, Iowa State University 1
Choice Margins and the Measurement of Ecological Benefits: The Case of
Urban Watersheds
V. Kerry Smith, Daniel J. Phaneuf, and Raymond B. Palmquist, North
Carolina State University 25
Discussant
John Powers, U.S. EPA, Office of Water 70
Summary of Q&A Discussion Following Session III, Part 1 73
in
-------
Recreation Demand Using Physical Measures of Water
Quality1
Kevin J. Egan, Joseph A. Herriges, and Catherine L. Kling
Department of Economics, Iowa State University
John A. Downing
Department of Ecology, Evolution and, Organismal Biology, Iowa State University
August 13, 2004
1 Funding for this project was provided by the Iowa Department of Natural Resources and the
U.S. Environmental Protection Agency's Science to Achieve Results (STAR) program. Although
the research described in the article has been funded in part by the U.S. Environmental Protection
Agency's STAR program through grant (R830818), it has not been subjected to any EPA review
and therefore does not necessarily reflect the views of the Agency, and no official endorsement
should be inferred.
1
-------
Abstract
This paper incorporates a rich set of physical water quality attributes, as well as site and
household characteristics, into a model of recreational lake usage in Iowa. Our analysis
shows individuals are responsive to physical water quality measures and WTP estimates are
reported based on improvements in these measures.
2
-------
1 Introduction
Over three decades have lapsed since the passage of the 1972 Clean Water Act (CWA), yet
progress towards meeting the standards set forth in the CWA has been slow in the area of
nonpoint source pollution. The most recent National Water Quality Inventory (USEPA,[17])
categorizes forty-five percent of assessed lake acres in the U.S. as impaired, with the lead-
ing causes of these impairments being nutrients and siltation. Moreover, few states have
developed the priority ranking of their impaired waters or determined the Total Maximum
Daily Loads (TMDLs) as required under Section 303(d) of the CWA.1 Legal actions by citi-
zen groups have prompted renewed efforts towards developing both the priority listing and
associated TMDL standards.2 However, the task facing both the EPA and state regulatory
agencies remains a daunting one. The prioritization process alone, which is all the more
important given current tight budgets, requires information on the cost of remediation and
the potential benefits that will flow from water quality improvements. Both types of infor-
mation are in short supply. The purpose of this paper is to help fill this gap by providing
information on the recreational value of water quality improvements as a function of detailed
physical attributes of the water bodies involved. The water quality values are obtained from
a recreation demand model of lake usage in the state of Iowa, combining trip and socio-
demographic data from the Iowa Lakes Valuation Project and an extensive list of physical
water quality measures collected by Iowa State University's Limnology Lab.
Recreation demand models have long been used to value water quality improvements,
but studies typically rely on limited measures of water quality. The most commonly used
indicators are fish catch rates (e.g., [3], [11]). However, catch rates are themselves endoge-
1TMDL's specify the amount of a pollutant that a water body can receive and still meet existing water
quality standards.
2 As of March 2003, there have been approximately 40 legal actions taken against the USEPA in 38 states
concerning the implementation of Section 303(d) of the CWA.
3
-------
nous, depending on both fishing pressure and the abilities of the anglers, and provide only
indirect measures of the underlying water quality. Physical water quality measures, such as
secchi depth and bacteria counts, are used only sparingly, in large part due to limitations in
available data. Phaneuf, Kling and Herriges [14] use fish toxin levels in their model of Great
Lakes fishing, but the toxin levels were available only for a limited number of aggregate sites
in the region. Parsons and Kealy [13] use dummy variables based on dissolved oxygen levels
and average secchi depth readings to capture the impact of water quality on Wisconsin lake
recreation. Similarly, Parsons, Helm, and Bondelid [12] construct dummy variables indicat-
ing High and Medium water quality levels for use in their analysis of recreational demand
in six northeastern states. These dummy variables are based on pollution loading data and
water quality models, rather than direct measurements of the local water quality. In all of
these studies, the physical water quality indicators are found to significantly impact recre-
ation demand, but, due to the limited nature of the measures themselves, provide only a
partial picture of value associated with possible water quality improvements.
Bockstael, Hanemann, and Strand's [2] analysis of beach usage in the Boston-Cape Cod
area has perhaps one of the most extensive lists of objective physical water quality attributes
included in a model of recreation: oil, fecal coliform, temperature, chemical oxygen demand
(COD), and turbidity. However, the study also points out one of the frequently encountered
problems in isolating the impact of individual water quality attributes - multicollinearity.
Seven additional water quality measures were available to the analysts: color, pH, alkalinity,
phosphorous, nitrogen, ammonia, and fecal coliform. These latter variables were excluded
from the analysis due to correlations among the various groups of water quality measures.
The five water quality variables used were chosen because they were either directly observ-
able by recreationists or highly publicized. While these choices are certainly reasonable given
limitations in the available data, the lack of direct information on how nutrient levels (phos-
4
-------
phorous and nitrogen) impact recreational usage is unfortunate in the context of setting
TMDL standards in midwestern states, where nutrient loadings are of particular concern.
The contribution of the current paper lies in our ability to incorporate a rich set of
physical water quality attributes, as well as site and household characteristics, into a model
of recreational lake usage in Iowa. Trip data for the study are drawn from the 2002 Iowa
Lakes Survey, the first of a four year project aimed at valuing recreational lake usage in Iowa.
The survey was sent to a random sample of 8,000 Iowa households, eliciting information on
their recreational visits to Iowa's 129 principle lakes, along with socio-demographic data
and attitudes towards water quality issues. The unique feature of the project, however, is
that a parallel inventory of the physical attributes of these lakes is being conducted by Iowa
State University's Limnology laboratory.3 Three times a year, over the course of a five year
project, eleven distinct water quality measurements are being taken at each of the lakes,
providing a clear physical characterization of the conditions in each lake. Moreover, due to
the wide range of lake conditions in the state, Iowa is particularly well suited to identifying
the impact of these physical characteristics on recreation demand. Iowa's lakes vary from a
few clean lakes with up to fifteen feet of visibility to other lakes having some of the highest
concentrations of nutrients in the world, and roughly half of the 129 lakes included in the
study are on the EPA's list of impaired lakes.
The remainder of the paper is divided into five sections. Section 2 provides an overview
of the two data sources. A repeated mixed logit model of recreational lake usage in Iowa is
then specified in Section 3. The mixed logit model allows for a wide variety of substitution
patterns among the recreational sites and for heterogeneity among households in terms of
their reaction to individual site characteristics. (See, e.g., [7],[10], and [16]). Parameter
estimates are reported in Section 4. In Section 5, we illustrate not only the implications of
3The limnological study is funded by the Iowa Department of Natural Resources.
5
-------
the model in terms of recreational value of meeting the objectives of the CWA (i.e., removing
all of the lakes in the state from the impaired water quality list), but also how the model
can be used to prioritize the remediation task. Conclusions from the paper are provided in
Section 6.
2 Data
Two principle data sources are used in developing our model of recreational lake usage in
Iowa: the 2002 Iowa Lakes Survey and the physical water quality measures collected by Iowa
State University's Limnology laboratory. As noted above, the 2002 Iowa Lakes Survey is
the first survey in a four year study of lake usage in the state. The focus of the survey was
on gathering baseline information on the visitation patterns to Iowa's 129 principle lakes,
as well as socio-demographic data and attitudes towards water quality issues. After initial
focus groups and pre-testing of the survey instrument, the final survey was administered
by mail in November 2002 to 8,000 randomly selected households in the state. Standard
Dillman procedures ([5]) were used to insure a high response rate.4 Of the 8,000 surveys
mailed, 4,423 were returned. Allowing for the 882 undeliverable surveys, this corresponds to
an overall response rate of sixty-two percent.
The survey sample was initially paired down to 3,859 households as follows. Those indi-
viduals who returned the survey from out of state were excluded (thirty-eight observations).
It is not feasible to ascertain whether these respondents have permanently left the state
or simply reside elsewhere for part of the year. Respondents who did not complete the trip
questions or did specify their numbers of trips (i.e. they simply checked that they had visited
a given lake) were excluded (224 observations). Lastly, anyone reporting more than fifty-two
total single day trips to the 129 lakes were excluded (133 observations). In the analysis
4 Complete details of the survey design and implementation can be found in [1],
6
-------
below, only single day trips are included to avoid the complexity of modeling multiple day
visits. Defining the number of choice occasions as fifty-two allows for one trip per week to
one of the 129 Iowa lakes. While the choice of fifty-two is arbitrary, it seems a reasonable
cut-off for the total number of allowable single day trips for the season.5 This last step elimi-
nated approximately three percent of the returned surveys. Finally, due to the large number
of respondents, the overall sample was randomly divided into three segments; specification,
estimation, and prediction portions. The analysis reported here comes from the specification
stage using 1,286 observations. Once the estimation stage is reached, the results will be free
from any form of pretest bias and the standard errors will be not be biased by the extensive
specification search.6
Table 1 provides summary statistics for trip and the socio-demographic data obtained
from the survey. The average number of total single day trips for all 129 lakes is 6.68 varying
from some respondents taking zero trips and others taking fifty-two trips. In general, the
survey respondents are more likely to be older, male, have a higher income, and to be more
educated than the general population. Schooling is entered as a dummy variable equaling
one if the individual has attended or completed some level of post high school education.
The physical water quality measures used in modelling recreational lake usage in Iowa
were gathered by Iowa State University's Limnology laboratory. Table 2 provides a listing
of the water quality attributes and 2002 summary statistics for the 129 lakes used in our
analysis. All of the physical water quality measures are the average values for the 2002
season. Samples were taken from each lake three times throughout the year, in Spring/early
Summer, mid-Summer, and late Summer/Fall to include seasonal variation.
Each of the water quality measures help to characterize a distinct aspect of the lake
5 Sensistivity analysis, raising the allowable number of trips per year above fifty-two, indicated that the
results were not sensitive to the choice of this cut-off.
6 Creel and Loomis [4] use a similar procedure in investigating alternative truncated count data estimators.
7
-------
ecosystem. Secchi depth indicates the lake depth at which the bottom of the lake can still be
seen, providing an overall water clarity measure. Chlorophyll is an indicator of plant biomass
or algae, which in turn leads to greenness in the water. Three nitrogen levels are gathered.
In addition to total nitrogen, NH3+NH4 measures particular types of nitrogen, such as
ammonia, that can be toxic, whereas N03+N02 measures the nitrate level in the water.
Total phosphorous is an important indicator of water quality in Iowa, as it is usually the
principal limiting nutrient which determines algae growth. Silicon is important to diatoms,
a key food source for marine organisms. The acidity of the water is measured by "pH" with
levels below 6 or above 8 indicating unhealthy lakes. As Table 2 notes, all of the pH levels
in this sample are tightly clustered between 7.3 and 10. Alkalinity is the concentration of
calcium or calcium carbonate in the water. Plants need carbon to grow and all carbon comes
from alkalinity, therefore alkalinity is an indication of the abundance of plant life. Inorganic
suspended solids (ISS) consist basically of soil and silt in the water due to erosion, where as
volatile suspended solids (VSS) consists of organic matter. Increases in either ISS or VSS
levels will decrease water clarity. With the exception of pH levels, Table 2 demonstrates that
there is considerable variation in water quality conditions throughout the state. For example,
secchi depth varies from a low of 0.09 meters (or 3.5 inches) to a high of 5.67 meters (over
18 feet). Total phosphorus varies from 17 to 453 ug/L, some of the highest concentrations
in the world.
In addition to trip and water quality data, two other data sources were used. First, the
travel costs, from each survey respondent's residence to each of the 129 lakes, were needed.
The out-of-pocket component of travel cost was computed as the roundtrip travel distance
multiplied by $0.25 per mile.7 The opportunity cost of time was calculated as one-third the
estimated roundtrip travel time multiplied by the respondent's wage rate. Table 3 provides
7PCMiler (Streets Version 11) was used to compute both roundtrip travel distance and time.
8
-------
summary statistics for the resulting travel cost variable. The average price of a recreational
trip to a lake is $136, although perhaps a more meaningful statistic is the average price of a
lake visit, $85.
Second, lake site characteristics were obtained from the Iowa Department of Natural
Resources [9]. Table 3 provides a summary of these site characteristics. As Table 3 indicates,
the size of the lakes varies considerably, from 10 acres to 19,000 acres. Four dummy variables
are included to capture different amenities at each lake. The first is a "ramp" dummy variable
which equals one if the lake has a cement boat ramp, as opposed to a gravel ramp or no
boat ramp at all. The second is a "wake" dummy variable which equals one if wakes are
allowed and zero otherwise. About 66% of the lakes allow wakes, whereas 34% of lakes are
"no wake" lakes. The "state park" dummy variable equals one if the lake is located in a
state park, which is the case for 38.8% of the lakes in our study. The last dummy variable
is the "facilities" dummy variable. Facilities include things like restrooms, picnic tables, or
vending machines. A concern may be that facilities would be strongly correlated with the
state park dummy variable. However, while fifty of the lakes in the study are located in state
parks and fifty have accessible facilities, only twenty six of these overlap.
3 The Model
The Mixed Logit model was chosen since it exhibits many desirable properties including, "it
allows for corner solutions, integrates the site selection and participation decisions in a utility
consistent framework, and controls for the count nature of recreation demand (Herriges and
Phaneuf, [7])."
Assume the utility of individual i choosing site j on choice occasion t is of the form
Uijt = V (Xif, fii) + eijt, i = 1,N; j = 0,..,J; t = 1 (1)
where V represents the observable portion of utility, and from the perspective of the
9
-------
researcher, £ijt} represents the unobservable portion of utility. A mixed logit model is defined
as the integration of the logit formula over the distribution of unobserved random parameters
(Revelt and Train, 1998). If the random parameters, f3i7 were known then the probability of
observing individual i choosing alternative j on choice occasion t would follow the standard
logit form
L*m= ;xp(t"'(ft)> . (2)
exp [Vikt (&)]
k=0
Since the (3^s are unknown, the corresponding unconditional probability, P^ (61), is ob-
tained by integrating over an assumed probability density function for the /Vs. The uncon-
ditional probability is now a function of 0, where 0 represents the estimated moments of the
random parameters. This repeated Mixed Logit model assumes the random parameters are
i.i.d. distributed over the individuals so that
Pjt = J Lijt(f3)f(f3\9)df3. (3)
No closed form solution exists for this unconditional probability and therefore simulation
is required for the maximum likelihood estimates of 6.%
Following Herriges and Phaneuf [7], a dummy variable, Dj, is included which equals one
for all of the one through J recreation alternatives and equals zero for the stay-at-home
option (j = 0). Including the stay-at-home option allows a complete set of choices, including
in the population those individuals who always "stay at home" on every choice occasion and
do not visit any of the sites. It is convenient to partition the individual's utility into the
stay-at-home option or choosing one of the J sites
TT /3 Zi ~\~ £iOt f A\
Uijt ~ nl , , -17' V4/
PiXij + ai + £ijt, J ~ 1, •••; J
8Randomly shifted and shuffled uniform draws are used in the simulation process (Hess, Train, and Polak,
[8]). The number of draws used in the simulation is 750.
10
-------
where ati is the random parameter on the dummy variable, Dj, which does not appear since
it equals one for j = 1,J and zero for j = 0. The vector z% contains socio-demographic
data such as income and age, and Xij represents the site characteristics that vary across the
lakes, including attributes such as facilities at the lake as well as water quality measures.
Notice the parameters associated with the socio-demographic data are not random as this
information does not vary across the sites.9
The random coefficient vectors for each individual, f3i and can be expressed as the
sum of population means, b and a, and individual deviation from the means, 8^ and 7^, which
represents the individual's tastes relative to the average tastes in the population (Train, [16]).
Therefore redefine
PiXij = b' Xij + S'iXij (5)
a,i = a + (6)
and then the partitioned utility is
TT ft Zi -\- Tji
uijt — 0> 11 'I T ' v' /
PiXij + a + 7)ijt, j — 1,..., J
where
£iot i 1 j • • • j -/V; t 1, ...,T
ijt~ Kxij + li + £ijt, j = i = l,-,N; t = l,...,T
is the unobserved portion of utility. This unobserved portion is correlated over sites and trips
due to the common influence of the terms 5[ and 7^ which vary over individuals. For example,
an individual who chooses the stay-at-home option for all choice occasions would have a
negative deviation from a, the mean of ctj, while someone who takes many trips would have
a positive deviation from a, allowing the marginal effect to vary across individuals. However,
the parameters do not vary over sites or choice occasions; thus, the same preferences are used
9 It is possible to interact the socio-demographic data with the sites, if one believed for example that
income would effect which lake was chosen.
11
-------
by the individual to evaluate each site at each time period. Since the unobserved portion
of utility is correlated over sites and trips, the familiar IIA assumption does not apply for
mixed logit models.
In particular, we model the utility individual i receives from choosing lake j on choice
occasion t as
TT P ^iOt /Q\
J —f3pPij + (3q Qj + (3® Aj + a>i + eijt, j = 1,J
where is the socio-demographic data summarized in Table 1, is the travel cost from each
Iowan's residency to each of the 129 lakes, as calculated with PCMiler (Table 3). The vector
Qj denotes the physical water quality measures (Table 2) and Aj represents the attributes
of the lake (Table 3). As shown in equation (9), notice that the parameters on the lake
attributes and the dummy variable, Dj, are random. These six variables are assumed to be
independently normally distributed with the mean and dispersion of each variable estimated.
Finally, we estimate two models. The first specification, model A, includes six physical
water quality measures. Included are the four paramount variables for nutrient criteria
(USEPA [17]): total phosphorus, total nitrogen, chlorophyll, and Secchi depth, as well as
inorganic suspended solids and organic suspended solids, which we consider to be crucial
indicators as well. A second model, model B, includes the complete list of eleven water
quality measures. Estimating two models allows us to observe the stability of the parameters
across different specifications.
4 Results
The results for Model A and B are divided into two Tables, 4a and 4b. For both models,
the coefficients for the socio-demographic data, price, and the random coefficients on the
amenities are given in Table 4a. Table 4b lists for both models the coefficients for the
physical water quality measures. All of the coefficients are significant at the 1% level except
12
-------
for a few of the socio-demographic data. For model B, with eleven physical water quality
measures, only the "male" dummy variable is not significant. In Model A, income, household
size, and the quadratic term on age are insignificant. Note that the socio-demographic data
are included in the conditional indirect utility for the stay-at-home option. Therefore, the
negative income coefficient indicates that as income rises the respondents are less likely to
stay at home and more likely to visit a lake (i.e. lake visits are a normal good). Males,
higher educated individuals, and larger households are all more likely to take a trip to a
lake. Age has a convex relationship with the stay-at-home option and therefore a concave
relationship with trips. For Model B, the peak occurs at about age 37, which is consistent
with the estimate of larger households taking more trips, as at this age the household is more
likely to include children.
The price coefficient is negative as expected and identical in both models. Now turning
to the amenities parameters, again all of the parameters are of the expected sign. As the
size of a lake increases, has a cement boat ramp, gains accessible facilities, or is in a state
park, on average leads to increased trips. Notice however the large dispersion estimates. For
example, in model A the dispersion on the size of the lake indicates 11.1% of the population
prefers a smaller lake, possibly someone who enjoys a more private experience. The large
dispersion on the "wake" dummy variable seems particularly appropriate given the poten-
tially conflicting interests of anglers and recreational boaters. Anglers would possibly prefer
"no wake" lakes and recreational boaters would obviously prefer lakes that allow wakes. It
seems the population is almost evenly split with 56.9% preferring a lake that allows wakes
and 43.1% preferring a "no wake" lake. Lastly, the mean of the trip dummy variable,
is negative indicating that on average the respondents receive higher utility from the stay-
at-home option, which is expected considering the average number of trips is 6.7 out of a
possible 52 choice occasions.
13
-------
The physical water quality coefficients are reported in Table 4b and are relatively stable
across the two models. For both models A and B, secchi depth is positive and the suspended
solids, both organic and inorganic (volatile), are negative, indicating the respondents strongly
value water clarity. However, the coefficient on chlorophyll is positive suggesting on average
respondents do not mind some variation of green water. The negative coefficient on total
phosphorus, the most likely principal limiting nutrient, indicates higher algae growth leads
to fewer recreational trips.
The only physical water quality coefficient to change qualitatively across the two spec-
ifications is total nitrogen which is positive in model A. Total nitrogen having a positive
coefficient is consistent with expectations given the negative sign on total phosphorus. With
such large amounts of phosphorus in the water, more nitrogen can actually be beneficial by
allowing a more normal phosphorus to nitrogen ratio. If the ratio becomes too imbalanced
more problematic blue-green algae blooms become dominant. Total nitrogen is negative in
model B, but two other forms of nitrogen are included with the nitrates form (N03+N02)
being positive, possibly for the same reason as just discussed.
Continuing with the additional measures in model B, alkalinity has a positive coefficient,
consistent with alkalinity's ability to act as a buffering capacity on how much acidity the
water can withstand before deteriorating. Since all of the lakes in the sample are acidic (i.e.
pH greater than 7) a positive coefficient for alkalinity is expected. The positive coefficient
on Silicon is also consistent since Silicon is important for diatoms, which in turn are an
important food source for marine organisms. Lastly, pH is entered quadratically reflecting
the fact that low or high pH levels are signs of poor water quality. However, as mentioned,
in our sample of lakes all of the pH values are normal or high. The coefficients for pH show
a convex relationship (the minimum is reached at a pH of 8.2) to trips, indicating that as
the pH level rises above 8.2, trips are predicted to increase. This is opposite of what we
14
-------
expected and further specifications will consider this fact.
5 Welfare Calculations
Given the random parameters, /3i5 the conditional compensating variation associated with a
change in water quality from Q to Q' for individual i on choice occasion t is
CVit = In
^exp(Vyt [Q'-,Pi])
.3=o
— In
J^exp(yiit [Q; /5J)
.3 =o
which is the compensating variation for the standard logit model. The unconditional
compensating variation does not have a closed form, but it can be simulated by
R
cv" = s E ^ ^
r=1 '
J^exP(Viit [Q';/3-])
J=o
— In
J^exp(Viit [Q;/3-])
.3=0
where R is the number of draws and r represents a particular draw from its distribution.
The simulation process involves drawing values of f3i and then calculating the resulting
compensating variation for each vector of draws, and finally averaging over the results for
many draws. Following Von Haefen [18], 2,500 draws were used in the simulation.
Three water quality improvement scenarios are considered with the results from
Model A used for all the scenarios. The first scenario improves all 129 lakes to the physical
water quality of West Okoboji Lake, the cleanest lake in the state. Table 5 compares the
physical water quality of West Okoboji Lake with the average of the other 128 lakes. All of
West Okoboji Lake's measures are considerably improved over the other 128. For example,
West Okoboji Lake has slightly over 5 times the water clarity, measured by secchi depth,
of the other lakes. Given such a large change, the annual compensating variation estimates
of $208.68 for every Iowa household seems reasonable (Table 7). Aggregating to the annual
value for all Iowans simply involves multiplying by the number of households in Iowa which
is 1,153,205.10 Table 7 also reports the average predicted trips before and after the water
10Number of Iowa Households as reported by Survey Sampling, Inc., 2003.
15
-------
quality improvement. Improving all 128 lakes to the physical water quality of West Okoboji
Lake leads to a reasonable 14.1% increase in average trips. As expected, the predicted trips
to West Okoboji Lake fall by 19.8% from 0.39 average trips per Iowa household to 0.31.
Iowans can now choose the nearest lake with the attributes they prefer, instead of traveling
further to West Okoboji Lake.
The next scenario is a less ambitious, more realistic plan of improving nine lakes to the
water quality of West Okoboji Lake (see table 5 for comparison). The state is divided into
nine zones with one lake in each zone, allowing every Iowan to be within a couple of hours of
a lake with superior water quality. The nine lakes were chosen based on recommendations by
the Iowa Department of Natural Resources for possible candidates of a clean-up project. The
annual compensating variation estimate is $39.71 for each Iowa household. As expected, this
estimate is 19.0% of the value if all lakes were improved, even though the scenario involves
improving only 7.0% of the lakes. This suggests location of the improved lakes is important
and to maximize Iowan's benefit from improving a few lakes, policymakers should consider
dispersing them throughout the state.
The last scenario is also a policy oriented improvement. Currently of the 129 lakes, 65
are officially listed on the EPA's impaired waters list. TMDL's are being developed for these
lakes and by 2009 the plans must be in place to improve the water quality at these lakes
enough to remove them from the list. Therefore, in this scenario the 65 impaired lakes are
improved to the median physical water quality levels of the 64 non-impaired lakes. Table 6
compares the median values for the non-impaired lakes to the averages of the impaired lakes.
The table indicates the median values of the non-impaired lakes seems an appropriate choice
with physical water quality measures higher than the averages of the 65 impaired lakes, but
much below those of West Okoboji Lake. This scenario is valued considerably lower than
the first two water quality improvement scenarios. The estimated compensating variation
16
-------
per Iowa household is $4.87. Consistent with this, the predicted trips only increase 0.3%
over the predicted trips with no improvement in water quality. A reasonable conclusion
is Iowan's have an abundance of lakes at this threshold level, and bringing the low quality
lakes up to this level is not much of a benefit.
6 Conclusions
The first year survey of the Iowa Lakes Project gathered recreation behavior to 129 of Iowa's
principal lakes. This data was combined with extensive physical water quality measures
from the same set of lakes gathered by the Iowa State University Limnology Lab. Our
analysis employing the repeated mixed logit framework, shows individuals are responsive to
physical water quality measures and it is possible to base willingness to pay calculations on
improvements in these physical measures. In particular we considered three improvement
scenarios, with the results suggesting Iowans more highly value a few lakes with superior
water quality rather than all recreational lakes at an adequate level, as determined by being
listed as an impaired lake by the Environmental Protection Agency.
A number of important practical findings come directly from this work. Limnologists and
other water quality researchers should be interested in the results of this paper, since the
general belief is that visitors care about water clarity as measured by secchi depth (how many
meters beneath the surface of the water a secchi dish is visible) or water quality in general.
By estimating the partial effects of a list of physical measures, we have determined which
significantly affect recreationist's behavior. Limnologists and water resource managers can
then use this information about what physical lake attributes visitor's trip behavior responds
to in designing projects for water quality improvements. Our results indicate water clarity
is very important as evidenced by the secchi dish and suspended solids parameters. Also,
nutrients in general are found to decrease recreation trips.
17
-------
The findings from this study also have direct relevance for environmental protection
managers and citizens concerned with the water quality in that they can be used to prioritize
clean-up activities to generate the greatest recreation benefits for a given expenditure. Not
only can the findings be used to determine which lakes and in what order to clean them, but
also the most efficient levels of improvement.
18
-------
Table 1. 2002 Iowa Lakes Survey Summary Statistics
Variable
Mean
Std. Dev.
Min.
Max.
Total Day Trips
6.68
10.46
0
52
Income
$56,140
$37,436
$7,500
$200,000
Male
0.67
0.46
0
1
Age
53.36
16.47
15
82
School
0.66
0.47
0
1
Household Size
2.61
1.32
1
12
Table 2. Water Quality Variables and 2002 Summary Statistics
Variable
Mean
Std. Dev.
Min.
Max.
Secchi Depth (m)
1.17
0.92
0.09
5.67
Chlorophyll (ug/1)
41
38
2
183
NH3+NH4 (ug/1)
292
159
72
955
N03+N02 (mg/1)
1.20
2.54
0.07
14.13
Total Nitrogen (mg/1)
2.20
2.52
0.55
13.37
Total Phosphorous (ug/1)
106
81
17
453
Silicon (mg/1)
4.56
3.24
0.95
16.31
PH
8.50
0.33
7.76
10.03
Alkalinity (mg/1)
142
41
74
286
Inorganic SS (mg/1)
9.4
17.9
0.6
177.6
Volatile SS (mg/1)
9.4
7.9
1.6
49.9
Table 3. Summary Statistics for Lake Site Characteristics
Variable
Mean
Std. Dev.
Min.
Max.
Travel Cost
135.79
29.47
94.12
239.30
Acres
672
2,120
10
19,000
Ramp
0.86
0.35
0
1
Wake
0.66
0.47
0
1
State Park
0.39
0.49
0
1
Facilities
0.39
0.49
0
1
19
-------
Table 4a. Repeated Mixed Logit Model Parameter Estimates (Std. Errs in Parentheses)"
Model A: 6 Water Quality Measures Model B: 11 Water Quality Measures
Variable
Mean
Dispersion
Mean
Dispersion
Income
—0.008*
(0.007)
-0.12*
(0.007)
Male
—4.98*
-0.31
(0.42)
(0.42)
Age
-0.24*
-0.58*
(0.07)
(0.08)
Age2
0.0001
(0.00006)
0.0078*
(0.0007)
School
-4.45*
(0.40)
-3.44*
(0.40)
Household
-0.41
-1.24*
(0.17)
(0.17)
Price
-0.17*
-0.17*
(0.0006)
(0.0007)
Log(Acres)
4.60*
(0.064)
3.81*
(0.057)
5.13*
(0.067)
4.05*
(0.06)
Ramp
11.60*
17.85*
14.87*
18.79*
(0.78)
(0.51)
(0.89)
(0.59)
Facilities
1.18*
18.09*
3.54*
16.78*
(0.26)
(0.28)
(0.24)
(0.25)
State Park
8.00*
15.15*
6.67*
13.99*
(0.26)
(0.27)
(0.24)
(0.27)
Wake
2.76*
15.81*
-1.64*
15.57*
(0.30)
(0.33)
(0.30)
(0.29)
a
-8.97*
3.01*
-9.19*
3.12*
(0.05)
(0.04)
(0.05)
(0.04)
* Significant at 1% level.
a All of the parameters are scaled by 10, except a (which is unsealed) and the income
coefficient (which is scaled by 10,000).
20
-------
Table 4b. Repeated Mixed Logit Model Parameter Estimates (Std. Errs in Parentheses)"
Model A: 6 Water
Model B: 11 Water
Variable
Quality Measures
Quality Measures
Secchi Depth (m)
0.78*
(0.05)
0.84*
(0.07)
Chlorophyll (ug/1)
0.054*
(0.03)
0.06*
(0.003)
NH3+NH4 (ug/1)
-0.002*
(0.0006)
N03+N02 (mg/1)
3.16*
(0.19)
Total Nitrogen (mg/1)
0.31*
(0.01)
-3.21*
(0.19)
Total Phosphorous (ug/1)
-0.0033*
(0.001)
-0.016*
(0.001)
Silicon (mg/1)
0.81*
(0.02)
pH
-136.72*
(5.83)
PH2
8.35*
(0.34)
Alkalinity (mg/1)
0.038*
(0.002)
Inorganic SS (mg/1)
-0.010*
(0.008)
-0.089*
(0.009)
Volatile SS (mg/1)
-0.18*
(0.01)
-0.28*
(0.02)
LogLik
-47,740.38
-47,494.17
* Significant at 1% level.
a All of the parameters are scaled by 10.
21
-------
Table 5. West Okoboji Lake vs. the other 128 Lakes
West Okoboji
Averages of the
Averages of the
Lake
other 128 Lakes
9 Zone Lakes
Secchi Depth (m)
5.67
1.13
1.23
Chlorophyll (ug/1)
2.63
41.29
40.13
Total Nitrogen (mg/1)
0.86
2.22
3.64
Total Phosphorous (ug/1)
21.28
106.03
91.11
Inorganic SS (mg/1)
1.00
9.49
9.52
Volatile SS (mg/1)
1.79
9.43
8.42
Table 6. 64 Non-impaired Lakes vs. the 65 Impaired Lakes
Median of the Averages of the
64 Non-impaired Lakes 65 Impaired Lakes
Secchi Depth (m)
1.27
0.70
Chlorophyll (ug/1)
23.25
56.76
Total Nitrogen (mg/1)
1.11
2.77
Total Phosphorous (ug/1)
58.79
153.70
Inorganic SS (mg/1)
3.51
20.42
Volatile SS (mg/1)
6.02
15.49
Table 7. Annual Compensating Variation Estimates using Model A
Average CV
per choice occasion
per Iowa household
for all Iowa
households
Predicted Trips
(9.80 with current
water quality)
All 128 Lakes
Improved to W. Okb.
$4.01
$208.68
$240,649,000
11.18
9 Zone Lakes
Improved to W. Okb.
$0.76
$39.71
$45,788,092
10.06
65 Impaired Lakes
Improved to Median
$0.09
$4.87
$5,612,219
9.83
22
-------
References
[1] Azevedo, C.D., K.J. Egan, J.A. Herriges, and C.L. Kling (2003). Iowa Lakes Valuation
Project: Summary and Findings from Year One. Final Report to the Iowa Department
of Natural Resources, August.
[2] Bockstael, N.E., W.M. Hanemann, and I.E. Strand (1986). Measuring the Benefits of
Water Quality Improvements Using Recreationd Demand Models, report presented to
the Environmental Protection Agency under cooperative agreement CR-811043-01-0,
Washington, D.C.
[3] Chen, H. Z., F. Lupi, and J. P. Hoehn (1999). "An Empirical Assessment of Multinomial
Probit and Logit Models for Recreation Demand," in Herriges, J.A., and C. L. Kling,
Valuing Recreation and the Environment: Revealed Preference Methods in Theory and
Practice, Cheltenham, UK: Edward Elgar, pp. 141-161.
[4] Creel, M.D., and J.B. Loomis (1990). "Theoretical and Empirical Advantages of Trun-
cated Count Data Estimators for Analysis of Deer Hunting in California." American
Journal of Agricultural Economics 72(2): 434-41.
[5] Dillman, D. A. (1978) Mail and Telephone Surveys - The Total Design Method, New
York: Wiley
[6] Hanemann, W.M. (1978) "A Methodoligical and Empirical Study of the Recreation Ben-
efits from Water Quality Improvements." Ph.D. dissertation, Department of Economics,
Harvard University.
[7] Herriges, J., and D. Phaneuf (2002). "Introducing Patterns of Correlation and Substi-
tution in Repeated Logit Models of Recreation Demand." American Journal of Agri-
cultural Economics 84: 1076-1090.
[8] Hess, S., K. Train, and J. Polak (2003). "On the use of randomly shifted and shuffled
uniform vectors in the estimation of a Mixed Logit model for vehicle choice." Working
paper.
[9] Iowa Department of Natural Resources (2004), Fishing Guide for Iowa Lakes.
[10] McFadden, D., and K. E. Train (2000). "Mixed MNL Models for Discrete Response."
Journal of Applied Econometrics 15(September/October): 447-70.
[11] Morey, E.R., R. D. Rowe, and M. Watson (1993). "A Repeated Nested-Logit Model
of Atlantic Salmon Fishing." American Journal of Agricultural Economics 75(August):
578-592.
[12] Parsons, G.R., E. C. Helm, and T. Bondelid (2003). "Measuring the Economic Benefits
of Water Quality Improvements to Recreational Users in Six Northeastern States: An
Application of the Random Utility Maximization Model." Working paper, University of
Delaware, July.
[13] Parsons, G.R., and M.J. Kealy (1992). "Randomly Drawn Opportunity Sets in a Ran-
dom Utility Model of Lake Recreation." Land Economics 68(1): 93-106.
[14] Phaneuf, D.J., C.L. Kling, and J.A. Herriges (2000). "Estimation and Welfare Cal-
culations in a Generalized Corner Solution Model with an Application to Recreation
Demand." The Review of Economics and Statistics 82(1) (February): 83-92.
23
-------
[15] Revelt, D., and K. Train (1998). "Mixed Logit with Repeated Choices: Households'
Choices of Appliance Efficiency Level." The Review of Economics and Statistics 80:
647-57.
[16] Train, K. (1998). "Recreation Demand Models with Taste Differences Over People."
Land Economics 74 (2) (May): 230-239.
[17] U.S. Environmental Protection Agency (2000). "Nutrient Criteria Technical Guidance
Manual: Lakes and Reservoirs." Office of Water, Office of Science and Technology,
Report EPA-822-B00-001, Washington, D.C.
[18] Von Haefen, R. (2003). "Incorporating observed choice into the construction of wel-
fare measures from random utility models." Journal of Environmental Economics and
Management 45: 145-65.
24
-------
Choice Margins and the Measurement of Ecological Benefits:
The Case of Urban Watersheds
V. Kerry Smith, Daniel J. Phaneuf, and Raymond B. Palmquist*
October 13, 2004
Draft please do not cite without permisssion
Abstract
This paper outlines a new method for measuring the marginal willingness to pay
for the services provided by ecological resources. The framework takes advantage of the
choices consumers make for observable private goods affected by one or more of these
services. Each of these decisions corresponds to a choice margin. The methodology uses
the distinction between long and short run choices to integrate a hedonic property value
model with recreation demand models differentiated by local housing neighborhoods.
The demand models are used to develop a consistent quantity index for the contribution
of the ecological services to the recreational activities that are expected to be possible in
different residential locations. The hedonic model estimates the marginal value for small
changes in this quality adjusted index for recreational opportunities. A new database on
recreational activities linked with housing sales is used to evaluate the services provided
by an urban watershed. The results support the proposed logic and indicate that marginal
benefits of protecting watersheds, measured under the two different perspectives, are
comparable, with the long run measure slightly larger than the benefits measured using ex
post recreation choices.
Key Words: ecological services, marginal willingness to pay, joint hedonic and random
utility models.
JEL Classification numbers: Q 26, Q 51, Q 57.
Thanks are due Jaren Pope for excellent research assistance with this project, as well as Brian Stynes,
Melissa Brandt, Michael Darden, Eric McMillan, and Vincent McKeever for assistance in the development
of the household survey. Thanks are also due Alex Boutaud and Jack Crawley for assuring the multiple
drafts of this paper were successfully completed. Partial support for this research was provided by the U. S.
Environmental Protection Agency through Star grant # R-82950801.
25
-------
I. Introduction
One of the most important challenges facing environmental policy analysts today
stems from the need to measure the gains or losses arising from changes in the services of
ecological resources.1 In this paper we consider one aspect of these challenges - the task
of measuring the welfare impacts of changes in water quality in a rapidly growing urban
watershed. In many areas of this country increasing demands for residential housing and
the subsequent development of supporting retail services have taxed the ability of
watersheds to provide basic ecological services. At the same time, much of the growth in
housing demand can be broadly viewed as amenity-driven. As a result, there is a
fundamental tradeoff to be faced between the largely private benefits from increased
development and the public costs of the amenity consequences stemming from
development-related land cover changes. Designing effective public policy to address
this tradeoff requires information of several types. Among these is an understanding of
the economic benefits from enhancing the amenities provided by urban watershed
services.
Three sources of benefit information are usually noted in providing responses to
these needs: (a) contingent valuation (or conjoint) studies with hypothetical plans to
improve (or avoid deterioration in) existing ecological resources such as wetlands (see
Johnston et al. [1999], Bateman et al. [2004] and Woodward and Wui [2001] as
examples); (b) hedonic property value models, using proxy measures, such as distance
1 There are many potential examples supporting this judgment. The U.S. Environmental Protection Agency
has recently established a federal advisory committee to consider the valuation of activities to protect
ecological systems and services. Similar efforts are currently underway under the auspices of the National
Academy of Sciences. Internationally the United Nations sponsored the Millennium Ecosystem
Assessment which sought to evaluate how changes in ecosystems services affect human well being and the
types of responses that can be adopted at local, national and global scales to improve ecosystem
management.
26
-------
between homes and the resource of interest to take account for the resource's effect on
property values, (Leggett and Bockstael [2000], Mahan, Polaksy and Adams [2000]);
and (c) travel cost recreation demand models estimating the value of amenity changes as
they relate to the value of recreation visits (see Phaneuf [2002] and Egan, Herriges,
Kling, and Downing [2004]). Applications using these approaches argue that they can
address the general problem of valuing changes in ecosystem services, but their results
remain somewhat disjointed. Each strategy exploits a different margin of choice and as a
result appears to arise from a different model. Little guidance has been offered to explain
the relationship between these methods' estimates of the values associated with enhanced
ecosystem services.2
This paper proposes a framework to address these limitations by combining
aspects of both the hedonic and recreation demand models. We develop a spatially
varying, theoretically consistent index that summarizes the effects of watershed quality
measures on the local recreation opportunities available because a household selects a
residential neighborhood. We use a hedonic model to estimate the effect of both this
index and the general amenity impacts on property values. Our strategy exploits the
distinctive margins of choice associated with the different benefit measurement models.
These choice margins relate to decisions with unique temporal dimensions. Short term,
local recreation choices are used to develop indexes of how watershed quality affects the
amount of recreation available. Asset values, the sales prices for private homes in our
case, are determined by long term residential choices. They also depend on amenities.
2 One of the reasons for this disparity stems from differences in the ways each method measures the
environmental services linked to choices of private goods. Efforts to use joint estimation linking two or
more methods have generally taken place where there is a well defined measure of these services (see
Cameron [1992a] as an example).
27
-------
To assure we can integrate the results from these two choices, we develop a framework
that distinguishes the influence of amenities on expected local recreation from amenities
as neighborhood attributes. Our choice margin approach exploits the logic of Pollak's
[1969] conditional demand framework in describing the use of long and short run
decisions to identify the value of water quality as it is reflected in both recreation
opportunities and general amenity effects.
We evaluate our proposal using a new database gathered for Wake County, North
Carolina. These data integrate property sales, recreation trips for a sample of over two
thousand homeowners, and measures of the characteristics of the sub-hydrologic units
comprising the county's watersheds. We estimate separate random utility models for
local recreation trips for each of nineteen housing areas. Each model takes account of the
travel time required for these short recreational outings and includes indexes of the
watershed quality in the hydrological areas containing each recreation site. Our
recreation quantity index is defined as the average of the conditional expected utility
associated with the opportunities available in each housing area. By measuring the value
of expected behavior arising from these short term trips (using the actual behavior of
current residents) we have a consistent index of the importance of watershed quality for
the outings available to each housing market area. Our framework provides a consistent
bound for the marginal willingness to pay. It is estimated using the influence of this
index on housing prices. We use a recent proposal to increase sewage capacity and
permit growth in Granville County, which adjoins our study area, to illustrate how the
framework can be applied. We treat the change as affecting the quality in a popular lake
28
-------
and assume for the calculations that it would lead to the loss of this recreational area for
local outings.
The remainder of the paper is organized as follows. In the next two sections we
discuss the conceptual basis for our approach and outline an operational model. This is
followed by sections describing the Wake County database, our empirical specifications,
and our estimates along with a policy application The final section provides discussion
of the general implications of our proposal.
II. Parsing Information from Different Choices
A. Conceptual Background
A choice margin describes an opportunity for an individual to make a decision
that leads to the acquisition of both a private good and the services of a non- market good.
These decisions can take place in both the long run and in day to day decisions,
sometimes characterized as the short run. Long run choices involve the selection of
neighborhoods and the purchase (or rental) of housing units. Short run decisions are
conditional on these longer term selections and can involve trips to local recreation sites
for short outings. Once a housing location choice is made, a household allocates
remaining monetary and time resources to other purchased goods, leisure, and recreation.
These decisions contribute to well-being in the short-run. As a result, it seems reasonable
to expect that when deciding on a residential location the household considers the
portfolio of amenities conveyed by the location, the accessibility of areas for recreation,
and how these (or other) amenities relate to the quality of recreation opportunities. These
29
-------
factors will contribute to the expected future gains from recreation trips originating from
the location
To model these decisions formally we artificially divide the decision process into
two steps, for each potential location and housing choice, a household is hypothesized to
evaluate the short run decisions that could be made for recreation outings to maximize
utility subject to its resource constraints at that location Preferences are assumed to be a
function of recreation trips, x(q), a numeraire good z, and housing services h(a,q) which
at this stage are treated as quasi-fixed. The term a designates a vector of housing
attributes including location specific characteristics. In equation (1) below we also
include the term e to include unobserved heterogeneity in households that is not known
by the analyst.
u = u(x(q\h(a,q\z,£) (1)
h(a,q) is treated as "given" from the perspective of the choices of x(q) and z. This
implies we can deduct its "cost" from income to derive the resources available for the
short run budget constraint. That is, the income available for the next stage of the process
- selecting x(q) and z - is constrained by the remaining disposal income (i.e.,
m = m* -ph(a,q) with m* the full income and ph(a,q) the hedonic price function in
annual terms).3 The budget constraint for these short run decisions is given in equation
(2).
m = m*-ph(a,q) = z + px-x(q) (2)
The first order conditions imply solutions for the recreation demands, other market
goods, with the indirect utility function V(px,m,q,e). This structure in turn implies that the
3 We assume one unit of housing is consumed so ph (a, q') is the annual expenditure on housing.
30
-------
realized ex post benefits for a given household from visits to recreation sites can be
defined by equation (3).
/ \ pCf V [',q,e]
MCS(q,£)=\-^4 Idp, (3)
i. I
where p°x denotes the choke price for visits to the recreation site.
Using recreation demand models to estimate the benefits from an improvement in
q requires describing how MCS(q, e) changes with q. This analysis expands the integral
given in (3) so that it considers changes in bothpx and q. We designated this expression
as MCs(q°,ql).
This measure corresponds to the area between the two Marshallian demand curves
at different values for q as defined in equation (4) below. The vertical line after the
expression defined by Roy's identity, (- Vjh jVm), designates that each Marshallian
demand is evaluated at a different value for q. This expression is the type of analysis we
discussed at the outset. It uses one type of choice margin, namely what is used to
describe the consumption of recreation trips under different amenity conditions, to
recover a measure of the value of the changes in amenity services from qo to qi. To
assure it is the full economic value for this change, conventional practice assumes x and q
are weak complements. In addition, a symmetry condition, parallel to that imposed with
multiple price changes, assures the consistency of line integrals. This logic follows from
Palmquist [2004] and is one interpretation of the Willig [1978] condition usually cited as
required for consistent measures of the consumer surplus arising from quality changes.
MCSA(q°,q>) = j'JU'f(- F„,/Vm\ V - j'jT(- Vp_/V,
qlj
\
dpx (4)
i y
31
-------
To consider the potential for linkages between models associated with long and
short run decisions, consider how the value of trips at a given quality level might
influence other decisions. Equation (3) provides a summary of the gains due to the
household's ability to take trips to the recreation site. It is the ex post benefit from the
access conditions giving rise to recreation trips at a given quality level. In making a
residential choice, it seems reasonable to suppose that households consider, ex ante, the
expectation of what these benefits would be for each possible neighborhood. In other
works, households consider the value of the recreation options implied by the choice of
each neighborhood. Each area, in principle, provides somewhat different access
conditions to recreation opportunities. As a result, we can argue that the expected
benefits available from a residential location can be seen as an attribute of the location.
With this simple model, we can define the expected benefits from recreation at a given
residential location by
EMCS(q) = E[MCS(q,e)] (5)
The expectation operator in this case is with respect to the heterogeneity across
households in the location, both observed and unobserved. The latter is identified in our
model by e. Equation (5) is not a household-specific measure. It is a measure of the
average recreation opportunities available because a household has the access defined by
one location compared to no access. Of course, in practice the choice is based on each
neighborhood's relative value, so the default of no access becomes irrelevant. The
expectation is across diverse households conditional on the level of q at each specific
location. Using a long run perspective, we hypothesize that this value would be
capitalized into housing prices in equilibrium.
32
-------
To consider the long ran version of our model we need to return to equation (1)
and assume for each location a household has evaluated the potential for different
patterns of local outings. This process implies that including EMCS(q) in the preference
function in equation (1) allows us to account for the sub-optimization that takes place
conditional on a location. More formally, equation (6) offers a simple description for this
objective function and (7) the relevant budget constraint
max u(h(a,q), EMCS(q\ z) (6)
m* = ph(a,q)+ px ¦ x(q,px,m(m *))+ z (7)
where x(q,px) represents the optimized value of x (given the allocation of income
relevant to choices of x with each housing location andph{.)). m(m*) acknowledges the
separability in the decision process that we imposed on this problem — recreation is
based on the housing decision so the income remaining m(m*) is the connection
associated with the budget decomposition in the choice process.
Ph(a,q) is the hedonic price function. With x and z assumed to be selected as part
of a separate decision process, optimal for each h(a,q), we can recast the problem in
equations (6) and (7) by holding EMCS(q) constant for a location Then the decision
process involves selecting the location that is best, given the recreation and z choices that
would be made for that location The first order conditions in this case reflect the re-
allocations in recreation use (captured through ) an(j the changes in housing as
dq
in equation (8).
33
-------
du dEMCS du 5h
dEMCS dq 5h 5q _ dph
du dq
dz
Households choose a residential location such that, at the margin, the value of expected
recreation plus aesthetic benefits of the location are balanced against the implicit
marginal purchase price of the amenities. Equation (8) implies that the elements of q can
influence home prices both directly through the amenity effect and indirectly through the
recreation effect.
Identifying these effects in hedonic models has proven challenging, and to our
knowledge no studies have isolated both the direct and indirect effects shown in equation
(8). Most hedonic studies rely on ad hoc proxy variables to control for the two effects.
In the next sub-section we suggest how information from a recreation survey can be
combined with housing sales information to operationalize our choice margin logic
B. Empirical Implementation
The approach outlined above, using the distinction between long run housing
choices and short run recreation decisions, requires that we augment housing sales price
information with recreation data. By matching the recreation usage of current residents
to specific homes we can characterize both the relevant choice alternatives and the
patterns of use. For example, suppose an urban watershed can be divided into J areas
corresponding to well-defined real estate markets. The spatial layout of the landscape
and existing amenity levels convey a similar portfolio of recreation opportunities (and
qualities) to each resident in each of these market areas. Denote the set of recreation sites
in each housing market area j by Kj and the amenities levels at these sites by qj. Given
34
-------
observations on visits to sites by residents of zone j it is possible to estimate a random
utility model of site choice. Denote the indirect utility for a visit by person i to site k by
uik = a + Ptik + 8qk + eik, k=l i = \,...,N] (9)
where is a measure for the time cost for visiting site k, (a,[3,8) denote parameters to be
estimated, e& can be assumed to be a random error term distributed type I extreme value,
and TV; is the number of person-trips observed in market area j. Given the error
distribution, estimation of the parameters in (9) for observed trips originating from
market area j is straightforward, and the expected maximum utility from the opportunities
available for a trip originating in market area j is given in equation (10).
Eu]{q)=}f-Yj\a
i =1
Zexp(^)
k=1
(10)
where vk is the predicted deterministic component of utility (i.e., j5tik + &/,., a cannot
be identified, but does not affect the properties of the index ).
The expected value of anticipated recreation derived from a location, as defined in
equation (10), can also be interpreted as a "quantity index" for the set of recreation
alternatives available in a given neighborhood. It is consistent with the outline we
developed above. In this case, a linear expression for the outing choice model (i.e.
equation (9)) allowed us to avoid considering the allocation of income to local outings.4
We have also avoided assumptions about "pricing" the time (4t) required for these trips,
because our objective is to measure an index for the recreational choice opportunities
conveyed by each location While we can use the model underlying (9) to measure the
4 This assumption of locally constant marginal utility of income implies that the Marshallian measure of the
value of trips and the Hicksian measure will be equal. Because our focus is on choices among locations for
trips without a stay-at-home option, we do not consider selection effects arising from our survey response
rate. This is an area for future research.
35
-------
economic value of changes in the attributes of that choice set, our primary objective is to
develop the quality adjusted quantity index5 A random utility framework assumes each
trip decision is independent of every other such choice. Thus, under these conditions
(and in the absence of income effects) equations (6) and (10) provide comparable
measures for the effects of recreation for housing choices. The random utility approach
has the added advantage of easily reflecting a wide array of site alternatives.
Thus, our index collapses a large amount of spatially explicit information on site
availability and quality into a single variable that varies across the urban landscape. With
a measure of the recreation alternatives available to homebuyers when they select each
location, it is possible to isolate the effects of changes in ecological services as they affect
local recreation. This strategy also does not preclude considering how amenities
contribute to neighborhood attributes. These two terms are the elements isolated in
equation (8). For our hedonic model, we adopt a semi-log specification for the price
functions as in equation (ll).6 q(d,) is the distance proxy used to describe the
neighborhood amenity effect of a resource that is described using a measure of the
distance between the house and that resource.
s
In pu=a0+Y,P,a u +Y\Euji + y2q{di) + ui (11)
i=i
A measure bounding the welfare effects associated with changes in the ecological
services water based recreation sites in urban watersheds can then be distinguished based
5 This strategy does not require that we measure the opportunity cost of time. It can be assumed to be a
source of unobserved heterogeneity. In separate research with this same sample we found that the
opportunity cost of time can vary with the amount of time required for these types of outings (see
Palmquist, Phaneuf, and Smith [2004]).
6 Cropper, Deck, and McConnell's [1988] simulation experiments suggest that when the independent
variables in hedonic models are replaced with proxy variables or the specifications are likely to be
incomplete, simpler specifications for the price function such as the semi log have superior properties based
on estimates of the marginal willingness to pay.
36
-------
on how the change in q influences Euj. With the semi log form these would be given in
equation (12).7
Our analysis requires that three different types of information be combined
consistently. The first involves information on the sales of private homes in Wake
County, North Carolina. These data were obtained from the Wake County Revenue
Department. This database includes detailed information on residential properties.
However, the format for these data was often not compatible with economic analysis. A
translation from administrative records to measures of structural features of the homes
was an important first step in our research. For example, the county database has
information on each home's floor plan. The pre-analysis of these records required
calculating the number of squared feet in different uses in each home. The top panel of
Table 1 provides definitions of variables derived from these sources. Most are self
explanatory. A set of qualitative variables for the condition of the house were defined
7 We could also consider how we would measure a bound for changes in site specific amenities. This
process requires the definition of a distance equivalent change for the change in q.
(12)
A8rec = estimated bound for annual benefits from change from q°
to q1 due to the location specific recreation effects for
market area j and property i
? = annualization factor
III. Data
37
-------
based on ratings of the physical condition of the structure, rating from A (the highest
score) to D (the lowest score).8
The second set of information was derived from a mailed survey to homeowners.
Using the records of home sales from 1992 to 2000, we selected owner-occupied
properties with sales prices greater than $50,000. Our sampling plan took advantage of
realtor defined sub-markets. There are nineteen zones identified by the Triangle Multiple
Listing Service as relatively homogenous sub-markets.9 These areas will be labeled as
the MLS zones. Figure 1 displays a map of the county with the spatial boundaries for
each zone. For the selection of our sample, these areas were combined into four larger
contiguous zones (i.e., approximately dividing the county into four quadrants). 9,000
records were drawn randomly from the records satisfying our initial criteria. The
resulting sampled units were then evaluated to assure a sufficient number of observations
in each of the sub-hydrologic units identified in a detailed separate analysis of the
watersheds in Wake County by a private consulting firm (see CH2MHill [2003]). The
sub-areas defined for this assessment are given in Figure 2.10 There are 81 sub-
hydrologic units in the CH2MHill classification scheme. When the initial sample did not
have 20 observations in a sub-hydrologic area, we evaluated the set of housing sales that
remained after drawing our initial sample of 9,000. If there were sufficient remaining
housing sales in the relevant zones, we randomly selected additional observations to raise
the number in each area to 20. If there were an insufficient number of sales, we simply
8 A very small number of sales were of houses rated E. These were combined with rating D before the
sample was drawn.
9 We analyzed the hedonic price function in two ways - as a single price equation for one equilibrium based
on structural attributes, but with our index of area specific opportunities as a determinant of price, and
alternatively as a price function with fixed effects for each sub-market. Then we evaluated the
determinants of these fixed effects (see below).
10 Because the hydrologic zones are generally smaller than the MLS zones, this restriction also assured
reasonable sample sizes for each MLS zone. See table 1A in the Appendix.
38
-------
selected all that met our criteria. Each owner's name and address was verified using the
current Wake County Property tax records. Only observations where the sales record
from our hedonic database could be cross-linked to the currently listed owners were
included in the sample.
A mailed survey was designed to collect information about each homeowner's
socio-economic characteristics, recreation behavior, and leisure time choices. One aspect
of the design of the questionnaire involved collecting information about whether
homeowners considered water-based recreation sites and their attributes in making their
housing choices. To address this issue we conducted two focus groups.11 These
discussions lead to the definition of a new class of recreation trips - which we designate
here as local outings. These trips are short excursions involving a few hours.
Surveys were mailed to 7,554 households with valid addresses where we also had
complete sales and property characteristics. Two mailings and a reminder postcard were
sent to each selected homeowner (i.e., following the Dillman [1978] format for mailed
surveys). Our survey took place between May 2003 and September 2003. We realized a
32% response rate, based on completed valid responses in comparison to the mailing
reaching the intended addresses.12 Each survey packet contained the survey
11 Two focus groups were conducted as part of the background research to develop the survey. The first
took place July 23, 2002 with 10 homeowners. The years they lived in Raleigh ranged from 5 to 36 years.
The second was October 9, 2002 with 14 individuals. Members of this group have lived in the area
between 2 and 26 years. The focus groups identified local outings as the primary type of recreation that
would be influential for selecting among neighborhoods in Wake County. Participants did not feel location
would be important to trips that involved a longer time period.
12 To gauge the potential for selection effects we used information from the 2000 Census at the block group
level to estimate a grouped logit model with the fraction returning a questionnaire specified to be a function
of the socio-economic characteristics of each block group. The results suggest areas with higher proportion
of white residents, in areas with older homes, and with residents that lived a longer time in the area were
more likely to return the questionnaire. There was some evidence that the response rates might be lower
from high income areas. The full results (withz statistics in parentheses) are given as follows:
fraction responding = 1.724 percent white + 0.159 median number of rooms
(6.73) (1.94)
39
-------
questionnaire, a letter, a map, and a legend for the recreation sites (as well as the
opportunity to identify sites not listed on the map). Appendix A provides the survey
assignments and the proportion returned by MLS zone.
The survey design allows each of these two databases (i.e. the residential housing
sales data and the household survey) to be linked (via the latitude and longitude of each
residence) to a set of geo-coded records developed for each of the over 200 recreation
areas. These sites were identified in the survey. They are the sites listed by the survey
respondents as the places for their recreation trips. The records for the housing sales,
survey responses, and the locations, plus the travel time and distance to each recreation
site, can also be linked to a separately developed database that is the third component of
our analysis.
The last database includes records for water quality readings for the county. The
water quality data combine twelve separate databases with technical indicators of water
quality characteristics.13 Our analysis in this paper is intended to be a first stage
13 Two are chemical monitoring data obtained from the N.C. Department of Natural Resources. These
include monthly readings form 1994 to 2000 for 61 variables. The definitions of the factors that are
measured and the method used are documented based on available records. These reports were
supplemented with the paper records required for major NPDES point sources. Nine variables were
collected from the monthly reports of these sources for the Neuse River. Four types of biological databases
are included. Single samples collected on benthic and habitat, characteristics in August 2001 by CH2MHill
for Wake County, and periodic readings for the state benthic communities were collected by N.C. Division
ofWater Quality from 1982 to 2003forthe sites in the Neuse River Basin and from 1983 to 2001 for the
Cape Fear River Basin. Chemical data for four variables describing water quality for major lakes in the
Neuse and Cape Fear watersheds are available periodically from 1981 to 2002. Chemical data for the
upper Neuse River Basin with 89 variables are reported monthly over the period 1990 to 2002. The U.S.
Geological Survey (USGS) also report chemical and flow data for sites within the upper Neuse and Cape
Fear basins monthly from 1989 to 2001. This database includes 33 variables. Chemical and flow data for
+ 0.021 median house age
(3.83)
+ 0.022 x 10"5 median house value
(1.50)
+ 33.976
(1.28)
Pseudo R2 = 0.018
40
-------
evaluation of the basic logic of the model. As a result, we focus on only two of the
available variables - a measure of the percent of the land area in each MLS zone covered
with impervious surface, and a qualitative variable recoding the CH2MHill rating of the
sub-hydrologic units in the county.
IV. Results
A. Describing Local Recreation Choices: RUM Estimates
The first step in developing our index for the role of local recreation alternatives
available to homeowners involves modeling local outings. As we noted, a simple random
utility model is used to describe these choices. We develop separate models for each of
the 19 housing market areas identified by the Multiple Listing Service. This strategy is
possible because our survey elicits for each sampled homeowner a record of three types
of recreation trips to water based recreation sites. Our questionnaire asks about the
recreation trips taken during a seven month period from May through November 2002.
The trips are distinguished as: short outings that involve less than four hours away from
home; day outings involving a full day of activity but no overnight stay; and experiences
that involve two day trips with an overnight stay. The number of each type of trip, the
sites used, and activities undertaken are each recorded separately. Over two hundred
locations were identified by the sample respondents. Each site was geo-located with a
latitude and longitude. As described in the footnotes to Table 1, distance and travel time
measures to every possible site were estimated for every sample respondent.
the lower Neuse River Basin are available in the LNBA database monthly from 1994 to 2002. Finally, the
USGS flow data for upper Neuse River Basin was assembled monthly from 1990 to 2002. All these
databases can be linked either through the latitude and longitude of the sampling location or other
identifying information to our various geographic area definitions.
41
-------
For the random utility models estimated to develop our index of recreation
opportunities, we define a zone specific choice set with all the sites identified by
homeowners in each MLS zone. Each random utility model provides the basis for an
index of the local recreation alternatives homebuyers are assumed to consider in
evaluating the selection of a residential location. We assume new buyers will focus on
the recreation sites that current residents use. Each of the zone specific choice sets varies
in size and composition. While our questionnaire did limit the space for reporting sites
used to seven alternatives, none of the individuals identified more than 6 sites for local
outings. The average number identified by a respondent who took local outings was
about 2 sites for each zone. This count offers a potentially interesting way to consider the
differences among the recreation trips in our survey. The count of sites an individual
reports that she uses reflects both her desire for variety and the supply of recognized
alternatives to meet each type of recreation.
While local outings are the focus of our recreation demand models, a comparison
of the factors influencing the stated number of sites used for each type of recreation helps
to confirm that people do consider these types of trips as distinct. Table 2 provides a
simple multivariate analysis of the reported counts of the sites used for each type of
recreation based on a Poisson regression model. The second column provides the results
for local outings. Columns three and four report the findings for one day and two day
trips respectively. Several socio-economic characteristics display different influences on
the three types of trips. The most notable of these is income, whichhas a significant
(with a p-value of 0.10) negative influence on the number of places respondents report
for their local outings, but the opposite effect for longer trips. Age and boat ownership
42
-------
have consistent effects on all three types of recreation. Small children appear
constraining for day trips but do not have an impact on the number of sites selected for
other types of recreation. Other variables such as respondent reports of appreciable time
limitations do not affect the count of sites listed but may well affect behavior in other
ways. The distinctive roles for income, race, and small children in these summaries
suggest people appear to evaluate the choice alternatives differently for each type of
recreation
Our simple random utility models for local outings follow the logic outlined in
equation (8). Trips are assumed to be independent choices. Travel time was considered
to be the primary "cost" of a local outing. We also add to this specification two measures
of the quality of the watershed that includes the site. The first of these measures is the
estimated percent of the land area covered with impervious surface. Schueler [1994] and
Cappiella and Brown [2001] have suggested this measure can serve as an indicator to
predict the negative effects of development-related changes in land cover on aquatic
systems. In addition, we use an expert rating of each sub-hydrologic zone as a second
indicator. More specifically, as we noted earlier, in 2003 a private consulting firm,
CH2MHill, completed a commissioned study of Watershed Quality. Wake County had
requested a study to evaluate the county's streams and watersheds as part of a planning
process intended to balance economic development with natural resource conservation
and environmental protection.
Three categories of information were assembled for their assessment: chemical
data on stream quality and concentration of pollutants, biological data on the number and
types of species sensitive to water quality, and physical characteristics related to habitat
43
-------
and geomorphology. Eight-one sub-hydrologic units were classified into healthy,
impacted, and degraded. The breakdown is given as follows:
Rating Number of Sub Hydrologic
Healthy 30
Impacted 38
Degraded 13
We coded a qualitative variable to distinguish sites in degraded hydrologic areas. All of
the sites visited for local outings were either degraded or impacted.
Table 3 provides the estimated random utility models for six of the nineteen
zones. In all of these cases, increases in travel time to reach a site reduced the likelihood
of selecting it.14 The signs for both the impervious surface measure and the qualitative
variable for degraded conditions varied across models estimated for each MLS zones. In
the case of the impervious surface measure, the majority of the estimated parameters
were negative and most of these were significantly different from zero. The qualitative
variable rating sub-hydrologic units (based on the CH2MHill evaluation) was less stable
- with both positive and significant and negative and significant estimates. Our a priori
interpretation of these variables implied that both would have negative effects on the
likelihood of visiting the recreation sites in areas with these conditions. Nonetheless, it is
important to acknowledge that both watershed quality measures refer to spatial zones that
include the recreation sites. They are not specific indexes for each site.15
Despite the mixed record for the influence of these watershed measures on site
choices, the overall effect of impervious surface on our index of water-based recreation
14 For the remaining zones, only one case resulted in a positive estimate for the travel time parameter.
15 In future research we plan to consider linking the available technical measures of water quality in our
database to each site. However, it is also reasonable to ask how the people selecting these sites would
know about these detailed measurements and use them to evaluate the water quality conditions. This issue
will be considered in our further research with the Wake County Database.
44
-------
opportunities provided by each neighborhood is consistent with a priori expectations.
That is, when we consider the mix of sites selected by our sample respondents in each
MLS zone and compute the average value for the expected maximum utility terms (i.e.
equation (10)), the index declines with increases in the impervious surface in the MLS
zone visited for short outings. Equation (13) provides a simple regression of the value of
our index of opportunities on the average impervious surface in the zones with sites
visited.
The numbers in parentheses below the estimated parameters are t-ratios.16
Eu = 31.50 - 2.57 Percent Impervious Surface,
; F ; (13)
(1.73) (-1.85)
n= 19
i?2 = 0.17
The estimates for the expected maximum utility (or the average log sum), EUj,
corresponds to the selections by respondents in each MLS zone, weighted by the
parameters from the zone specific random utility model. The average measure for
impervious surface considers the values for the MLS zones visited through the selection
of sites for local recreation. It is a trip weighted average of these measures. This
expression indicates that the overall pattern described with the index is what we would
have expected. Recreation opportunities are given a lower (quality adjusted) "score"
when there are higher proportions of the land areas in impervious surface in the locations
of the sites visited. There was no significant association when the same analysis was
16 This model is not intended as a test, since the dependent variable is defined using values of the
independent variable from a set of RUM estimates. It is a convenient summary of the net outcomes of the
reported choices in each MLS zone. The objective is to evaluate whether the recreation quantity index
signals to property markets consistently the quality of the recreation sites being selected. In this case it
describes whether increases in the weighted average value for the impervious surface measure are
associated with declines in our quality adjusted index for recreation opportunities.
45
-------
conducted using the average log sum and the average scores based on the CH2MHill
rating.
B. Hedonic Property Value Estimates
Table 4 provides the estimates for our hedonic model, based on sales in 1998 and
1999 using a semi-log specification. The second column reports the estimated effects of
structural characteristics along with two measures of water quality related effects. The
first of these is our quality adjusted index of "value" of access to recreation sites for local
outings, based on equation (10) and the 19 estimated random utility models. The second
is an index of general water based amenities that are also hypothesized to be relevant for
each neighborhood. It relies on the conventional logic that if a house is located on or
near a lake this proximity may be an amenity for the residents, which may influence
property values. To compute this measure, we used Wake County GIS Services to
provide an Arcview shapefile of all lakes in the county. The distance of each house from
all lakes was calculated and the distance to the nearest lake was determined. Since the
amenity effect of lake proximity would decline rapidly with distance from the lake and
would fall to zero at some distance, an index for lake proximity was developed. The
Lake Distance index = max
1-
{ d
V^max J
;0
where d is the distance of the house from
the nearest lake and dmax is the maximum distance where the lake has any effect on the
house value. This index is between zero and one and is convex. As noted in table 1, a
value of 2,640 feet (one-half mile) was used for dmax. The third column reports the means
46
-------
for each of the conventional housing attributes as well as for the sales price (in the row
corresponding to the intercept for the model).
Both measures of the effects of watershed related amenities are significant,
positive determinants of property values. This result suggests that the effects of the
quality adjusted recreation index can be distinguished from a more general index of the
neighborhood related amenities provided by urban watersheds.17 All of the other
structural variables in the model (with the exception of an indicator variable for the
presence of a swimming pool) are significant determinants of the sales prices. The only
potentially implausible relationship implied by the estimates is for the measure of average
commuting time. We expect that increases in commuting time to work associated with
the different home sites would reduce property values. However, this measure could
easily be serving as a proxy variable for the more rural areas in the county and, as a
result, reflect the influence of rural amenities which would also imply greater distances
from employment centers and longer average commutes.
Thus, the hedonic estimates provide strong confirmation for our efforts to
distinguish the long and short run aspects of the influences amenities have on individual
behavior. Our framework implies that a model that describes the role of housing choices
as conditioning factors influencing short term recreation decisions addresses the "double
17 Our quality adjusted index for the amount of each recreation is the average of the expected maximum
utilities as defined in equation ( 10 ). Assigning this to all housing transactions based on their MLS zone
might arguably introduce an errors-in-variables problem (see Moulton [1990]). To evaluate whether this
interpretation affected our estimates for the role of Eu, we considered an alternative estimation strategy.
We estimated fixed effects for each zone in the hedonic model and then used the estimated fixed effects in
a feasible GLS (using the relevant partition matrix from the OLS variance covariance matrix for the
covariance matrix, see Nevo [2001 ] as an example). The estimated effect for the recreation index is nearly
identical to the one step hedonic findings (t-ratios are in parentheses)
ML S_F ixed_effect = 10.881 + 0.004 Eu
(2297.6) (32.271)
47
-------
counting" concerns raised by McConnell [1990], Moreover, with information on
recreation choices it is possible to consistently estimate the distinct roles for
neighborhood amenities and the amenities conveyed through local recreation in an
integrated framework.
C. A Policy Application
One important use of hedonic property value models is for estimating bounds for
the tradeoffs homeowners would be willing to make to improve the amenities available in
their neighborhoods. We can use this logic to evaluate the plausibility of our estimates
for the effects of access to, as well as the quality of, local recreation sites. Equation (12)
provides the algebraic description of the logic involved. To make the analysis tangible
we selected a recent proposal to expand the capacity of a waste water treatment plant
serving a growing community, Butner, NC, that is outside Wake County.18 However, the
change would influence important watersheds in the county. The expansion would
increase the plant discharge from 5.5 to 7.5 million gallons daily. This increase implies
that nitrogen loadings into a tributary of the Neuse River would double.19 The Neuse
Watershed is the most important in Wake County. This river also is the source for water
to Falls Lake, one of the popular recreation sites in Wake County, and a drinking water
source for homeowners in Raleigh and elsewhere in the county. The lake has already
18 Our description is based on newspaper accounts of the proposal in The Raleigh News and Observer,
August 5, 2004, August 7, 2004, and September 24, 2004.
19 The proposed increased discharge from the Butner waste water treatment plant is possible under current
regulations because the Butner facility purchased emission permits from the Bay River Metropolitan
Sewage District near the mouth of the Neuse River. The emissions from the Bay River facility are, for
practical purposes, directly into the Pamlico Sound at New Bern, N.C. The purchase of 6,113 pounds of
this facility's permits translates into 61,130 pounds 200 miles up river because it is estimated that only 10
percent of the added nitrogen up river would reach the end of the Neuse. The permits are defined
exclusively on the basis of nitrogen entering the estuary. They do not consider the intermediate effects of
increased discharges on the river and ecosystem throughout the 200 mile stretch.
48
-------
begun to experience water quality problems even without this expansion. The upper
portion of the lake has measured concentrations of chlorophyll A, exceeding state
standards. Chlorophyll A is related to algae blooms and water quality. Increased
nutrients will accelerate these problems.
To illustrate how our linked model can be used to consider the effects of this
change, we assume that granting the expansion permit would imply Falls Lake was no
longer an attractive recreation site for local outings due to the continued degradation
associated with the increased nitrogen loadings into the lake. To represent this change
we removed Falls Lake from the choice set describing available sites in each of the 17
MLS zones where it was a choice alternative. This process allows us to compute the
change in expected maximum utility, our quality-adjusted index of the "quantity" of local
recreation opportunities available to each housing market. We then use the estimated
hedonic price function to compute an upper bound for homeowners' annual willingness
to pay to avoid this change. Our estimates use the adjusted predictions for the housing
prices and a five percent rate to compute the annual estimate of the bound for the
willingness to pay.20
Table 5 reports these estimates in the second column along with the proportionate
change in the expected maximum utility (in the third column), the average sales prices for
housing by MLS zone (in the sixth column), and some other information to gauge the
importance of the Falls Lake site for the survey respondents in each MLS zone. These
summary statistics are given in the last four columns. First, we list the total number of
local outings reported to be taken by our survey respondents and the average per
20 The adjusted price is an approximation to reduce the bias in predicting the price from a semi log model.
For simplicity we assume housing is completely durable, so the annual value is the discount rate times the
sales price.
49
-------
respondent in the seven month period covered by our survey. To evaluate the potential
importance of Falls Lake to these residents, we also computed the number of these
outings that were to the Lake and the average outings per user. Economic importance is
not exclusively associated with the count of trips. It will also depend on the alternatives
available with comparable proximity and quality.
Comparison of the values measured for the loss together with the total outings
versus the outings to Falls Lake confirms this conclusion Some of the larger values arise
when there are a number of outings to other sites, as in the case of MLS zones 2, 4, and 7.
The Falls Lake site makes an important contribution to the quality adjusted index of the
amount of recreation opportunities available in an area, thus relying on the pattern of use
alone would be misleading. Of course, the largest values for MLS zone 14 (close to the
Lake) arise where residents perceive few alternatives.
Finally for comparative purposes we report the average value for the change in
per trip consumer surplus (also in 1998 dollars) due to the loss of Falls Lake as a choice
alternative by MLS zone. This estimate is based on each zone's random utility model
and the change in the expected maximum utility with and without Falls Lake in the
choice set. There are two added steps required to compute it. First, the difference in the
average log sums with and without Falls Lake is divided by the absolute magnitude of the
parameter estimated for travel time. We could consider this ratio as expressing the
willingness to avoid the loss, in time units - the amount of free time a person would give
up rather than close Falls Lake from their choice set. To monetize this time, we make use
of some related research with this same sample (see Palmquist, Phaneuf, and Smith
[2004]). This work hypothesized that the opportunity cost of time varies based on the
50
-------
amount and timing of the time required for recreation. We use the time allocations of our
survey respondents along with their willingness to pay to substitute market services for
some home production to estimate this opportunity cost. Our analysis suggests the
marginal opportunity cost varies with the amount of time required. For these
computations we used the marginal value for a 4 hour trip and adjusted the estimated
parameters for time costs. Using this average value by MLS zone (and given in column
six) together with the willingness to give up time, it is possible to develop an approximate
estimate of the per trip consumer surplus. If we scale this willingness to pay by the
average number of outings taken by our sample respondents to all sites (given in column
eight) we see the product is generally less than the long run value implied by the hedonic
estimates. While there is no reason to expect the short run and long run estimates would
be equal, there is clear consistency between the two. More specifically, the two methods
are monetizing the same increment in the index for the change in recreation opportunities.
The hedonic uses the long run market capitalization of these opportunities (in annualized
terms) by the housing market. The monetizing of the same index uses another market -
based on labor/leisure choices and time allocation choices when respondents were offered
short run options for adjustment. In the absence of uncertainty and with limited
adjustment costs, we could specify an envelope condition that would imply equality in
these values.21 The close correspondence for our approximation implies there is scope
for using housing markets together with structural models of how ecosystem services
contribute to people's activities in developing revealed preference estimates for fairly
complex patterns of spatial effects on behavior, provided we can rely on people observing
how these services contribute to the quality of their activities.
21 This condition is what McConnell [1990] was implicitly describing.
51
-------
V. Implications
Measuring people's valuation of water quality and watershed services with
hedonic property value models has proved difficult. Leggett and Bockstael [2000]
suggest that despite consumers' reports indicating they want to live near water resources
for the recreational opportunities they offer, there are often few opportunities for analysts
to observe sufficient local differences in recognizable water quality conditions to measure
their effects. As a result, these authors highlight the distinction between the geographic
extent of the housing market and the likely spatial variation in water quality conditions.
To estimate consumers' responses to differences in the services provided by improved
water related resources within a hedonic framework there must be sufficient -variation in
the measure hypothesized to characterize these services. Often this is not the case.
Properties on a single lake are unlikely to experience markedly different water quality.
Their analysis of the sales of waterfront properties on the western shore of the
Chesapeake Bay was successful in estimating a water quality effect using a distance
weighted average of and index for the bacterial contamination (i.e. the fecal coliform
counts from 104 monitoring stations). The water quality measure exhibited sufficient
spatial variation to evaluate its effect on coastal property values. Mahan et al. [2000] also
found that proximity to streams, lakes and some types of wetlands increase property
values in Portland. However, their efforts to estimate second stage inverse demand
models were not successful. They also acknowledge the important role the spatial extent
of the market plays in these types of analyses.
52
-------
Our research suggests that there are several distinct roles for the services of
environmental resources. A single proxy index is unlikely to be able to adequately reflect
all of them. One of the first of these roles is as a neighborhood amenity. This
contribution is the one most widely recognized in the hedonic literature. A second role
often acknowledged in discussions of the importance of ecological services but with little
specific discussion in the hedonic literature arises when their services make a supporting
contribution to other activities. Some of these involve people and their outdoor
recreation trips. Others involve related natural resources, such as groundwater, whose
quality and recharge rate can be influenced by the characteristics of watersheds.
The spatial boundaries relevant for these various influences across different sets
of activities need not be the same. Neighborhood amenity effects are likely to be
associated with the immediate proximity of a house, as our index of access to close lakes
implied. It is less clear how to characterize the roles for other influences. Most hedonic
studies have relied on some distance based index. We have suggested an alternative
approach.
Our framework considers the decisions used in revealed preference models
applied to environmental services as alternative strategies to recover information about
the importance of these services to people. Each describes a different choice margin. To
integrate their results, the various choices must be described consistently. In this paper
we used the long run/short run distinction to integrate local recreation choices with
residential housing decisions.
Efforts to propose some type of integration are not new to non- market valuation.
One was the basis for Cameron's [1992a, 1992b] proposal to use revealed preference
53
-------
behavior to impose "budget discipline" on the stated preference choices people make for
the same resource. A proposed strategy for integration is also the basis for the
maintained assumptions that Smith, Pattanayak, and Van Houtven [2002] use to calibrate
preference functions for benefit transfers. In our case here, however, there is an
important distinction. Revealed preference models are used to construct a quantity index
that collapses the recreation opportunities available to those living in a neighborhood.
This index is derived from a model of recreation demand. The model reduces the
complexity of all the attributes and availability measures for the local recreation sites into
a consistent quantity index. It also defines the spatial domain of influence through the
choice set of recreation sites considered relevant for the model.
The equilibrium housing price will be influenced by this index because the
opportunities differ across the neighborhoods comprising a housing market. McConnell
[1990] describes this prospect as a potential source for double counting. He argues that
property values capitalize the expected future values derived from the available
recreation services due to a location. Our use of the expected value of the maximum
utility available from a recreation choice set is consistent with his suggestion that it is not
a large leap to propose that".. .the present discounted value of pollution damage from an
ex ante concept, containing valuation of expectations of future choices" (p. 126). The
potential for connections does not stop here. Rather, our proposed logic offers the means
to consider other watershed services, provided there is a basis for using current choice
margins to describe how these services contribute to current decisions.
Our empirical example exploits prior information describing the types of
recreation likely to be associated with choices among alternative housing neighborhoods.
54
-------
The measure of these recreation alternatives in a location was a consistent and significant
determinant of housing prices and the specification also controls for the effects of more
general amenity effects of proximity to local water resources. We developed estimates
for the value of avoiding the loss of a popular recreation site in the northern portion of
Wake County using both the hedonic bounds and the random utility models. The results
are consistent with interpreting the hedonic as ex ante bound for the incremental value of
the expected future services from protecting the lake.
55
-------
Table 1: Primary Variables for Empirical Analysis
Name
Definition
A. Hedonic Variables
lprice
Natural log of sale price of property
baths
Numb er of b athroom s
regheatarea
Main heated living area in square feet
age
Age of structure, calculated as sale year-year built
acreage
Lot size in acres
sewer
Community sewer system
bsmtheat
Basement heated area in square feet
atticheat
Attic heated area in square feet
encporch
Enclosed porch area in square feet
scrporch
Screened porch area in square feet
opnporch
Open porch area in square feet
garage
Garage area in square feet
Deck
Deck area in square feet
fireplaces
Number of fireplaces
detgarage
Dummy variable indicating presence of detached garage
wallduml
Dummy variable indicating presence of brick walls
bsmtduml
Dummy variable indicating presence of full basement
bsmtdum2
Dummy variable indicating presence of partial basement
floorduml
Dummy variable indicating presence of hardwood floors
poolres
Dummy variable indicating presence of residential
swimming pool
condadum
Dummy variable indicating house is of condition A
(highest)
condcdum
Dummy variable indicating house is of condition C
condddum
Dummy variable indicating house is of condition D
commuting time
Average travel time to work computed for workers 16 years
and older by block group based on 2000 census
Lake Distance Index
r f d Y2 1
this variable is measured in feet as the max 1- ,0
V^max J
where d is the distance of each house to the nearest lake and
dnmx is the maximum distance, assumed to be '/2 mile (2,640
feet)
B. Recreation Variables
Travel time
Time in minutes for one way trips from respondent's home
to recreation sitea
3 The travel time and travel distance between each survey respondent's house and each recreation site were
calculated using the PCMiler software. PCMiler calculates distances between lat/long points along a road
network, and then estimates travel times using speed limit information. It has been commonly used in
travel cost models to calculate travel distances and times, however it is designed for the trucking industry.
56
-------
Distance
One-way distance in miles from respondent's home to the
recreation timea
C. Watershed Variables
Percentage Impervious
Surfaceb
Measure of fraction of land area in MLS zone covered with
impervious surface
Sub Watershed Rating
Classification of 81 sub-hydrologic units in Wake County as
healthy, impacted, or degraded, based on the CH2MHill
evaluation of the state of the County's watersheds
D. Household Survey
less high
qualitative variable = 1 if less than high school education
Fine
family income (in dollars)
male
qualitative variable = 1 if respondent is male
white
qualitative variable = 1 if respondent is white
age of respondent
age in years
children less than 6
number of children less than 6 years of age
time limited
qualitative variable = 1 if respondent indicates leisure time
is limited
boat own
qualitative variable = 1 if respondent owns a boat
Thus one of PCMiler's drawbacks is that the road network it uses to calculate the times and distances is
composed of roads that are accessible to trucks. The error in estimating travel times and distances using a
network of major roads accessible to trucks is likely to be largest for the sites used for local outings.
To decrease the measurement error that might be introduced with PCMiler in these cases, an alternative
strategy was developed using Arcview. Using a comprehensive road network including minor roads
developed by "Tigerline" (Census 2000 TIGER/Line Data is provided by the U.S. Bureau of the Census)
for Wake County, travel times and distances from survey households to local recreation sites were
calculated within Arcview. One exception was the calculation of times and distances to Jordan Lake, a
popular local recreation destination located just outside of Wake County. By calculating the travel time
and distance to the County line, and the adding this time and distance to the PCMiler estimate from the
county line to the site, a more accurate time and distance was generated to this recreation site. To
determine if the other Arcview estimates were more accurate than the PCMiler estimates for the local
recreation site, a sample of 20 households and 10 recreation sites were compared to estimates produced by
Mapquest, an online service that also uses major and minor roads in their calculations. Based on Sum of
Squared Errors, it appears that the Arcview estimates were more accurate than the PCMiler estimates for
the local recreation sites. Thus we replaced the PCMiler times and distances with the Arcview times and
distances for local recreation sites.
b To create a measure of percent imperviousness for other geographic areas in our study, we used the same
procedure employed by CH2MHill. Land use types were classified into 17 classes. We used the
CH2MHill estimates for percent impervious surface measures for each of the 17 land use classes. The
amount of each land type in each area was then weighted by these percentages to measure the impervious
surface for the geographic area of interest.
57
-------
Table 2: Determinants of Count of Sites Used by Type of Recreation Tripa
Independent
Variables
Local Outing
Day Trip
Two Day Trip
less high( =1)
-0.142
(-0.40)
0.114
(0.25)
0.382
(0.85)
Fine
-0.067xl0"6
(-1.94)
-0.022x10"6
(-0.47)
0.073xl0"6
(2.10)
male (= 1)
-0.041
(-0.93)
0.033
(0.53)
-0.060
(-1.23)
white ( = 1)
0.116
(1.64)
0.085
(0.85)
0.262
(3.07)
age of respondent
-0.013
(-5.18)
-0.013
(-3.75)
-0.010
(-3.65)
children less than 6
-0.023
(-0.80)
-0.144
(-3.20)
-0.014
(-0.45)
time limited ( = 1)
0.030
(0.57)
-0.009
(-0.13)
-0.029
(-0.50)
boatown (= 1)
0.263
(4.80)
0.586
(8.49)
0.296
(5.11)
intercept
0.807
(5.93)
0.232
(1.22)
0.178
(1.19)
no. of observations
1572
1354
1641
pseudo R2
0.013
0.028
0.015
3 These estimates are based on a Poisson regression model with the number of recreation sites treated as a
count variable. The numbers in parentheses are ratios for the estimated coefficients to their asymptotic
standard errors.
58
-------
Table 3: A Sample of Random Utility Models by MLS Zone
Independent
MLS Zone"
Variables
1
5
7
14
15
18
Percent Impervious
-0.035
0.032
-0.306
-0.028
0.005
-0.215
Surface
CffiMHill
Rating = Degraded
(=1,0 otherwise)
(-3.71)
0.773
(8.25)
(4.97)
0.333
(7.92)
(-26.27)
3.406
(23.89)
(-1.11)
-0.645
(-2.03)
(0.69)
-0.863
(-6.93)
(-7.57)
2.882
(9.241)
Travel Time
-0.070
-0.109
-0.126
-0.138
-0.193
-0.125
(-8.32)
(-30.94)
(-36.66)
(-13.37)
(-19.99)
(-14.39)
3 The numbers in parentheses are the ratios of the estimated parameters to its estimated asymptotic standard
error for the null hypothesis of no association.
59
-------
Table 4: Hedonic Property Value for Sales in 1998 and 1999
Independent Variables
Model"
Means b
Index of Recreation Access'
0 004
CS 50)
Luke Distance Index1'
o o|4
(2.57)
age
-0.002
(-17.54)
11.37
(15.06)
baths
0.036
(23.97)
2.46
(0.68)
acreage
0.042
(29.84)
0.45
(0.62)
regheatarea
0.039xl0"/
(204.43)
1,914
(681.24)
detgarage
0.085
(18.83)
0.03
(0.17)
fireplaces
0.068
(29.16)
0.91
(0.34)
deck
0.019xl0_/
(33.42)
159.35
(145.05)
sewer
0.013
(5.60)
0.83
(0.37)
floorduml
-0.015
(-4.35)
0.10
(0.30)
scrporch
0.034xl0"/
(25.40)
16.78
(55.88)
atticheat
0.023xl0"/
(42.96)
43.21
(142.27)
bsmtheat
0.058x10"'
(11.22)
48.39
(199.66)
garage
0.030x10"^
(69.80)
289.46
(248.82)
poolres
0.006
(0.76)
0.01
(0.09)
bsmtduml
0.133
(30.93)
0.052
(0.221)
bsmtdum2
0.138
(35.70)
0.063
(0.243)
3 The numbers in parentheses are t-ratios for the null hypothesis of no association.
b Numbers in parentheses are standard deviations.
c The Index of Recreation Access corresponds to the average values across properties sold for the log sum
derived from the parameter estimates for the random utility model associated with each house's MLS zone.
The Lake Distance Index is
l-(d/265o¥>0
with d the distance from the home to the nearest lake.
60
-------
wallduml
0.038
(14.22)
0.113
(0.317)
encproch
0.196x10"'
(7.32)
3.72
(28.11)
opnporch
0.169x10"'
(19.10)
68.46
(85.17)
condadum
0.231
(62.58)
0.06
(0.233)
condcdum
-0.139
(-28.22)
0.03
(0.158)
condddum
-0.310
(-19.09)
0.02X10"1
(0.041)
commuting time
0.005
(47.50)
28.61
(8.37)
Year of Sale = 1999
0.036
(24.22)
Intercept
10.853
(2,224.10)
180,202®
(75,564)
no. of observations
38,725
R2
0.861
e The average value of the sales price.
61
-------
Table 5: WTP Bounds for Removing a Recreation Site
MLS
Zone
Hedonic
Bound for
WTPa
Proportionate
Change in
Recreation
Index
Benefits -
RUM
per tripb
Marginal
Value
of Time
(per hour)
Average
Housing
Price
Local Outings
Total
- All Areas
Per User
Falls Lake Outings
Total PerUser
1
1.49
0.039
0.30
32.24
208,851
920
5.75
82
3.7
2
3.40
0.090
0.31
27.18
188,630
3,414
10.25
485
6.6
3
1.59
0.061
1.18
45.25
107,387
143
4.09
13
6.5
4
7.32
0.235
0.37
20.96
136,030
854
8.80
6
1.5
5
0.16
0.004
0.02
28.78
207,699
3,150
7.93
54
2.8
6
0.15
0.006
0.01
20.85
123,880
523
7.58
4
2
7
12.87
0.279
1.15
31.12
225,316
2,058
6.77
958
8.3
8
3.03
0.106
0.14
25.72
134,764
698
5.97
118
5.1
9
0.57
0.014
0.08
19.74
195,453
1,124
4.89
21
3.5
10
-
-
-
18.33
232,327
657
7.30
0
0
11
1.34
0.049
0.28
26.77
123,567
333
4.01
56
4.3
12
17.31
0.830
0.48
28.89
104,258
96
4.00
47
5.9
13
10.31
0.332
C
15.52
123,211
58
2.64
14
2.3
14
58.86
1.606
3.86
19.95
177,824
310
4.25
217
5.7
15
0.06
0.002
0.005
32.57
174,766
869
6.30
3
1
16
2.14
0.069
0.57
26.53
138,287
366
3.62
6
2
17
—
—
—
20.00
155,304
142
4.58
0
0
18
0.41
0.012
0.03
20.02
157,528
401
4.66
10
1.7
21
1.81
0.056
1.09
30.99
148,560
383
4.30
168
4.8
3 These estimates are in 1998 dollars. They use the predicted price and adjust for the bias in converting from the predicted In/? to a predicted price.
p = exp (in p)-(l + y2 var(ln p)fl (see Kennedy[1983] forfurther details).
b Converted to 1998 dollars using the Consumer Price Index.
c The estimated parameter for time cost of travel was positive for this model. As the seventh column indicates, this MLS zone had the smallest number of local
outings generated, with only 22 users.
62
-------
Figure 1: MLS Spatial Zones for Housing Submarkets in Wake County, NC
63
-------
-------
References
Atasoy, Mary, Raymond B. Palmquist, and Daniel J. Phaneuf, 2003, "Land Use Patterns
and Pollution in the Upper Neuse," presented at Water Resources Research
Institute Annual Conference, Raleigh, April.
Bateman, Ian J., Matthew Cole, Phillip Cooper, Stavros Gerogiou, David Hadley and
Gregory L. Poe, 2004, "On Visible Choice Sets and Scope Sensitivity," Journal of
Environmental Economics and Management, 47(January): 71-93.
Cameron, Trudy Ann, 1992a, "Combining Contingent Valuation and Travel Cost Data
for the Valuation of Non-market Goods Land Economics, 68(3): 302-317.
Cameron, Trudy Ann, 1992b, "Nonuser Resource Values," American .Journal of
Agricultural Economics, 74(5): 1133-1137.
Cappiella, Karen and Kenneth Brown, 2001, Impervious Cover and Land Use in the
Chesapeake Bay Watershed, report by the Center for Watershed Protection for
U.S. EPA Chesapeake Bay Program, January.
CH2MHill Inc., 2003, "Wake County Final Plan," Wake County Comprehensive
Watershed Management Plan, Charlotte, NC, February. (This document and other
related documents can be found at http://projects.ch2m.com/WakeCounty).
Cropper, Maureen L., Leland B. Deck and Kenneth E. McConnell, 1988, "On The Choice
of Functional Form for Hedonic Price Functions," Review of Economics and
Statistics, 70(November): 668-675.
Dillman, Don A., 1978, Mail and Telephone Surveys: The Total Design Method, (New
York: John Wiley & Sons).
65
-------
Egan, Kevin J., Joseph A. Herriges, Catherine L. Kling, and John A. Downing, 2004,
"Recreation Demand Using Physical Measures of Water Quality," Department of
Economics, Iowa State University, Working paper, (August).
Farber, Stephen and Brian Griner, 2000, "Valuing Watershed Quality Improvements
Using Conjoint Analysis Ecological Economics, 34(1): 63-76.
Feenstra, Robert C., 1995, 'Exact Hedonic Price Indexes," Review of Economics and
Statistics, 77(4): 634-653.
Feitelson, E., 1992, "Consumer Preferences and Willingness to Pay for Water-Related
Residences in Non-Urban Settings: A Vignette Analysis," Regional Studies,
26(1): 49-68.
Johnston, Robert J., Stephen K. Swallow and Thomas F. Weaver, 1999, "Estimating
Willingness to Pay and Resource Tradeoffs with Different Payment Mechanisms:
An Evaluation of a Funding Guarantee for Watershed Management," Journal of
Environmental Economics and Management, 38(July): 97-120.
Kennedy, P.E., 1983, "Logarithmic Dependent Variables and Prediction Bias," Oxford
Bulletin of Economics and Statistics, 45( ): 389-392.
Leggett, Christopher G. and Nancy E. Bockstael, 2000, "Evidence of the Effects of Water
Quality on Residential Land Prices "Journal of Environmental Economics and
Management, 39(2): 121-144.
Mahan, Brent L., Stephen Polasky, and Richard M. Adams, 2000, "Valuing Urban
Wetlands: A Property Price Approach," Land Economics, 76(1): 100-113.
McConnell, K.E., 1990, "Double Counting in Hedonic and Travel Cost Models," Land
Economics, 66(2): 121-127.
66
-------
Moulton, Brent R., 1990, "An Illustration of a Pitfall in Estimating the Effects of
Aggregate Variables in Micro Units," Review of Economics and Statistics, (May):
334-338.
Nevo, Aviv, 2001, "Measuring Market Power in the Ready-to-Eat Cereal Industry,"
Econometrica 69 (2): 307-342.
Palmquist, Raymond, 2004, "Weak Complementarity, Path Independence, and the Willig
Condition," Journal of Environmental Economics and Management (in press).
Palmquist, Raymond B., Daniel J. Phaneuf and V. Kerry Smith, 2004, "Differing Values
of Time for Non-Market Valuation: A Theoretical and Empirical Comparison,"
Paper presented to American Agricultural Economics Association, Denver,
Colorado, (August).
Parsons, George R., 1991, "A Note on Choice of Residential Location in Travel Cost
ModelsLand Economics, 67(3): 360-364.
Phaneuf, Daniel J., 2002, "A Random Utility Model for Total Maximum Daily Loads:
Estimating the Benefits of Watershed-Based Ambient Water Quality
Improvements," Water Resources Research, 38 (11): 36-1 - 36-11.
Pollak, Robert A., 1969, "Conditional Demand Functions and Consumption Theory,"
Quarterly Journal of Economics, 83 (February): 60-78.
Poor, P. Joan, Kevin J. Boyle, Laura O. Taylor, and Roy Bouchard, 2001, "Objective
versus Subjective Measures of Water Clarity in Hedonic Property Value Models,"
Land Economics, 77(4): 482-493.
Schueler, T., 1994, "The Importance of Imperviousness," Watershed Protection
Techniques, 1(3): 100-111.
67
-------
Smith, V. Kerry, George L. Van Houtven and Subhrendu K. Pattanayak, 2002, "Benefit
Transfer via Preference Calibration: Prudential Algebra for Policy,"Land
Economics, 78(1): 132-152.
Willig, Robert D., 1978, "Incremental Consumer's Surplus and Hedonic Price
Adjustment," .Journal of Economic Theory, 17: 227-253.
Woodard, Richard T. and Yong-Suhk Wui, 2001, "The Economic Value of Wetland
Services: A Meta-Analysis " Ecological Economics, 37(May): 257-270.
-------
Appendix A
Table 1A: Survey Assignments and Returns by MLS Zone
MLS Zone
Surveys Mailed3
Survevs Returned
Proportion Returned
1
358
107
0.299
2
1055
334
0.317
3
289
101
0.349
4
179
62
0.346
5
1401
430
0.307
6
281
87
0.310
7
840
260
0.310
8
481
149
0.309
9
525
162
0.323
10
334
108
0.342
11
284
97
0.361
12
97
35
0.457
13
81
37
0.306
14
222
68
0.352
15
270
95
0.296
16
274
81
0.417
17
48
20
0.265
18
230
61
0.334
21
305
102
0.317
Total
7554
2396
0.317b
a Errors in record keeping caused the version number of four surveys to be omitted from the tracking system.
As a result, they are included here but not used in subsequent analysis.
b Proportion returned for the county as a whole.
-------
Session III: Keeping Water Fresh: The Value of Improved Fresh Water Quality
Discussant Comments — John Powers
October 26, 2004
Summary of Egan et al.
The purpose of the research is to provide information on the recreational value of water
quality improvements as a function of detailed physical attributes of water bodies. The
research supports Iowa's effort to comply with Clean Water Act requirement to develop
TMDLs for impaired waters. The policy effort involves identifying priority waterbodies
and strategies for allocating resources through the use of cost and benefit information on
remediation efforts.
The research includes developing a recreation demand (travel cost) model of recreational
lake usage in Iowa. Unique data collected by Iowa State University's Limnology
Laboratory an extensive array of physical attributes, including Secchi depth, chlorophyll,
3 nitrogen measures, phosphorus, silicon, acidity (pH), alkalinity, and 2 suspended solids
measures. The researchers are also collecting site and household characteristics data .
The authors use a mixed logit model that integrates site selection and participation
decisions in a utility consistent framework. They estimate model specifications that
differ in the numbers of physical water quality measures in order to test for the stability
of parameter estimates.
The estimated parameters are generally of the expected sign, with water clarity being
valued highly. Some variation in water color is acceptable, but high algae levels lead to a
reduced number of trips taken.
The results indicate that policy makers can maximize benefits by spreading
improvements across the state, and by improving a smaller number of lakes to high
quality rather than raising a large number of lakes to average quality.
Comments
Overall, this is a nice paper, but it is only a small fraction of the bigger project. I like the
idea of testing to see how recreation demand is affected by physical characteristics of
water quality. The results suggest people know about and respond to easily observed
characteristics, especially water clarity. But, the impact of nitrogen is less observable,
and also very valuable, since it affects the nitrogen-phosphorous balance, and has impacts
in estuaries.
Several questions for the researchers
How does limited information about water quality affect the welfare estimates? Does the
valuation methodology affect your answer (e.g., SP rather than RP)? Does knowing the
source of contamination affect "value"?
70
-------
Summary of Smith et al.
The purpose of the research is primarily methodological. The authors examine choice
margins using hedonic and recreation demand (travel cost) models through integrated
estimation of hedonic property values (long run decisions) and recreation demand values
(short run decisions). Expected recreation benefits are viewed as a (public) attribute of
home location. The results are used to estimate the impact (lost water quality benefits) of
recent proposal to expand capacity of wastewater treatment plant serving a growing
community (Butner, NC). In this example, the authors estimate willingness to pay to
avoid the adverse impact on recreation and drinking water in Wake County, NC of
nitrogen loads into the Neuse River.
This study attempts to estimate choice margins across a larger set of alternatives, by
using hedonic and travel cost methods in an integrated manner. This analysis involves
using GIS-based models and data, socio-economic data, water quality measure, property
sales, and recreation data for over 2,000 households. Several indexes are used in this
study, including a Watershed Quality Index , which is used to summarizes effects of
watershed quality measures on local recreation opportunities, a Recreation Quantity
Index (Index of Recreational Opportunities), which is used to capture the expected
benefits from recreation, and a Lake Distance Index, which is an index of lake proximity,
accounting for distance of the house from the nearest lake, and the maximum distance
from where the lake has any effect on house value.
Comments
I like the idea of integrating different sources and types of benefit information, although I
am concerned about the complexity. I also think the use of the indexes is quite
interesting, though I would like to see greater attention given to index structure and
theoretic rationale. Finally, does it matter whether we measure from the household
location to the resource location, or vice versa?
General comments
The "commodity" definition is important to benefit transfer so that apples-to-apples
transfers are possible. If we think of an ecosystem production function, then we can
obtain value estimates for ecosystem outputs, such as safe drinking water or safe water
for swimming (e.g., CWA designated uses), or ecosystem inputs, such as biophysical
characteristics (e.g., pollutant concentrations). What is/are the "policy-relevant"
definition(s) of water quality? And are certain valuation methods are to different
definitions?
Indexes are alluring but can be tricky. They can help facilitate analysis, and can help
communicate complexity, but without a simple theoretical structure, they can also be
easily criticized.
71
-------
How do we reconcile these tensions? Benefit transfer is commonplace, as policy analysts
at all levels of government look to the literature for benefits information. How do you
(researchers) feel knowing that your published findings could be used to estimate the
benefits of a policy beyond the immediate scope of your study (e.g., benefits of
agricultural nutrient controls in New England, or Minnesota, or the whole US). For the
researchers, how does it feel to know that your research findings may be used in a
transfer? Also, how does the geographic scale of your work affect your current
research? Could you shift your research into a regional or national scale?
How high are the transaction costs associated with multidisciplinary research?
72
-------
Summary of the Q&A Discussion Following Session III (Part 1)
Steve Swallow (University of Rhode Island)
Directing his question to Joseph Herriges, Dr. Swallow stated that he noticed most, or
perhaps all, of the water quality dimensions were linear. He said he was "just thinking
about how a lake ecosystem works—maybe I should talk about nutrients, maybe I should
talk about Secchi depths, whatever, but if you can see to the bottom of a lake, that's a
lake that doesn't have a lot of nutrients in it." He commented that although that may be
aesthetically pleasing, it's not necessarily good for fish and, therefore, might be affecting
the "recreation quality." Dr. Swallow continued: "If you raise the Secchi depth up to
zero, that means it's eutrophic—everything is growing, and it probably stinks, too." He
closed by asking whether Dr. Herriges has "thought about trying to do some non-
linearities where there might be a peak in the quality from the perspective of what
humans are valuing but a difference from the peak in the quality from the perspective of a
pristine ecosystem that some of your ecology friends might be focused on."
Joseph Herriges (Iowa State University)
Dr. Herriges explained that he didn't really have time to talk a lot about the specification
search part of the study, but a lot of that came from talking with, in this case, the
limnologist. He continued, " I showed you a real simple version of the model, but what
we're doing in the specification search is looking over a variety of models with both
linear and non-linear effects—non-linear effects in terms of Secchi depth and things like
that—so we have looked at a whole range of different models and we have found non-
linear effects in a number of the variables. In the specification stage, we're searching
over both whether to include a variable in the model and also what non-linear form we
have." Dr. Herriges stated that "one of the things coming out of this conference is that
we need to go back and look at some more non-linearities in the process. The limnologist
has been particularly helpful in pointing out which variables might be the ones we focus
on because they have physical signs that people visiting the lake might see. So, that's
why we had the six variables I showed you—those are the ones we focused on initially
because the nutrients and so on have particular physical attributes that people can see."
He concluded by reiterating that they "have all the other variables in and have looked at a
lot of different non-linearities."
Steve Swallow
Dr. Swallow added, "It also might affect the difference you're seeing between different
user groups—the people who get in the water versus the people who are on top of the
water fishing."
Joseph Herriges
Dr. Herriges replied, "In fact, that's something we haven't done, which I think would be
useful to do. We have not looked at segmenting the population. The problem there is
that different user groups are somewhat endogenous—if you don't like certain types of
water quality, you may not be a swimmer because you don't like the physical attributes,
so there's a bit of a problem modeling what people choose to do."
73
-------
David Widow sky (U.S. EPA/OPP)
Identifying himself as "both a producer and consumer of ecological benefits analysis,"
Dr. Widawsky said he wanted to bring the focus back to the subtitle of the workshop:
Improving the Science Behind Policy Decisions. Referring to "the talks we heard just
now and some of the talks we heard this morning," he stated "policy decisions are often
presented as a choice between one set of biophysical properties and another set of
biophysical properties. We know that in getting to that different set of biophysical
properties, the real decision is not choosing that set of properties but choosing a set of
land use decisions and behaviors that are linked to the properties and which we can value
through the biophysical models that are kind of the challenge. As we heard this morning
from Nicole and in the keynote address, the challenge is with respect to having an
integrated model between an ecological assessment model, an ecological valuation
model, and an economic model. My question is: To what degree do you gentlemen
incorporate biophysical models to describe how all of this gets you to the sub-
characterization of value and what challenges were expected getting to an integrated
model, and . . . how would you address those challenges?"
Joseph Herriges
Dr. Herriges responded, "The quick answer is: We did not look at that." He continued,
"What we're doing in our project, for example, is trying to look at the value that the
households place in certain attributes of the lakes, certain water quality levels. I think the
question you're addressing is that there's a cost associated with that. You have to
understand that if you're really going to evaluate whether to adopt a [program] to try to
get these lakes up to a given level of quality, we want to know the benefits of that, and
that's really what our project is looking at. But, you also need to know the costs of doing
that, so you need to be able to model the fate and transport of the various pollutants
getting in, how different incentives might cause changes in land use and how those then,
in turn, affect the water quality." Reiterating that that's a different issue outside the
scope of their study, Dr. Herriges commented, "There is actually a project going on at
Iowa State University in the Center for Agricultural and Rural Development trying to do
exactly that—trying to pair up our work on getting at the benefits with their own work of
trying to model the cost of achieving different levels of water quality through different
incentives on land use and set aside and so on.
Kerry Smith (North Carolina State University)
Dr. Smith said, "I should say that I was, as usual, not very clear on what the benefit we
measured was. The benefit measured, which I presented at the very end, was just the
elimination of the site, in this case Falls Lake, from the choice set. Ray [Palmquist] and
Dan [Phaneuf] have done some separate work that, as we develop this index function I'm
talking about, would be capable of being used, but it's . . . sort of a reduced form model.
What they've done is they've put their variety of different measures based on monitoring
74
-------
data in the Neuse River watershed of ambient concentrations of different pollutants—
nitrogen, phosphorus, and so forth—and then set up a spatial model that takes account of
both the timing of the monitoring and the timing of the activities, and in this case it's
changes in land cover and land use at points that are upstream of the places where the
measurements are taken. Now, in principle, if we had those physical attributes that they
are describing in the largely reduced form model, conveyed through our index, then it
would be possible to make somewhat of a connection. The difficulty is that the closed
loop . . . isn't in our model—the closed loop being: suppose we were to take this reduced
form model that they've developed that looks at land cover changes and new
development (new building permits, new land conversion, and so forth) and it takes that
and it links it to total measured phosphorus or nitrogen or something else at a particular
point in the river. That gets conveyed through our index up to our housing model, and
we say "Okay, no problem, we're just going to put some limits on here—we'll refer to
them as brand new versions or something else so it will simulate that effect in the reduced
form model that they've got, then that will connect to our index, and we'll just value it in
the Hedonic model. The problem is that the Hedonic equilibrium is different because it's
restricting the nature of the land use that's associated with getting the outcome in the
beginning of the model, so the feedback would be passively put in there.
So, the short answer that I should have said was: No, we didn't do that. Those of you
who have listened to me so far today know that I never give any short answer to
anything."
Robert LaFrance (Connecticut Department of Environmental Protection)
Mr. LaFrance said, "I've listened to a lot of your academic discussions and I'm
wondering: How do you guys relate to local and state officials, both those who are
elected and those who are not? Maybe you can give me some sense of your interaction
with them, because that's kind of where I'm at and I'm trying to take some of this and
bring it back to my job."
Joseph Herriges
Dr. Herriges responded, "I'm not sure how to answer that question. This gets back to this
whole issue about interdisciplinary research, too. There are interactions between
ecologists and economists, and getting the communication between those two different
disciplines and communicating with the local and state regulators in the process [is often
difficult]. There are costs associated, but there are huge benefits as well. In our project
the limnologist actually started this interaction by calling and saying that they were doing
this extensive study and would like some economic numbers at the end—economists are
used to being called in at the very end. Well, what's happened in the process of both of
us talking to each other is that the project has evolved into something bigger and
broader." He said this expansion of the project involved "bringing in local people and
finding out what matters to them in terms of the lake—what changes they're looking at,
their interest in local economic impact versus what economists would say in terms of
changes in value, etc." Dr. Herriges summarized by saying, "Communication is
75
-------
extremely important. We've learned a lot by talking to state regulators, and what matters
to them is they want to know that people will actually do something as a result and that
they'll contact their state legislator and there will be action coming out of this. So,
learning how to communicate with each other is extremely important in this process.
Kerry Smith
Dr. Smith advised Mr. LaFrance to talk to his colleague Dan Phaneuf about that issue,
"because he's really had much more experience, not only in the context of developing
this model I refer to, but in the context of working with some folks at RTI (RTI
International, Inc.) at integrating a watershed model with an economic model for a large
local project." He went on to relate this story: "Many years ago I was asked to pretendI
was an expert witness at a mock trial that took place in New York City—this was about
twenty years ago—and the best way of characterizing me interacting with public officials
was what was said after I pretended I was an expert witness and was supposed to be
presenting purely the results of an economic model. A retired judge who was listening to
this looked at the people who had hired me and said, " Where did you find this person?"
That has often been the response I get."
Clay Ogg (U.S. EPA, National Center for Environmental Economics)
Dr. Ogg identified himself as "the Project Officer on the other project that you mentioned
where they're looking at the production costs . . . and we did ask them if this cost
analysis is directly linked to the lakes that you're looking at, and I think the answer was
"No." . . . They did look at one lake though, and for the first lake that they analyzed I
think there was a report that indicated you could actually take all the land out of
agriculture and that the benefits would be sufficient to pay for that. But, if you're talking
about making Iowa lakes look like Okoboji, I think you are talking about something
fairly drastic there in terms of taking land out of agriculture. So, it might be useful at
least to look at the size of the watershed you're talking about."
Joseph Herriges
Dr. Herriges admitted to not knowing exactly what project Dr. Ogg was speaking about,
but said, "They're working on a number of projects, and some of them are very much at a
smaller watershed level. The project I'm talking about actually is with the Iowa DNR
(Department of Natural Resources), and I think it's a different project than the one you're
referring to." He added, "I'm not on the project, so I can't tell you exactly what they're
doing, but my understanding is that they're trying to give the state some information
about the cost side of achieving some improvements in water quality. I don't know how
broad it is, but that's the kind of thing you have to look at. We're trying to model the
benefits, but if we're trying to achieve some of these improvements in water quality, you
need to understand the costs and how that works."
END OF SESSION III (Part 1) Q&A
76
-------
Valuation of Ecological Benefits: Improving the Science
Behind Policy Decisions
PROCEEDINGS OF
SESSION III, PART 2: KEEPING WATER FRESH:
THE VALUE OF IMPROVED FRESH WATER QUALITY
A WORKSHOP SPONSORED BY THE U.S. ENVIRONMENTAL PROTECTION
AGENCY'S NATIONAL CENTER FOR ENVIRONMENTAL ECONOMICS (NCEE)
AND NATIONAL CENTER FOR ENVIRONMENTAL RESEARCH (NCER)
October 26-27, 2004
Wyndham Washington Hotel
Washington, DC
Prepared by Alpha-Gamma Technologies, Inc.
4700 Falls of Neuse Road, Suite 350, Raleigh, NC 27609
-------
ACKNOWLEDGEMENTS
This report has been prepared by Alpha-Gamma Technologies, Inc. with funding from
the National Center for Environmental Economics (NCEE). Alpha-Gamma wishes to
thank NCEE's Cynthia Morgan and the Project Officer, Ronald Wiley, for their guidance
and assistance throughout this project.
DISCLAIMER
These proceedings are being distributed in the interest of increasing public understanding
and knowledge of the issues discussed at the workshop and have been prepared
independently of the workshop. Although the proceedings have been funded in part by
the United States Environmental Protection Agency under Contract No. 68-W-01-055 to
Alpha-Gamma Technologies, Inc., the contents of this document may not necessarily
reflect the views of the Agency and no official endorsement should be inferred.
11
-------
TABLE OF CONTENTS
Session III, Part 2: Keeping Water Fresh: The Value of Improved Fresh Water
Quality
Valuation of Natural Resource Improvements in the Adirondacks
Spencer Banzhaf, Dallas Burtraw, David Evans, and Alan Krupnick,
Resources for the Future 1
The Value of Regional Water Quality Improvements
W. Kip Viscusi, Harvard University; Joel Huber, and Jason Bell, Duke
University 52
A Consistent Framework for Valuation of Wetland Ecosystem Services
Using Discrete Choice Methods
J. Walter Milon, David Scrogin, and John F. Weishampel, University of
Central Florida 84
Discussant
Kevin J. Boyle, University of Maine 92
Summary of Q&A Discussion Following Session III, Part 2 96
in
-------
Valuation of Natural Resource Improvements in the Adirondacks
Spencer Banzhaf, Dallas Burtraw, David Evans, and Alan Krupnick*
Resources for the Future
October 2004
* Corresponding author. Resources for the Future, 1616 P Street NW, Washington, DC, 20036;
krupni ck@rff. org.
-------
Valuation of Natural Resource Improvements in the Adirondacks
Spencer Banzhaf, Dallas Burtraw, David Evans, and Alan Krupnick
Resources for the Future
Abstract
For 20 years acid rain has been a central issue in the debate about clean air regulation, especially
in New York State's Adirondack Park. Based on a contingent valuation survey of a random
sample of New York residents, our study quantifies for the first time the total economic value of
expected ecological improvements in the Park from likely policies. Our preferred estimates of
the mean willingness to pay using the base case characterization of ecological improvements
range from $48 to $107 per year per household in New York State. The alternative scope case
yields mean WTP ranging from $54 to $159. Multiplying these population-weighted estimates
by the approximate number of households in the state yields benefits ranging from about $336
million to $1.1 billion per year. The instrument passes an external scope test, a test of sensitivity
to bid, and a test of sample selection.
Key words: Adirondacks Park, air pollution, contingent valuation, ecological values, New York,
non-market valuation, scope test.
2
-------
Valuation of Natural Resource Improvements in the Adirondacks
1. Introduction
For 20 years acid rain has been a central issue in the debate about clean air regulation and
the controversy has centered on the Adirondack Park, which covers some six million acres in
New York State. The park was prominent when Congress created the 1980 National Acid
Precipitation Assessment Program (NAPAP), which coordinated the expenditure of roughly
$500 million to study the effect of acid precipitation on the Adirondacks' ecosystem and other
natural resources in the United States. The 1990 Clean Air Act amendments, an important
legislative milestone in the protection of air quality, dedicated a separate title to the reduction of
acid rain that initiated the well-known sulfur dioxide (SO2) emission allowance-trading program.
More recently, the Environmental Protection Agency (EPA) has cited the reduction in acid
precipitation as a benefit of further reductions in SO2 and nitrogen oxides (NOx) in its support of
the Bush administration's Clear Skies legislative initiative and its regulatory alternative, the
Clean Air Interstate Rule. New York State justifies its own regulatory policies and lawsuits
against utilities by emphasizing the benefits of reduced acid deposition in the Adirondacks.
Until now, all of these abatement initiatives have taken place in the absence of economic
estimates of the total benefits that would result from improvements to the park's ecosystem.1 In
part, this mismatch is explained by the large health benefits that independently justify most
policies that reduce acid rain precursors as in U.S. EPA 1999. But it has resulted primarily from
an inadequate link between the ecological science and social science necessary to enable
1 The NAPAP research effort did include a partial assessment of benefits, including an estimate
of $4-15 million annual recreational fishing benefits in the Adirondacks, from a 50 percent
reduction in acid deposition (NAPAP 1991). No study has ever attempted to estimate the total
value of improvements in the Adirondacks.
3
-------
economic valuation of the benefits of emission reductions. This mismatch has also resulted from
a lack of information on the ecological effects of changes in emissions and deposition to support
that linkage.
Accordingly, while analyzing the environmental pathways linking changes in emissions
to economic benefits, Burtraw et al. (1998) identified the quantification of nonuse values as a
key gap in the literature and thus a priority area for future research. Indeed, the need for
improved estimates of nonuse benefits from ecosystem protection has arisen in many policy
contexts. Consequently, the EPA and other agencies are placing increased emphasis on gathering
this information, as seen for example in the recent formation of the EPA Science Advisory Board
Committee on Valuing the Protection of Ecological Systems and Services.
This study seeks to fill this gap within the important context of air pollution policies by
estimating the change in the total economic value (the sum of use and nonuse value) to New
York State residents that would result from an improvement in the Adirondack Park ecosystem
through further reductions in air pollution. Because stated preference is the only method capable
of estimating nonuse values and because our research application focused on a total value rather
than a value function of attributes, we employed a contingent valuation survey. The survey was
administered both on the Internet and via mail, providing a comparison of mode of
administration and an indirect test of convergent validity. While these different modes have their
pros and cons, the key survey results are remarkably consistent across modes.
This survey was designed to meet or exceed the stringent protocols for stated preference
surveys developed by the NOAA Panel on Contingent Valuation (1993) and the OMB (2003).
One of these protocols stresses that the "commodity" being valued map closely to the underlying
science. Following this guideline, we interviewed a number of top experts on ecological damages
4
-------
in the park and developed a summary of the science report (Cook et al, 2002).2 The report serves
as the foundation for the description of the park's condition as well as the commodity being
valued, that is, the type and magnitude of improvements reasonably following further reductions
of acid deposition precursors.
A major effort of our research was to accurately but meaningfully distill this information
and convey it to a general audience. To this end, during development of the survey we convened
31 focus groups and conducted two major pretests to develop and extensively assess alternative
text, debriefing questions, and graphics.
Our scientific review indicated that there remains much uncertainty about the future
status of the park in the absence of intervention and about the benefits of intervention.
Nonetheless, focus group results clearly indicated that credibility of the survey depended on
respondents believing that scientists understand the problem and how to fix it. Consequently we
developed two versions of the survey to span the range of opinion about the status of the park.
We use the terms base case to refer to the survey that describes a constant baseline (in the
absence of a policy intervention) paired with small ecosystem improvements (in the presence of
an intervention) and scope case to refer to a gradually worsening baseline paired with larger
ecosystem improvements. This design choice has the added advantage of permitting an external
scope test of preferences, a key test of contingent valuation performance highlighted by the
NOAA Panel. We find strong evidence that our instrument is in fact sensitive to scope.
A common criticism of contingent valuation is that the hypothetical nature of the exercise
tends to yield overestimates of willingness to pay (WTP). In response, we typically followed a
cautious or conservative approach when faced with questions of appropriate survey design by
2 A draft of this report was peer-reviewed by field scientists, advocates, and staff at the New
York Department of Environmental Conservation (NYDEC).
5
-------
characterizing the science, presenting information, and applying statistical methods in ways that
are expected to yield estimates likely to understate rather than overstate the true WTP for the
improvements described.
Our preferred estimates of the mean WTP using the base case characterization of
ecological improvements range from $48 to $107 per year per household in New York State. The
alternative scope case scenario yields mean WTP ranging from $54 to $159 per year per
household. Multiplying these population-weighted estimates by the approximate number of
households in New York State yields benefits ranging from about $336 million to $1.1 billion
per year.
The results of this study help complete the two-decade-long project of integrated
assessment across natural and social sciences, resulting in economic estimates that can be used to
guide policymaking to address the ecological effects of acid rain in North America. The above
values exceed cost estimates of reducing SO2 and NOx emissions from power plants subject to
the Clear Skies initiative if the cost share is determined according to the share of these emissions
actually being deposited in the park.
2. From Science to Survey
Comprising both public and private lands, the Adirondack Park covers 20 percent of New
York State, encompassing nearly three times the area of Yellowstone National Park. One-sixth of
the park is designated as wilderness—85 percent of all wilderness area in the northeastern United
States. The park has 2,769 lakes larger than 0.25 hectares, six major river basins, and the largest
assemblage of old growth forests east of the Mississippi River. Thirty tree species, along with
numerous wildflowers and a multitude of shrubs, herbs, and grasses, are native to the park. These
attributes draw nine million visitors each year.
6
-------
The Adirondack^' watersheds are particularly sensitive to potential acidification from
atmospheric deposition of sulfates and nitrates, in part because they tend to have shallow soils
and bedrock with low acid-neutralizing capacity. However, as is said in the survey, "[m]ost of
the lakes affected by past air pollution are small; they are typically much smaller than Central
Park in New York City. The large lakes that you may have heard of (such as Saranac Lake or
Lake George) are much bigger than Central Park and are not lakes of concern."
Table 1 shows some of the conclusions reached in our analysis of the scientific research
and how they translated into descriptions in the survey. Currently, a small fraction of the lakes in
the park are acidic due to natural causes (roughly 10%), but most degradation is a result of
acidification linked to emissions from power plants and other sources. About half of the lakes are
degraded in quality, some of these without fish populations. The actual cause of declining
populations of fish is often increased aluminum concentrations, a by-product of the process of
acidification.
The future baseline for the park's ecosystem depends largely on nitrogen saturation. If a
watershed becomes nitrogen saturated, then increased nitrogen deposition will lead to greater
chronic acidification of the receiving water body. Significant reductions in SO2 and NOx
emissions resulting from the 1990 Clean Air Act Amendments (CAAA) have led to some
recovery of acid-neutralizing capacity and surface water pH in the Adirondacks, but not in
proportion to the drop in emissions (Driscoll et al., 2001a; 2001b; 2003; Stoddard et al., 1999).
Estimates of the time scale for reaching saturation vary considerably from watershed to
watershed. Some may never become saturated at current or forecasted deposition levels; others
may and would thus require further reductions in deposition for recovery.
7
-------
This variability and underlying uncertainty implies a range for the future baseline of
chronically acidic lakes (assuming constant future deposition) from great degradation to a
modest improvement. Assuming full implementation of the 1990 CAAA and no further emission
reductions, the share of lakes that are chronically acidic could rise from 19 percent in 1984 to 43
percent or more by 2040 with saturation at 50 years or fall to 11 percent or less by 2040 if
saturation is never reached (EPA, 1995). Our response to this information was to develop base
case and scope case alternatives.
We found widespread scientific consensus that acidification also has harmed forests
(Driscoll et al., 2001a; 2001b; Lawrence, 2001). In particular, because acid deposition has been
implicated in declines of high-elevation spruce stands, in the base case scenario respondents are
told that the improvement program would yield small benefits to these stands. Moreover, there is
mounting but as of yet not definitive evidence that damage to sugar maple and white ash stands
also can be caused or exacerbated by acidification.
In the scope case scenario the described damage to the spruce stands is greater, damage to
sugar maple and white ash is described, and it is stated that the stands are expected to decline in
the future. Improvements from the current and future state of the forests also are more significant
in the scope version of the survey.
There is also mounting evidence that acidification is affecting some bird populations. In
the base case scenario, acidification is implicated in reduced, but stable, populations of the
common loon and hooded merganser. The improvements to these species as a result of the policy
intervention are characterized as minor in the base case scenario. In the scope case ecosystem
acidification also is implicated in loss of nesting places and changes to songbird populations of
8
-------
wood thrush and tree swallow in the park. In the scope case, all four species are expected to
gradually worsen without the policy intervention.
3. Description of Survey Instrument
To develop an estimate of societal WTP to avoid the effects of acidification, we
employed a contingent valuation (CV) survey, an approach that has been used since the early
1960s (Davis, 1963) to determine both use and nonuse values and has been extensively examined
(Mitchell and Carson, 1989; Haab and McConnell, 2002) in a wide variety of applications. Of
the thousands of CV instruments administered to date, there is generally a handful of studies that
are considered models. One relatively famous example is the application of the CV technique to
estimate damages from the Exxon Valdez oil spill in Prince William Sound in 1989 (Carson et
al., 2003). A later, widely known, and thorough application by the same team of researchers
estimated damages from the Montrose Corporation's release of DDT and PCBs off the coast of
Los Angeles (Carson et al., 1994). These studies served as models for the organization and
treatment of information in our study provided below.^ This information is followed by
treatment of several thematic issues, which arose from our objective of developing a cautious,
valid WTP estimate grounded in science and useful for policy.
Context
The introductory section of the survey is designed to place the proposal into a broad
context of household and public decisionmaking and address the embedding problem, which is a
tendency of respondents to expand the commodity definition to include many other things than
3 A burgeoning literature on valuing ecosystems is increasingly able to inform policy but it is
rarely capable of providing estimates of specific value that can be used in benefit-cost analysis
(for example, Nunes et al., 2003; Simpson et al., 1996). In a limited application Morey and
Rossman (2003) use stated preferenece methods to measure the value of delaying damage to
cultural materials from acid deposition.
9
-------
those intended to be valued (Kahneman and Knetsch, 1992). Respondents are helped to think
about substitutes to the proposal without explicitly asking them to choose among different goods.
The opening is austere, with the title "Policy Priorities Study: Adirondacks Version," giving
respondents the impression that there are many different versions of the survey addressing
different issues and public policy priorities. Respondents are asked if they felt their income taxes
are too high or low.
To encourage consideration of public goods trade-offs, subjects are asked to specify
whether more or less state spending in various areas (such as crime prevention or providing and
maintaining natural areas) is called for. They are explicitly reminded that spending increases or
decreases may result in higher or lower taxes. Respondents are then told that their version of the
survey deals with a tax-and-spending program to improve the health of lakes in the Adirondack
Park, while other versions focus on such diverse topics as infant health care and fire protection.
Baseline
Subjects are next introduced to the Adirondack Park and educated about damages to the
ecosystems of the park's lakes with specific attention paid to their altered fish populations. We
call the affected lakes the "lakes of concern," a sterile term intended to discourage overly dire
interpretations of their status.4 We state that about half (1,500 lakes of approximately 3,000
total) are lakes of concern. We emphasize that these lakes are generally smaller and less well
known than the large lakes, such as Saranac Lake or Lake George, that attract most of the park's
visitors. In the base case, the condition of forests and bird populations is also characterized. In
4 Initially we defined lakes as "healthy," "sick," or "dead," and found in focus groups that many
subjects had graphic images of "dead" lakes and "sick" lakes and thought that a "dead" lake
could not be recovered. We found that using the term "lakes of concern" did not create such a
vivid mental image and allowed a more dispassionate description of the commodity. Similarly,
we used sterile black-and-white pictures to introduce the affected animals.
10
-------
the scope case more forest and more bird species are characterized as damaged. Subjects learn
that the cause of these problems is acid deposition, acting directly and through aluminum
leaching from the soil.
Respondents learn that acid deposition has slowed dramatically thanks to programs to
reduce air pollution and that, in the base case, acid deposition is not expected to harm any
additional lakes in the future, but nor will the lakes improve on their own. As seen in many polls
(Bowman, 2004), in general people believe the environment is worsening over time. That view
applied to the Adirondack Park would be erroneous, based on our understanding of the science.
We appealed to the authority of scientists studying the lakes and the Environmental Protection
Agency (as our focus groups indicate great trust in these groups) to refute this preconception. In
the scope case we say that the lakes, forests, and bird populations will worsen slowly without
intervention.
A potentially troublesome concern in creating the survey was that the respondents would
associate human health damages with damage to the lakes. There are no direct human health
hazards from contact with the affected lakes. To address this issue respondents are told that the
acidity of the lakes is no more than that of orange juice, that they are safe for swimming, and that
there are no health effects from eating affected fish. They are also told that there is no
commercial market for these fish.
Scenario
A scenario for the improvement must be plausible to respondents but, as seen in the
Montrose and Exxon Valdez surveys and many others, need not be a real scenario currently acted
upon or even under consideration. What is important is that the improvement approach is
credible, is a public good requiring payment by individuals and not so expensive or cheap to
11
-------
make cost an issue. A perfect scenario is transparent and uses a payment vehicle that avoids any
bias in WTP responses. Telling the truth—that imposing reductions on power plants and other
sources of air pollution is the best way to fix the problem—could very well lead to biased
responses, which is what we found in initial focus group settings.
Our solution was to develop a fictional program that "scientists determined to be the
safest and most practical" for improving the Adirondacks ecosystem, involving the application of
a Norwegian technology to lime lakes (each year for ten years) and, in the scope case, forests by
airplane. In fact, liming of lakes to reduce acidity on an individualized basis does occur, but it
remains controversial and, to our knowledge, an application of the scale described in the survey
has never been recommended. However, liming constitutes an active, public program that would
require the collection of additional taxes—and, hence, the opportunity to elicit WTP.
The ten-year improvement period is probably in reality too short a time for the ecological
improvement from reducing acidification precursors to be fully realized. We choose a ten-year
horizon for benefits for two reasons. Practically, focus group participants equated long time
frames with uncertainty of outcome, which reduced the perceived effectiveness of the
intervention and thus biased WTP downward. Furthermore, emission reductions under Title IV
of the Clean Air Act Amendments have shown a change from trend in the Adirondacks lakes in
less than ten years since the program took force in 1995, so that important improvements could
in fact be expected in this time frame (Driscoll et al., 2003).
Focus group responses pointed to distrust of the ability of New York State government to
implement the program as described, and there was concern that the government would use the
taxes raised for the program for other purposes. Consequently, we invoked "an independent
Adirondacks Management Board of scientists, a representative from the U.S. Environmental
12
-------
Protection Agency, and other experts" that would oversee the program. In focus group testing,
this board appeared to deflect many of the concerns about management credibility.
In response to concerns that anglers will reap benefits and should pay their share of the
costs of improvements, we said that, where necessary, the fish will be restocked using revenue
from fish license fees. To fill out the scope case, we said that a tree-planting program would
supplement the liming of the forest.
Commodity
The effects of this program, and the commodity to be valued, vary for the base and scope
cases. In the base case, the improvement^ is to 600 lakes of concern (out of 1,500), which will
take place over a ten-year period, after which the lakes will be stocked with fish. Small
improvements in the populations of two bird species and one tree species will also occur, limited
to areas surrounding the affected lakes. In the scope case, improvement is to 900 lakes, plus two
additional bird species and two additional tree species. The status of the lakes with and without
the intervention is summarized in a pie graph and recapped in a summary table along with the
baseline and changes to tree and bird populations. For the scope version, improvements to the
forests are displayed using a pictograph with each square on a grid representing some number of
trees of various species, and their health, as a portion of total forests in the park.
Payment Vehicle
5 In early focus groups, we described the resource as being "restored," but found considerable
evidence of loss aversion in voting decisions as many people felt that ethics demanded we "clean
up our messes." We believed that, though such ethical perspectives are an important element of
the policy debate, the issue is independent from a measure of the benefits from the particular
resource. A cautious approach to valuing benefits required that we divorce stated WTP for the
particular improvements from the general desire to rectify past harm. As a solution, we turned to
"improvement" over "restored."
13
-------
Respondents (speaking for their households) are then presented with an opportunity to
pay increased taxes annually for ten years, if the majority of voters agree. To strengthen the
certainty of the government's commitment, the funding instrument for the program is a revenue
bond that must be paid off by the increased tax revenue. Prior to voting, respondents are
presented a balanced list of three reasons they may want to vote for or against the program. They
are also presented with "cheap talk" language that warns the respondent of a tendency by people
to answer survey questions about WTP in a different way than they would behave in actual
decisions and to try to consider their choice as though it was an actual decision.
Eliciting WTP
Finally, we elicit a vote in referendum format for or against the program, plus a single
follow-up vote in referendum format, motivated by the idea that engineering costs are uncertain.
Based on the results of two pretests, we targeted initial annual payment (bid) levels at
approximately the median and the 30th and 70th percentiles of the WTP distribution for the base
case improvements. We also sought information in the right-hand tail given that estimates of
mean WTP can be particularly sensitive to distributional assumptions in that region. Initial bids
were set at $25, $90, $150, and $250. Follow-up bids, conditional on a "no" or "yes" response on
the initial bid in the first vote were set at ($10, $50), ($50, $150), ($90, $250), and ($150,
$3 50).6 The first number in the follow-up bid is if they voted "no" initially and the second is if
they voted "yes."
Debriefing
After they voted for the program, we asked participants several debriefing questions. The
primary purposes of these questions were: (i) to solicit respondents' beliefs about the information
6 In addition, one of the pretests, used in the final data analysis, had initial bids set at $35, $85,
$150, and $200.
14
-------
and improvement scenario they were provided; (ii) to give them some limited opportunities to
revote when their beliefs were at odds with the survey's intent (if they believed there were health
effects,^ if they voted "no" only because New York State was responsible for implementing the
improvement plan or if they voted "no" because they believed upwind electric utilities should
pay); and (iii) to examine their more general attitudes and beliefs that might lead them to provide
"nay-saying," or "yea-saying" responses (see below). We also asked demographic questions in
this section, including one rarely asked about respondent's future family income. This question
was asked because the payment was to be over a ten-year period. This variable turned out to be
more significant than current income in explaining WTP.
After the demographic questions, at the end of the survey we inform respondents that the
liming program is not being considered by the New York State government and is not feasible.
Respondents are also told that these improvements would actually occur through further
reductions in pollution and who the sponsors of the survey were.
Expansive Priors
One may reasonably ask why we bother to introduce the effects of the intervention on
forests and birds in the base case if these endpoints do not improve significantly as a result of the
intervention. Initially our approach was to simply limit the description of the damage to the
aquatic ecosystem in the base case. However, we discovered in focus groups that omitting
mention of forests and birds in the base case was inconsistent with respondents' prior beliefs.
Because it was judged so unlikely that forests and birds were neither being currently damaged
nor would be helped by an improvement plan, respondents substituted their own expansive
7 52 percent accepted that there were no human health effects, 38 percent thought that there may
be minor health effects, and 10 percent thought there were important health effects. Of those who
had voted for the program and thought there were health effects, about 12 percent changed their
vote to "no" when asked to suppose there were no health effects.
15
-------
priors, ascribing much broader and larger effects to our improvement plan than we intended or
that the science can substantiate. There is some evidence that this substitution had the effect of
actually making their WTP higher for the base case than for the scope case. Accordingly, we
validated respondent priors by both narrowly identifying effects on forests and birds and
describing their improvements as minor. In focus groups we found this change made the
information treatment more credible, so that respondents suspended their priors and accepted our
characterization. 8 A similar challenge was to make credible and certain the characterization of a
constant future baseline and limited health effects, as discussed above.
Yea-saying and Warm Glow
One potential concern with contingent valuation is a presumed tendency of respondents
to vote "yes" for programs in a pro forma way, perhaps out of a sense of obligation or desire to
please the survey administrator, but in any case without truly registering the economic trade-offs
involved and hence without truly revealing preferences. A special case is "warm glow," in which
respondents value the giving per se as much as the commodity acquired (Andreoni, 1990).
Including warm glow would overstate values for the actual commodity, in this case the
Adirondacks.
As noted previously, the introductory pages of the survey are designed to make
respondents immediately think about the opportunity cost of paying for the program, and before
voting they were reminded of costs and other reasons to vote "no." In addition, we took pains to
use line drawings and other design features to minimize embedding and to avoid emotional
triggers. Also, we asked a series of debriefing questions that could be used to identify this type of
vote. In particular, we asked respondents if they agreed that "costs should be a factor when
8 Moreover, this nuance in fact made the scenario more consistent with the science than the
simplistic no-terrestrial-effects description.
16
-------
protecting the environment." Fully 75 percent of the respondents agreed that costs should be a
factor, suggesting they believe in the trade-offs inherent in a willingness-to-pay exercise.
Moreover, of the others, one-fifth exhibited implicit acceptance of the maxim when they
switched their vote to "no" in the follow-up valuation question when the bid was changed.
Nay-saying
In contrast to yea-saying, nay-saying is a tendency for respondents to vote against a
program for reasons that are extraneous to its benefits and costs. This includes respondents who
reject the scenario or choice construct as presented or who use their vote to register some other
protest. For example, some people vote against the program on the principle of limiting taxes or
because they do not trust the New York State government to implement the program or because
they don't think the program will work. Although our cautious approach made us more tolerant
of nay-saying than yea-saying, we nevertheless designed the survey to limit and identify this
phenomenon. About 79 percent of the sample agreed in principle that there are programs that
could justify new taxes. But as with our debrief targeting warm glow, actions speak louder than
words here, with almost half of the remaining 21 percent voting for the program and its tax
increases at some bid level. Moreover, as discussed below we asked several debriefing questions
on beliefs about the baseline and the feasibility of our program, with most respondents accepting
the scenario.
4. Survey Protocol
The survey was administered by Knowledge Networks (KN) from August 2003 through
February 2004 to residents of New York State. We selected this population for several reasons.
First, New Yorkers are most likely to hold nonzero values for improvements to the Adirondacks.
Second, designing an acceptable method of payment was easier if the sample were limited to
17
-------
New York. Third, by ignoring people out of state we were being cautious in our total benefit
estimates.
Table 2 summarizes the total sample, useful completions, and response rates for the
different modes of survey administration. Results from a second pretest were included in the data
analysis.^ Response rates to KN's preselected panel were, as expected, quite high, ranging from
84 percent for the pretest to 74 percent for the final implementation. The group comprises 53
percent of our total completed surveys.
To boost the sample size provided by Knowledge Networks and to examine the potential
sample selection caused by attrition in the KN panel, the survey was given to a group that had
withdrawn from the panel. This version was administered over the Internet and, with the
exception of some demographic debriefing questions, was the same as the version given to the
panel. The response rate for this group was 14 percent and totals 16.8 percent of our completed
surveys. Although this response rate seems low, it is not surprising from a group of subjects who
had already declined participation in one venue.
As a formal test of mode of administration, as an additional check on the KN panel, and
to further boost sample size, a final wave was mailed using a random-digit selection of telephone
9 An initial pretest was omitted from the final analysis, as it was too different from the final
instrument.
18
-------
numbers that were in turn matched to available addresses. The response rate for the mail survey
was 24 percent, and the group constitutes 31.3 percent of our completed surveys. 10
Table 3 presents descriptive statistics of the demographics of each sample. It illustrates
the difference among the samples and, where possible, compares them to the general population
of New York State. While there are some differences across the samples (for example, the mail
sample had the oldest average age, while the withdrawn sample had the youngest), in general
they display fairly consistent attributes. On each measure, the samples are proximate to the
characteristics of the general adult population in New York State.
5. Results
With the NOAA Panel protocols and OMB guidelines putting the burden of proof
squarely onto the researchers to show that their results are valid, we start with showing the
validity of our results before actually summarizing what they are.
Measures of Validity
We present three basic measures of validity: the external scope test, sensitivity of vote to
bid, and construct validity, that is, the extent to which patterns in the data reflect common sense
and expectations based on economic theory.
The external scope test examines whether two separate samples have different average WTP for
differing scales of environmental improvements (Boyle et al., 1994). It is a test both of the
10 Techniques used to induce response from respondents varied amongst the samples. Members
of KN's panel received compensation equivalent to about $10 in Internet service in exchange for
completing the survey while withdrawn and mail respondents received $10. In addition to these
incentive payments, subjects received reminders to complete their surveys. Members of the panel
received reminder e-mails encouraging completion of the survey. Members of the withdrawn and
mail samples received follow-up phone calls and reminder letters. For the mail sample up to five
attempts at person-to-person calls were made to contact the potential respondent to directly
request that they take the survey.
19
-------
subjects' comprehension of and attention to the scenario and vote, as well as warm glow and
embedding, or what Mitchell and Carson (1989) call "part-whole bias." The scope test has been a
major standard for contingent valuation since the NOAA Panel report.
A fundamental issue in designing a scope test is determining which dimensions of the
resource or service to expand. For example, Boyle et al. (1994) failed to find sensitivity to the
scope of a program to save migratory waterfowl from oiling themselves in dirty ponds. The
scope was measured as a variation in the number of birds (in three different versions, 2,000,
20,000, or 200,000 birds would be saved respectively). Some have criticized this scope test on
the grounds that the commodity is mistakenly defined: people might care more about the
availability of the clean ponds themselves than the birds or perhaps measured birds in flocks
rather than individuals or, again, in percentage terms rather than numbers. On the other hand,
Carson et al. (1994) passed a test of scope when comparing a project that would improve the
health of two fish species in Los Angeles Harbor to a project that would in addition improve the
health of bald eagles and peregrine falcons. This approach to a scope test is in contrast open to
the criticism that the scope of a commodity has not been measured at all, but rather an entirely
new commodity that is more greatly valued than fish.
Our approach to the scope test attempted a compromise between the narrow more-
individuals and the broad more-commodities approaches. We defined the resource to be scaled as
the health of the Adirondack Park as a system of lakes, forests, and animals. Specifically, we
varied the quantity of lakes improved (analogous to Boyle et al.) and also varied the number (and
quantity, in percentage terms) of tree and bird species improved. Our results provide strong
evidence of sensitivity to scope. Table 4 reports the share of "yes" votes at each bid level for the
base and scope versions.
20
-------
Several approaches can be used to test for scope sensitivity using these data. The most
nonparametric and perhaps most persuasive is to test for differences in the mean share voting for
the program at each bid level. P values for this chi-square test are provided in the final column of
the table. As seen in the table, more respondents vote for the program under the scope scenario at
each bid level, and the difference is statistically significant. Respondents are thus willing to pay
more when they understand there will be greater improvements. In addition, estimates of mean
WTP are higher for the scope version under a variety of model specifications (see the section on
willingness to pay estimates below), and these differences are also statistically significant.
Finally, other results corroborate the interpretation that respondents were paying careful
attention to the description of the resource. For example, when we asked whether respondents
accepted our description of the baseline state of the Adirondack Park, in the base survey
instrument 24 percent of the sample said that it was probably worse than we described it,
compared to only 6 percent with the scope instrument, a statistically significant difference.
Similarly, 15 percent of the sample thought that the survey was biased in favor of the program
with the base instrument, but 27 percent thought so with the scope instrument. The relatively low
numbers here overall are also evidence of content validity.
The second important statistical test is sensitivity to the level of the bid, that is, whether
fewer respondents vote "yes" when the bid level is increased within each given scenario. In fact
we find that responses are strongly statistically significant for both the base and scope versions
(with the exception of those cells with few observations, accounted for by the pretest). Even
including these cells, the difference is statistically significant according to a chi-squared test of
the equality of means. Moreover, according to Kendall's tau test these differences are
statistically ranked monotonically by bid, showing a consistent increase.
21
-------
Figure 1 concisely illustrates the sensitivity to scope and bid, omitting the sparse cells
from the pretest. Sensitivity to scope is indicated in each bid category by the higher percentage
who voted "yes" in the scope scenario than the base scenario. Sensitivity to bid is indicated by
the decline in the share of respondents willing to vote for the program as the initial bid level is
increased, for both the base and scope scenarios.
The third set of construct validity tests verifies that the other patterns in the data conform
to theory and common sense. We find that they generally do. Table 5 provides a representative
regression output covering three types of variables: demographic, attitudinal, and the degree to
which respondents accept the concepts in the survey and other information provided to them
(protests or indications of yea- or nay-saying). Model 1 contains only demographical and
attitudinal variables, model 3 contains only the protest variables, and model 2 contains both.
Some of these variables may be considered endogenous, an issue we return to below in the
discussion of willingness to pay.
Models 1 and 2 in the table show households with the highest incomes have the highest
WTP, as expected. The poorest households are also more likely to vote for the proposal,
presumably because they do not expect to have to pay for it, but the effect is not significant.
Consistent with the permanent income hypothesis and with the fact that payment would occur
over a ten-year period, those who expected their future income to be higher are willing to pay
more than those who thought otherwise. Household size is also a consistently significant factor,
with larger households less likely to vote for the program, although the effect is not significant.
On the other hand, holding household size constant, households with more children are
22
-------
significantly more likely to vote for the program. 11 Other standard demographic variables (age,
race, sex) are unsigned as hypotheses and were not considered in our analysis.
Measures of personal stake in the resource are also important. Households that frequently
visit the park (more than ten times a year) are willing to pay more for the program than others
who visit less frequently. In addition, those living farther from the park are willing to pay less,
with WTP falling by about $0.08 per kilometer from the household's closest entrance (by road)
to the park and with an elasticity of WTP to distance of about 0.4 when controlling for indicators
of protests (model 2). 12 This information is important for this study because of the inferences
one might make about WTP of households outside of New York. The finding is consistent with
previous work (Johnson et al., 2001).
Regarding the effect of attitudes on voting, self-classified environmentalists are more
likely to vote for the program, just as self-proclaimed conservatives and those who think taxes
are too high are more likely to vote against. We also asked people in the beginning of the survey
if they are interested in government spending more on nature and wildlife programs and on air
and water pollution control programs, among other things. Those who favored more government
spending on the environmental programs are more likely to vote for the program. In alternative
models, we replaced these variables with indicators for those who describe themselves as
"liberal" or "conservative," and find that the former are more likely to vote for the program
while the latter are less likely to do so.
Willingness to Pay
11 The model includes an indicator variable for the presence of children, plus a linear term for the
number of children. The former is negative, but the latter is positive and offsets the former at two
children.
12 After conditioning on distance, those living within the park's boundaries do not appear to pay
more than other households.
23
-------
We designed our strategy for estimating willingness to pay to limit three potential sources
of bias: the representativeness of the sample, anchoring in the follow-up vote on the program,
and yea-saying or nay-saying votes.
The first potential source of bias is the possibility of an unrepresentative sample,
especially for the KN panel of regular survey takers. To address this potential problem, first we
weighted all responses by all observable demographics, including location of residence, to reflect
the New York State population. To address unobservable factors, we included a random mail-
based sample of the entire New York population as a check on the KN panel. After weighting the
data to account for the differing demographics of the sample (see Table 3), we could not reject
the hypothesis of equal WTP from the differing survey modes.
Furthermore, one of the advantages of the KN panel is that Knowledge Networks elicited
initial background demographic and attitudinal questions for all its panel members. Thus, we
have individual-level details about the nonrespondents. This information provides a unique
opportunity to estimate sample-selection models against both those currently on the panel, but
not completing our survey, and those who have dropped out of the panel over time. We estimate
a Heckman sample selection model with a joint normal distribution between the unobserved
component of responding to our survey (among all those ever on the KN panel) and the
unobserved component of voting for the program. With this model, we cannot
reject the hypothesis that the correlation is zero, again implying no differences among the
samples. 13
The second potential source of bias is the use of the follow-up dichotomous choice
question, giving double-bounded rather than single-bounded data. Using double-bounded data
13 The test is in the context of a model implying that WTP is lognormally distributed, one of the
econometric models presented below.
24
-------
provides gains in efficiency (Hanemann, Loomis, and Kanninen, 1991), but may induce bias if
the WTP distributions differ across the two equations, for example because the new price in the
follow-up question sends a signal about the program quality or suggests that a strategic game
may be being played (see Haab and McConnell, 2002, for discussion). Estimating willingness to
pay with a lognormal distribution and restricting a completely general binary probit model to be
consistent with a single distribution (Cameron and Quiggin, 1994), we reject the hypothesis of
identical distributions at the one percent level using a log-ratio test. 14 Although we cannot reject
the hypothesis of constant median WTP, estimated mean WTP is lower using the double-
bounded data, a typical finding. Still, as with others using dichotomous choice data, we prefer to
make use of the additional efficiency afforded by the follow-up question. Moreover, any
potential bias introduced by this approach is downward, which is consistent with our cautious
philosophy.
The third type of potential bias is yea-saying or nay-saying. If these problems came
undetected, they would contaminate the estimates of WTP for the intended commodity with
values for other commodities. As discussed above, we attempted to identify such problems by
probing people's beliefs about the scenario and their willingness, in principle, to make trade-offs
between taxes and public goods. Table 6 summarizes the key probes and divides them into those
tending to bias WTP upward (yea-saying) and downward (nay-saying). It also shows the share of
respondents whose answers raised flags and our response. In some cases, when we identified
14 However, employing a nonparametric test suggested by Haab and McConnell (2002), we find
that, for those bid levels used both in initial votes and in follow-ups ($150 and $250), the
percentage voting "yes" in the second vote, conditional on voting "yes" in the first vote, was
higher than the unconditional percentage voting "yes" in the first vote. This finding is consistent
with the existence of a single distribution, and so constitutes a failure to reject the hypothesis of a
constant values across votes (compared to the alternative hypothesis of falling values in the
follow up). Admittedly, this is a weak test.
25
-------
problems (such as a belief that human health would be improved by the program), we asked
respondents to hypothetically accept our premise and revote. If they did not change their vote, or
in cases where we did not ask them to revote, we then have the opportunity to eliminate the
respondents from the sample or to control for them econometrically. As discussed below, our
results are robust to these differing treatments.
Using the full double-bounded data, we used standard methods for analyzing interval data
(Hanemann, Loomis, and Kanninen, 1991; Haab and McConnell, 1997), and assume that the
responses are distributed according to Weibull and lognormal distributions. These distributions
imply that WTP is always positive. They generally provide similar estimates of the effect of
covariates, but mean WTP is generally larger with the lognormal distribution because of a
thicker right-hand tail.
We estimated population-weighted interval models of the WTP distribution, controlling
for indicators of scenario or task rejection. In estimating WTP, we did not control for
demographic and attitudinal variables, such as those in models 1 and 2 of Table 5, as these
variables have no "right" answer and can simply be integrated over in computing mean WTP as
long as they are properly weighted to reflect the New York population.
In order to control for yea-saying and nay-saying, we either dropped respondents or controlled
for them econometrically. Model 3 in Table 5 presents the regression results for the all-
econometric-control case using the lognormal distribution. Table 7 presents the full array of
mean WTP estimates arising from different combinations of these two approaches (drops and
controls) for the base case survey. Table 8 does the same for the scope case. Each cell contains
estimates of the mean WTP for the lognormal and Weibull models. The columns represent
26
-------
adjustments made for nay-saying controls, while the rows represent adjustments made for yea-
saying controls.
To decide which variables to target for dropping (instead of adjusting econometrically),
we ran a series of regressions to determine which variables had the most important affect on
WTP. For yea-saying, the most significant variables are the attitude that costs should not be a
factor when protecting the environment and the belief that health effects are important; for nay-
saying, the most important variables are the belief that taxes should not be raised under any
circumstances and the belief that the liming program was not practical.
In the various treatments indicated in Tables 7 and 8 these variables are either dropped or
else controlled for econometrically by calculating WTP from the estimated regression
coefficients after redefining the targeted variable's value appropriately (for example, setting the
"thought there were health effects" variable to zero). All other variables of concern listed in
Table 6 were similarly controlled for econometrically.
Looking down the first column of data on Table 7, note how close the estimates are to
one another, ranging from $58 to $80. This implies that the results are remarkably robust to
various attempts to correct for warm glow and other yea-saying effects. The results appear to be
less robust when adjusting solely for the nay-sayers in the first row of the table, but still fairly
robust overall. As expected, the lognormal model produces substantially larger estimates than
that of the Weibull model, although results based on the latter model are fairly similar.
The other rows and columns of this table provide results for various combinations of
controls on the different groups of people in the sample. The diagonal is particularly important,
as it represents a symmetric treatment of yea- and nay-sayers. Across all the cells with some
form of control (either econometric controls or dropping specific variables) the results are quite
27
-------
close to one another, ranging from $156 to $266. This suggests that our estimates are quite robust
to the choice of dropping or controlling econometrically for these responses. However,
comparing the middle of the table with various choices of dropping or controlling
econometrically to the first row and column, it is clear that our results are sensitive to treating
various yea-sayers and nay-sayers in some form versus not at all. We tend to favor models with
more controls, but a case could be made for omitting some controls if it is believed responses to
the debriefs are endogenous with responses to the vote. In other words, after the vote, people
might look for additional reasons to justify their vote when they respond to the debriefing
questions. 15
We consider the symmetric, all-econometric-controls option as our preferred model in
Table 7 since it maintains the largest sample size and symmetrically controls for both yea-sayers
and nay-sayers. The Weibull model gives an estimate of $159, while the lognormal model gives
an estimate of $213. Turning to Table 8 for the scope case, the corresponding best estimates are
$179 and $308 per household per year.
The range of results across the cells of the table, and between the Weibull and lognormal
models, represents model uncertainty in the WTP estimates. The range between the base and
scope estimates represents the scientific uncertainty about the baseline state of the Adirondacks
and the effects of policy interventions. Each of these estimates is further subject to statistical
uncertainty, as indicated in Figure 2. The figure reflects uncertainties in WTP from the
regressions with only econometric controls and shows that statistical uncertainties are quite small
for the Weibull model and considerably larger for the lognormal model. Ninety-five percent
15 One might in particular accept respondents' statement that they would not have changed their
vote if there were no health effects or even if New York State were not involved, obviating any
need to control for those responses. Dropping those controls lowers estimates by about one-
fourth and widens the differences between the base and scope cases.
28
-------
confidence intervals for the latter are two to two-and-one-half times greater than the mean on the
high side, compared to about a 25 percent confidence interval on the Weibull models.
An even more cautious approach, using the most conservative design, is to estimate the
Turnbull lower bound (Carson et al. 1994, Haab and McConnell, 1998). This approach considers
the WTP of each household to be the lower bound of each interval of data. For example, if a
respondent answers "yes" to an initial bid of $25 and "no" to $50, this approach would interpret
$25 to be the actual WTP. In fact it is the lower bound of the $25-to-$50 interval. Besides being
unassailably cautious, this approach has the advantage of avoiding any distributional
assumptions. 16
Using the double-bounded referendum format, the Turnbull lower-bound estimates in the
base case scenario yield an estimate of mean WTP of $53 per household using all the data and
$46 per household dropping those who say costs should not be a factor, who believed there were
significant health effects, or who thought taxes should not be raised for any reason. The Turnbull
lower bound estimates of the mean WTP for the scope case are $155 and $111 per household
respectively.
6. From Survey to Policy
The foregoing results, although they are weighted to represent the population of New
York, lack three elements to make them policy relevant. The first is that they are developed from
a particular temporal phasing of payments and benefits that is unique to the survey. Converting
them to annualized benefits over an infinite time period would make them more generally useful.
Second, they provide total values for improvements at the park. But in some applications it may
16 Note that because it is a nonparametric estimator, the Turnbull lower bound cannot control for
protest attitudes. Thus, respondents must be either maintained in the sample or dropped.
29
-------
be important to have some idea of the use and nonuse value components. Third, as with any
estimate of benefits, the question arises: are these big numbers or small numbers? That question
is answered by comparing the estimates to a cost benchmark.
Discounting
The WTP estimates computed directly from responses as provided above are for
payments over a ten-year period beginning immediately to obtain a stream of benefits that won't
begin in full until the end of that ten years. For use in a benefit-cost analysis, we need to convert
these estimates into an annualized infinite stream. Assuming benefits phase in linearly over ten
years, the equation below provides this conversion:
Using the factor associated with a three percent discount rate, for instance, the $159 best Weibull
estimate with economic controls for all yea-saying and nay-saying variables provided above is
multiplied by 0.3, reducing WTP to $48 per year per household for a benefit phased in over ten
years and continuing indefinitely. As an upper bracket on the range of values for the base case,
we take the lognormal estimate of $213 times 0.5 (the factor associated with the five percent
discount rate), for a WTP of $107. For the scope case we similarly take the estimates of $179 and
$308 from the same cell, times the respective three and five percent adjustment factors, for a
WTP range of$54to$154.
Total Value versus Use and Nonuse Values
9
Annualization Factor
lOr
where 8 = and r is the discount rate.
1 + r
30
-------
People who recreate in the Adirondacks hold use and nonuse values. People who do not
recreate in the Adirondacks hold nonuse values. Thus, it is possible to get some insight into the
subcategories of values by examining WTP for the two groups. To do this, we first regressed
variables for frequency of use and the standard variable list against vote responses. We found
that the only significant distinction was between those whose visit frequency is over ten times in
the previous five years (23 percent of the sample) and those with less frequent visits (or no
visits). Using this variable, we predicted WTP for the two groups and found that the frequent
users had a WTP about 70 percent higher than that of the infrequent and nonusers, implying
relatively large use values.
Are the Benefits Large?
Are our WTP estimates "big" numbers? First, note what is not included in these numbers
that would be relevant to a formal benefit-cost analysis concerning reductions in acid deposition
precursors. They omit benefits to residents of other states, be they users or nonusers of the
Adirondack Park. Our results on the effect of location on WTP suggest that such benefits may be
smaller per household than those enjoyed by New York State residents. They also exclude
benefits to other ecological assets and those to other types of endpoints, most importantly the
health effects related to fine particulate exposure.
Second, these numbers can be compared to a cost benchmark. EPA (2004) has estimated
the costs of its Clear Skies proposal to utilities to be $4.3 billion in 2010, rising to $6.3 billion
per year by 2020.17 Clearly only a fraction of these costs should be attributable to improvements
in the Adirondacks because only a fraction of utility emissions affects that region. Although
there is no universally accepted way to make such allocations, a reasonable approach is to assign
17 Clear Skies ultimately would lead to reductions of 75 percent in SO2 and 65 percent in NOx by
sometime after 2020 when allowance banks are exhausted.
31
-------
cost shares to each utility's in accordance with the fraction of their emissions falling in the
Adirondacks. Using the TAF model (Bloyd et al., 1996) for the source-receptor relationships to
do this and model runs that provide the costs by electricity-producing region, we find that on
average, 2.3 percent of utility SO2 emissions fall on the Adirondacks. Multiplying each region's
share by their costs for implementing Clear Skies gives an estimate of $86 million in 2010 and
$126 million in 2020 for costs attributable to Adirondack improvements. These cost estimates
are significantly less than our estimates of the benefits.
7. Conclusions
This paper has presented the first-ever results for the total value of the ecological
improvements to the Adirondack Park that might be expected from another round of reductions
in air pollution emissions. These estimates matter because damage to the Adirondacks has been a
focal point in the clean air debate for over 20 years. Further emissions reductions are being
justified, in part, by how they will improve this unique resource. How much these improvements
are worth to the public is important to understand.
Not surprisingly, there are a large number of results, reflecting uncertainties in the
science, the underlying model of people's preferences for such improvements, normal statistical
uncertainties, and a variety of assumptions. Because these results have policy significance, we
work through these uncertainties and assumptions to provide a range of best estimates for use in
the policy process. We adopt a cautious interpretation of the natural science and cautious design
and analytical decisions to provide a value for an ecological outcome that scientists and
economists can agree would be achieved at a minimum by policy proposals to reduce precursor
emissions.
32
-------
The resulting cautious, best defensible estimates of the mean WTP using the base case
characterization of ecological improvements and adjusting for discount factors ranging from
three to five percent range from $48 to $107 per year per household in New York State. The
alternative scope case scenario yields mean WTP ranging from $54 to $159 per year per
household. Multiplying these population-weighted estimates by the approximate number of
households in New York State yields benefits ranging from about $336 million to $1.1 billion
per year. Accounting for statistical uncertainties underlying these estimates could halve them or
more than double them.
This study was designed to adhere closely to scientific information about the park and to
build a bridge between the natural and social sciences that could allow people to meaningfully
express a willingness to pay for ecological improvements in the Adirondacks. The methodology
adheres to all the appropriate protocols suggested by the NOAA Panel and OMB and passes their
suggested tests, most importantly the scope tests. As such our results are the culmination of over
two decades of a major federal research effort and provides long-sought and valuable
information about the benefits of air pollution policy.
33
-------
References
American Association for Public Opinion Research. 2004. Standard Definitions: Final
Dispositions of Case Codes and Outcome Rates for Surveys, Third Edition. Lenexa,
Kansas: American Association for Public Opinion Research.
Andreoni, James. 1990. Dlmpure Altruism and Donations to Public Goods: A Theory of Warm
Glow Giving.~ Economic Journal 100(401): 464-77.
Bowman, Karlyn. 2004. DPolls on the Environment^ AEI Studies in Public Opinion
http://www.aei.org/publications/pubID.14888/pub_detail.asp
Boyle, Kevin J., William H. Desvousges, F. Reed Johnson, Richard W. Dunford, and Sara P.
Hudson, 1994. DAn Investigation of Part-Whole Biases in Contingent Valuation
Studies.~ Journal of Environmental Economics and Management 27(1): 64-83.
Bloyd, Cary., et al. Tracking and Analysis Framework (TAF) Model Documentation and User's
Guide, 1996. Argonne National Laboratory, (December), ANL/DIS/TM-36.
Burtraw, Dallas, Alan J. Krupnick, Erin Mansur, David Austin, and Deidre Farrell, 1998. "The
Costs and Benefits of Reducing Air Pollutants Related to Acid Rain,D Contemporary
Economic Policy, 16, (4), 379-400.
Cameron, T.A., and J. Quiggin, 1994. ~Estimation Using Contingent Valuation Data from a
'Dichotomous Choice with Follow-Up' Questionnaire. ~ Journal of Environmental
Economics and Management 27: 218-34.
Carson, R. T., M. Hanemann, R. J. Kopp, J. A. Krosnick, S. Presser, P. A. Ruud, and V. K.
Smith, 1994. Prospective Interim Lost Use Value Due to DDT andPCB Contamination
in Southern California, Report to the National Oceanic and Atmospheric Administration.
Natural Resource Damage Assessment, Inc. La Jolla, California.
Carson, R.T., R.C. Mitchell, M. Hanemann, R.J. Kopp, S. Presser, and P.A. Ruud, 2003.
~Contingent Valuation and Lost Passive Use: Damages from the Exxon Valdez Oil
Spill.~ Environmental and Resource Economics. July 2003, vol. 25, no. 3, pp. 257-286.
Cook, Joe, Anthony Paul, Terrell Stoessell, Dallas Burtraw, and Alan Krupnick. 2002.
"Summary of the Science of Acidification in the Adirondack Park," Unpublished
manuscript, Resources for the Future.
Cummings, Ronald G., and Laura O. Taylor, 1999. Unbiased Value Estimates for
Environmental Goods: A Cheap Talk Design for the Contingent Valuation Method,~
American Economic Review 89(3): 649-665.
Davis, R.K., 1963. The Value of Outdoor Recreation: An Economic Study of the Maine Woods.
Dissertation, Harvard University.
Driscoll, Charles T., Kimberley M. Driscoll, Karen M. Roy, and Myron J. Mitchell, 2003.
"Chemical Response of Lakes in the Adirondack Region of New York to Declines in
Acidic Deposition," Environmental Science & Technology, 37(10): 2036-42.
34
-------
Driscoll, C.T., G.B. Lawrence, A.J. Bulger, T.J. Butler, C.S. Cronan, C. Eager, K.F. Lambert,
G.E. Likens, J.L. Stoddard, and K.C. Weathers. 2001a. Acidic Deposition in the
Northeastern United States: Sources and Inputs, Ecosystem Effects, and Management
Strategies. Bioscience 51(3): 180-98.
. 2001b. Acid Rain Revisited: Advances in scientific understanding since the passage of
the 1970 and 1990 Clean Air Act Amendments. Hubbard Brook Research Foundation.
Science Links Publication. Vol. 1, no. 1.
Haab, Timothy C., Kenneth E. McConnell. 1997. "Referendum Models and Negative
Willingness to Pay: Alternative Solutions." Journal of Environmental Economics and
Management 32: 251-270.
. 1998. ^Referendum Models and Economic Values: Theoretical, Intuitive, and Practical
Bounds on Willingness to Pay.D Land Economics 74: 216-29.
. 2002. Valuing Environmental and Natural Resources. Cheltenham, UK: Edward Elgar.
Hanemann, W., J. Loomis, andB. Kanninen, 1991. ~Statistical Efficiency of Double-Bounded
Dichotomous Choice Contingent Valuation,~ American Journal of Agricultural
Economics73: 1255-63.
Johnson, F. Reed, Richard W. Dunford, William H. Desvousges, and Melissa Ruby Banzhaf.
2001. DThe Role of Knowledge in Assessing Nonuse Damages: A Case Study of the
Lower Passaic River.D Growth and Change 32: 43-68.
Kahneman, D., and J. Knetsch, 1992. "Valuing public goods: The purchase of moral
satisfaction," Journal of Environmental Economics and Management, 22: 57-70.
Lawrence, Greg. 2001. DForest and Terrestrial Systems. ~ In Acid Rain: Are the Problems
Solved? (Conference Proceedings), ed. James C. White. Rochester:Center for
Environmental Information Inc.
Mitchell, Robert Cameron, and Richard T. Carson. 1989. Using Surveys to Value Public Goods:
The Contingent Valuation Method. Washington, DC: Resources for the Future.
NOAA Panel on Contingent Valuation (Arrow, Kenneth, Robert Solow, Edward Learner, Paul
Portney, Roy Randner, and Howard Schuman), 1993. 15 CFR Chapter IX, Federal
Register 58(10): 4601-4614.
Morey, Edward, and Kathleen Greer Rossman. 2003. "Using Stated-Preference Questions to
Investigate Variations in Willingness to Pay for Preserving Marble Monuments: Classic
Heterogeneity, Random Parameters, and Mixture Models," Journal of Cultural
Economics, Vol. 27 No. 3/4, (November 20), 205-229.
Nunes, Paulo A.L.D., Jeroen C.J.M van den Bergh, and Peter Nijkamp, 2003. The Ecological
Economics of Biodiversity, Northampton MA: Edward Elgar.
Simpson, R. David, Roger A. Sedjo, and John W. Reid, 1996, "Valuing Biodiversity for Use in
Pharmaceutical Research," Journal of Political Economy, Vol. 101: 163-85.
Stoddard, J.L., et al. 1999. "Regional trends in aquatic recovery from acidification in North
America and Europe," Nature 401:575-578.
35
-------
U.S. Environmental Protection Agency, 1995. Acid Deposition Standard Feasibility Study:
Report to Congress. EPA 430-R-95-001a, October.
. 1999. The Benefits and Costs of the Clean Air Act, 1910 to 2010. Report to Congress.
U.S. Office of Management and Budget, 2003. Regulatory Analysis. Circular A-4 (September
17).
U.S. National Acid Precipitation Assessment Program, 1991. 1990 Integrated Assessment
Report.
-------
Table 1. Summary of the Science of Acid Precipitation at Adirondacks Park
Science
Instrument
Approximately 3,000 lakes, mostly small;
half degraded or devoid of fish.
In description.
Fish decline attributable to acidification
through aluminum mobilization; some from
natural causes.
In description.
Effect on forests, but less well understood.
Possible effect on birds.
Base case: Effect on one tree, two bird species.
Scope case: Effect on three tree, four bird species.
1990 CAAA reductions leave stable
ecological baseline or improving slightly;
potential of nitrogen saturation.
Base case: Baseline not worsening, not improving.
Scope case: Baseline worsening.
Uncertainty in time period for recovery;
uncertain time period to nitrogen saturation.
Uncertainty excluded.
No health effects.
Explicitly addressed and excluded in instrument.
Expected changes from lower acidification
include improvements in between 20% and
40% of lakes; small improvements in forests
and bird populations.
Base case: 20% increase in lakes that support fish
in ten years. Slight improvements to forests, birds.
Scope case: 40% improvement in lakes that
support fish in ten years; larger improvements in
more types of forest and bird populations.
37
-------
Table 2. Summary of Survey Administration
Administration
Mode
Versions
Surveyed
Useful
Responses^
Response
Rate
Share
of
Sample
Pretest
KN Panel
Web
TV/
Base
*
141
118
84%
6.5%
Internet
Main
KN Panel
Web
TV/
Internet
Base and Scope
1,143*
841
74%
46.2%
KN Withdrawn
Internet
Base and Scope
2,120*
293
14%
16.8%
Mail
Paper
Base
2,372n
570
24%
31.3%
ALL
Total/Base/
Scope
5,776/4,150/1,626
1,822/1,254/568
To exhaust the New York residency on KN's panel certain households had multiple members
surveyed. If a household had one or more surveys where there was a response to the first
referenda question, the first member of the household that completed the survey was kept as part
of the sample and the remaining members were not counted as surveyed. If there was no
response from the household, only the member of the household first solicited to take the survey
is retained in the calculation of response rates. Thus, the response rates should be viewed as a
household response rate.
^Useful responses include those surveys where respondents answered at least the first referendum
question and completed the survey in a reasonable amount of time. Respondents who indicated
they had not realized that the payment was over a ten-year period were also excluded from being
a useful response.
^3,905 mail surveys were distributed. The reported figure is adjusted for the number of
addresses that were not English-speaking residences or were forwarded to addresses outside New
York. The response rate for mail is calculated using "response rate one" (RR1) in American
Association for Public Opinion Research (2004), defining cases with no answer during a fifth
disposition reminder call as ineligible.
38
-------
Table 3. Mean and Standard Deviation of Demographic and Attitudinal Questions, by
Survey Wave and for New York Population
Variable Description
Panel
(N=959)
Withdrawn
(N=293)
Mail
(N=570)
Total
(N=1822)
New York
State Adult
Population^
Age in years
48.2
(14.3)
42.1
(13.1)
51.4
(15.1)
48.2
(14.7)
45.5
Female
58.2%
46.8%
41.4%
51.2%
52.7%
Non white
22.8%
20.8%
12.4%
19.3%
32.1%
Household size
2.50
(1.43)
3.33
(1.35)
2.67
(1.72)
2.68
(1.54)
2.61
Number of children per HH
0.55
(0.99)
1.11
(1.10)
0.64
(1.13)
0.67
(1.07)
0.65
Annual household income*
$57,928
($38,903)
$72,021
($39,001)
$67,411
($49,071)
$63,078
($42,585)
$57,171
Expectation of income in five years
Lower than current
55.0%
52.1%
57.1%
55.2%
N/A
Same as current
14.2%
10.2%
18.6%
15.0%
N/A
Higher than current
30.8%
37.7%
24.3%
29.9%
N/A
High school educated
96.4%
99.6%
96.3%
96.8%
79.1%
Heard of Adirondack Park
90.1%
91.1%
92.0%
90.8%
N/A
Distance (mi) to Park
entrance
149.4
(63.3)
150.5
(61.0)
144.7
(64.0)
148.1
(63.2)
N/A
Reside in a metropolitan
area
93.3%
94.5%
89.0%
92.2%
92.1%
NY resident 5+ years
96.2%
96.1%
97.0%
96.4%
91.8%
Paid NYS taxes last year
84.5%
91.5%
85.1%
87.0%
N/A
Environmentalist
12.3%
11.7%
22.0%
15.4%
N/A
Self-identified political persuasion
Liberal
18.4%
17.6%
18.8%
18.5%
N/A
Moderate
67.0%
65.1%
61.5%
65.0%
N/A
Conservative
14.4%
17.3%
19.7%
16.5%
N/A
*Computed assuming each household is at the midpoint of its income range.
^Drawn from 2000 U.S. Census.
39
-------
Table 4. Share Voting for Program by Bid and Scenario
First Vote Bid Level
Base Scenario
Scope Scenario
P-value
25
65.6%
73.5%
0.10
(291)
(147)
*
35
44.8%
—
(29)
(0)
*
85
39.3%
—
(11)
(0)
90
50.9%
63.4%
0.02
(275)
(142)
150
41.8%
57.9%
<0.01
(316)
(140)
200*
32.3%
—
(10)
(0)
250
36.3%
51.5%
<0.01
(289)
(134)
P-value
(chi-square):
<0.01
<0.01
P-value
0.03
0.04
(Kendall's tau/
(Sample Size in Parentheses)
"Bid values were used in the second pretest only, so sample sizes are small.
^The chi-square test provides a test of simple joint inequality across bid levels; Kendall's tau is a
stronger test of monotonic ordering.
40
-------
Table 5. Log-Normal Econometric Models (Base Instrument)
Variable Model 1 Model 2 Model 3
Constant 4.6277*** 4.0116*** 4.4076***
(5.53) (5.23) (20.08)
Sigma 1.5432 1.2737 1.3825
Income < $20k 0.5221 0.3929
(1.41) (1.53)
Income $20-35k 0.2331 0.1210
(0.94) (0.54)
Income $35-50k 0.2464 0.2208
(1.09) (1.10)
Income >$125k 0.5569** 0.6611***
(2.41) (2.90)
Future income higher 0.2191** 0.2267***
(2.32) (2.81)
Household size -0.0849 -0.1036
-(0.87) -(1-22)
Presence of children (0/1) -0.4767 -0.5286*
-(1.41) -(1-90)
Number of children 0.2717* 0.3491**
(1.73) (2.39)
Female -0.0688 -0.1846
-(0.38) -Q.30)
Black (Not Hispanic) -0.2058 -0.5227*
-(0.68) -(1 93)
Other (Not Hispanic) 0.0608 0.1080
(0.16) (0.34)
Hispanic -0.3397 -0.4557
-(0.82) -(1 50)
Age -0.0181 0.0200
-(0.61) (0.75)
Age2 0.0002 -0.0002
(0.50) -(0.58)
Reduce spending on clean air & water -0.7298 -0.4684
-(1.38) -(1.08)
41
-------
Increase spending on clean air & water
0.9370***
(3.89)
0.3284
(1.53)
Environmentalist
0.7453***
(3.83)
0.4484***
(2.57)
Frequent visitor to park
0.4874**
(2.35)
0.2903*
(1.65)
Live in park
-0.5603
-(1.21)
0.0027
(0.01)
Distance to park (km)
-0.0019**
-(2.01)
-0.0008
-(1.08)
Protect environment at any cost (warm glow)
1.3655***
(7.34)
1.3915***
(7.22)
Health effects (minor)
0.5519***
(3.41)
0.7562***
(4.43)
Health effects (significant)
1.4671***
(4.87)
1.1131***
(3.12)
Future w/o liming is worse than survey depicts
0.2899
(1.63)
0.2939
(1.62)
Future w/o liming is better than survey depicts
-0.3603
-(1.18)
-0.1557
-(0.54)
Other animals effected
0.1029
(0.69)
0.0986
(0.63)
Didn't pay taxes
0.3207
(1.32)
0.3112
(1.45)
Liming not practical
-1 1779***
-(4.24)
-1.4697***
-(4.63)
Not confident in NY State to admin, program
-0.4930***
-(3.56)
-0.5117***
-(3.23)
Don't raise taxes for any reason
-0.2508
-(1.31)
-0.2225
-(1.18)
Vote doesn't matter
-0.1460
-(1.00)
-0.0138
-(0.09)
Upwind polluters are at fault
-1.1976**
-(2.40)
-1.8245***
-(2.71)
N
938
872
1056
Log Likelihood
-925.75
-678.10
-921.22
(Z-scores in parentheses.)
42
-------
Table 6. Identification of Possible Yea-saying and Nay-saying Votes
Indicator
Share of Final
Sample Treatment
Yea-saying
Costs should not be a factor 24.9%
Some health effects 38.1%
Significant health effects 10.3%
The future status of Adirondacks is worse 18.4%
than described
Other animals are affected beyond those 58.5%
mentioned
Does not pay taxes 13.0%
Dropped or controlled.
Given chance to revote. Others
controlled.
Given chance to revote. Others
dropped or controlled.
Controlled.
Controlled.
Dropped or controlled.
Nay-saying
Taxes should not be raised for any reason 21,1 %
The future status of Adirondacks is better 8.3%
than described
Not confident in New York State 37.1%
government to run the liming program
Liming program not practical 15.0%
Voted against program solely because 19.1%
upwind polluters should reduce instead
Controlled.
Controlled.
Given chance to revote. Others
controlled.
Dropped or controlled.
Controlled.
43
-------
Table 7. Base Improvement: Mean WTP, By Yea-saying Controls, Nay-saying Controls,
and Distributional Assumption (L=Lognormal, W=Weibull)*
Nay-saying Controls
None
All
econometric
controls
Econometric
controls,
drop tax
haters
Econometric
controls, drop
if tax hater
AND lime
rejector
Econometric
controls,
drop if tax
hater OR
lime rejector
Distribution
L W
L W
L W
L W
L W
Yea-saying Controls
None
324 180
N=1175
986 565
N=1099
817 505
N=884
1037 609
N=1024
721 441
N=795
All
econometric
controls
66 58
N=1102
213 159
N=1056
203 159
N=848
223 173
N=983
201 156
N=763
Econometric
controls,
drop warm
glower
77 63
N=841
223 156
N=800
194 156
N=638
Econometric
controls,
drop if
warm
glower
AND health
embedder
67 58
N=1050
231 166
N=1004
238 180
N=932
Econometric
controls,
drop if
warm
glower OR
health
embedder
80 63
N=810
266 179
N=770
233 180
N=544
*N applies to both lognormal and Weibull distributions.
44
-------
Table 8. Scope Improvement: Mean WTP, By Yea-saying Controls, Nay-saying Controls,
and Distributional Assumption (L=Lognormal, W=Weibull)*
Nay-saying Controls
None
All
econometric
controls
Econometric
controls, drop
tax haters
Econometric
controls, drop
if tax hater
AND lime
rejector
Econometric
controls, drop
if tax hater
OR lime
rejector
Distribution
L W
L W
L W
L W
L W
Yea-saying Controls
None
730 316
N=532
1791 841
N=515
2597 1962
N=424
1962 782
N=497
2921 866
N=388
All
econometric
controls
135 98
N=523
308 179
N=506
316 135
N=417
336 179
N=488
346 144
N=384
Econometric
controls,
drop warm
glower
122 98
N=394
331 192
N=380
341 142
N=307
Econometric
controls,
drop if warm
glower AND
health
embedder
134 98
N=487
308 180
N=470
337 180
N=452
Econometric
controls,
drop if warm
glower OR
health
embedder
123 99
N=359
332 188
N=346
384 155
N=252
*N applies to both lognormal and Weibull distributions.
45
-------
Figure 1. Share Voting for Program by Bid and Scenario
-------
Figure 2. Model and Statistical Uncertainty of Mean WTP
for All Econometric Models
Lognormal Weibull
Base
Lognormal Weibull
Scope
47
-------
Appendix A: Geographic distribution of respondents within New York State.
48
-------
Appendix B: Screen and page captures from survey
Map illustration of size of affected lakes.
Close-Up of One Area in the Park
This map illustrates one small part of the Adirondack State Park. This part is located where the
red dot is on the inset map. Most of the lakes affected by past air pollution are small; they are
typically much smaller than Central Park in New York City. The large lakes that you may have
heard of (such as Saranac Lake or Lake George) are much bigger than Central Park and are not
lakes of concern.
49
-------
Chart representing how the program will affect the lakes.
Here is a picture of how the program will affect the lakes:
About 600 lakes of concern have the right soils and chemistry to
benefit from liming. The other 900 lakes of concern would not be
helped by the improvement program.
Today
Around 2014
Lakes of
Concern
(50% lakes)
Healthy
Lakes
(50% lakes)
1,500
With
Program
Without
Program
1,500
Improved
Lakes
(20% lakes)
Healthy
Lakes
(50% lakes)
1,500
Healthy
Lakes
(50% lakes)
Lakes of
Concern
(30% lakes)
Lakes of
Concern
(50% lakes)
-------
Representation of forest improvement
Here is a picture of how the program will affect the forests:
Scientists expect that the area of the Adirondacks with healthy
forests will increase from 90% to 99% as a result of the program. In
addition, no more forests will get worse. As the forest improves, the
wood thrush and tree swallow populations in the Adirondacks will
increase from about 80% to about 95% of what they once were.
Healthy forest
Healthy red spruce, sugar maple and white ash stands
Forests of concern (10% of forest Today, 1% of forest With Program)
Worsening forests (2% of forests)
-------
ISSN 1045-6333
HARVARD
JohnM. Olin Center for Law, Economics, and Business
THE VALUE OF REGIONAL
WATER QUALITY IMPROVEMENTS
W. Kip Viscusi
Joel Huber
Jason Bell
Discussion Paper No. 477
06/2004
Harvard Law School
Cambridge, MA 02138
This paper can be downloaded without charge from:
The Harvard John M. Olin Discussion Paper Series:
http://www.law.harvard.edu/programs/olin_center/
52
-------
JEL Codes: Q25, K32
The Value of Regional Water Quality Improvements
W. Kip Viscusi,* Joel Huber, and Jason Bell
Abstract
Four years ago, Magat, Huber, Viscusi, and Bell (2000) reported pretest results that
introduced an iterative choice approach to valuing water quality improvements. This paper
applies this approach to a nationally representative sample of over 1,000 respondents. We find
that the method provides stable, policy relevant estimates of the amount people are willing to pay
for improvements. Willingness to pay for a one percentage point improvement in water quality
has a mean value of $23.17 with a median of $15, and appropriately increases with family
income, age, education, and the likelihood of using lakes or rivers. In addition, the method passes
an external scope test demonstrating that greater gains in the percent of water rated "good"
increase the likelihood that the respondent will choose the alternative with better water quality.
We tested the appropriateness of a national web-based panel of respondents and find that the
Knowledge Networks sample does not fall prey to difficulties that could plague such panels.
First, the sampled web-based panel matches United States demographics very well, and
predictors of sample responsiveness, such as the likelihood to take a long time to respond to the
survey, have minimal impact on the critical estimates of the value of good water. Second, the
results are quite insensitive to doubly censored regression that accounts for the portion of
respondents who indicated an unboundedly high or low estimate for the value of cleaner lakes
and rivers. Finally, the stability of the benefit values is further demonstrated by the selection-
corrected estimates that adjust for people invited to participate but who did not successfully
complete the survey.
Keywords: water quality, environmental benefits, survey, contingent valuation
*
Cogan Professor of Law and Economics, Harvard Law School, Hauser 302, Cambridge, MA 02138
ph: (617) 496-0019 kip@law.harvard.edu
This research was supported by EPA Cooperative Agreement R-827423-01-4 with Harvard University. Viscusi's
research is also supported by the Harvard Olin Center for Law, Economics and Business. Helpful comments were
provided by Dr. Alan Carlin, John Powers and Mahesh Podar.
53
-------
The Value of Regional Water Quality Improvements
W. Kip Viscusi, Joel Huber, and Jason Bell
© 2004 W. Kip Viscusi, Joel Huber, and Jason Bell. All rights reserved.
1. Introduction
The economic benefit of water quality improvements is society's willingness to pay for
increases in water quality. Early measures of water quality were derived from travel cost values
of recreational benefits1. Subsequent benefit assessments, which remain in use in some policy
applications, consist of analyzing the value of improvements in the water's ranking on a water
quality ladder.2 This unidimensional water quality index assumes that there is a hierarchy of
quality levels in terms of whether the water is drinkable, swimmable, fishable, or boatable.
Thus, water that is drinkable also meets acceptability criteria for all lower ranked uses.
Unfortunately, this hierarchical characterization is problematic, as these categories of uses do not
reflect our current scientific understanding of the empirical ordering of water quality. That is, if
one examines the pattern of quality levels across states, there is almost no evidence of such a
hierarchy.3 The focus of the survey results reported here is on people's willingness to pay for
water that is rated "good" based on an overall index, developed by the U.S. Environmental
Protection Agency (EPA), that initially merges benefits with respect to fishing, swimming, and
the quality of the aquatic environment. An additional survey component makes it possible to
1 See Berkman and Viscusi (1973).
2 Mitchell and Carson (1989) and Carson and Mitchell (1993) provide benefit assessments using this approach,
which was consistent with the previous scientific literature at that time. A different perspective is provided by Smith
and Desvousges (1986).
3 Examples of these differences using data from EPA's National Water Quality Inventory appear in Magat, Huber,
Viscusi, and Bell (2000), pp. 10-11.
54
-------
separate the component values.4 The survey results reported here will focus on the overall water
quality valuation component.5
This paper expands and tests the methodology developed by Magat, Huber, Viscusi, and
Bell (2000), where water quality values are derived from hypothetical market choices. These
values are based on simple choices between regions that differ on water quality and cost of
living. A series of such choices yield bounds on the value of water quality improvements for
each individual. The method has the advantage of generating estimates of the private value of
improvements in water quality from a simple understandable task.
This paper discusses econometric stability of these estimates as well as some reliability
and sampling questions that arise in this use of iterative choice to assess private values. The
study is based on over 1,000 new surveys implemented through web-based interviewing.
Generally, we find that water quality valuations follow expected economic patterns: factors such
as income, education, and visits to lakes or rivers are appropriately related to the value of water
quality. Further, a scope test indicates greater valuations for larger changes in water quality
gains, increasing confidence in the metric quality of the results. We assess the reliability of this
approach by testing for the stability of the results given different econometric assumptions, with
particular focus on those responses for which the dollar value of water quality could only be
bounded on one side.
A second important improvement in this study is the use of a national web-based panel
rather than the recruitment to regional central sites or mall intercepts used in the Magat et al.
(2000) study. The use of respondent panels for policy has emerged as a response to increasing
difficulty and expense attached to recruiting probability-based random samples. It is
4 See Magat, Huber, Viscusi, and Bell (2000).
5 The attributes of good water quality will be addressed in a separate survey to be administered by the authors in
2004.
55
-------
fundamentally an empirical question whether a panel-based sampling approach will produce
acceptable results. We find that the demographic characteristics of the final sample closely
correspond to that of the target universe of U.S. adults. Additionally, we show that that the
results are not affected by factors that might distinguish between those who take the survey
against those who do not. Finally, a sample selection procedure adjusts the water quality
valuations for the probability that a panel member will not take or successfully complete the
survey. These estimates differed little from the unadjusted means, providing assurance that they
are relatively independent of possible panel selection biases.
Section 2 describes the overall study design, the survey methodology and the iterative
choice method for generating values for improvements in water quality. Section 3 explores the
logical adequacy of the results, including an exploration of consistency tests for the responses as
well as the variation of the valuation responses conditioned on demographics. Section 4
provides tests of survey and sample validity. The survey was internet-based, using the
Knowledge Networks panel. We examine the extent to which attrition bias from the panel and
other aspects of this survey mode influence the water quality values. As indicated in the
concluding Section 5, the results are quite robust and meet a wide variety of tests for rationality
and consistency.
2. Study Design
The survey used a computer-based methodology and was administered to a representative
national sample.6 The average respondent completed the survey in 25 minutes. The instrument
initially acquainted the respondent with the meaning of regional differences in lake and river
water rated of good quality and differences in annual cost of living. This introductory section
6 While our survey uses an iterative choice format, it is related to contingent valuation surveys, though it uses a
different survey approach. For discussions of contingent valuation, see among others Bishop and Heberlein (1990),
Fischhoff and FuA>y (1998), and Mitchell and Carson (1989), and Schkade and Payne (1986).
56
-------
establishes the cognitive groundwork for the respondents so that a choice between regions
differing in these aspects can be reliably answered.
Introductory section in the survey
The key valuation task involves choices between regions differing in their levels of water
quality and the annual cost of living. A critical part of the method involves introductory sections
that encourage the respondent to think about these tradeoffs. This process begins with some very
general questions to encourage the respondent to think about the value of freshwater bodies. It
also elicits information on the frequency of visits to lakes and rivers as well as related activities,
such as boating, fishing, or swimming. The primary reason for asking about usage is to
encourage respondents to think about why they might value differences in water quality.
However, it may also be the case that respondents reporting greater usage of lakes and rivers
have higher valuations of improvements in the quality of those water bodies.
Immediately following the introduction to water usage, the survey explains the meaning
of cost of living and elicits the respondent's level of concern with an annual increase in cost of
living of $200. Respondents then respond to a question that tests comprehension involving a
simple choice between two regions, identical except that one is more expensive. The few
respondents who chose the more expensive location are provided a brief educational module
before being asked to proceed.
Next, respondents are introduced to the criteria that define what it means for water
quality to be "good." Consistent with definitions used by EPA's National Water Quality
Inventory, the survey provides the following definition:
The government rates water quality as either
* Good, or
* Not Good.
57
-------
Water quality is Good if the water in a lake or river is safe for all uses. Water
quality is Not Good if a lake or river is polluted or unsafe to use.
More specifically, water quality is Good if the lake or river
* Is a safe place to swim,
* Fish in it are safe to eat, and
* Supports many plants, fish, and other aquatic life.
Water quality is Not Good if the lake or river
* Is an unsafe place to swim due to pollution,
* Has fish that are unsafe to eat, or
* Supports only a small number of plants, fish, and other aquatic life.
The survey then explicitly excludes drinking water from the valuation task.
Once familiar with the concepts of water quality and cost of living, these contexts are
framed within context of a region, defined as "within a 2-hour drive or so of your home, in other
words, within 100 miles." A 100 mile radius is appropriate because it reflects a reasonable 2-
hour drive for the recreational use of bodies of water, and about 80 percent of all recreational
visits for lakes, rivers, and streams are within such a radius.7 This text explanation of region
contrasts with the method reported in Magat et al. (2000) where respondents viewed pictorial
representations of the region size. However, our pretest interviews indicated that the 100-mile
region radius could be well understood when described through the text used.
After they learned about water quality and the region, respondents received a warm-up
choice. In this case they were asked to choose between two regions that differed in the
percentage of water bodies with quality rated good. Respondents who preferred the region with
a lower percent of lakes and rivers rated good received a brief interactive tutorial on the meaning
of the benefit measure and the error in their response.
Key Valuation Choice Task
7 Data generated by the EPA NCEE Office for this study indicate that 77.9% of boating visits, 78.1% of fishing
visits, and 76.9% of swimming recreational visits are within a 100 mile radius. Calculations were made by Jared
Creason of NCEE using the 1996 National Survey on Recreation and Environment.
58
-------
Once respondents learn about water quality, cost of living and their application to a
region, they are ready for the iterative choice questions. This key valuation task is designed to
elicit the respondent's tradeoff between water quality and cost of living in choices between
different regions. These regions are "the same in all other ways, including the number of lakes
and rivers near your home." As a final warm-up question respondents are asked to make a
choice where one alternative dominated another on both cost of living and water quality. That is,
they choose between two regions, where one region had more quality lakes and rivers and lower
cost of living. Respondents who erred received a remedial tutorial that reviewed the nature of
the choice being made.
The critical choice questions take the form shown in Figure 1. It is noteworthy that the
task itself is not complex, which past evidence suggests should enhance the validity of the survey
approach.8 We will also present a series of rationality tests of the survey responses as validity
checks of the methodology.
If a respondent was indifferent in the initial choice presented in Figure 1, then the
iterative choice process is complete, yielding a cost of living willingness to pay value for the
illustrated choice of ($300-$ 100) / (60%-40%) = $10 per 1 percent improvement in water
quality. A choice of either alternative led to successive choices that terminated either at
indifference or a narrowly bounded value estimate. Specifically, if we let C, be the cost of living
in region i, i=l,2; and let G, be the percent of water in region i rated good, then the value V of
water quality benefits is given by
V = (C2-Ci)/(G2-Gi).
8 DeShazo and Fermo (2002) show that complex choice sets can pose difficulties with respect to respondents' ability
to process the choices and give consistent responses.
59
-------
Figure 2 displays the logic of the iterative choice questions. The program iterates
choices, each time degrading the desirable aspect of the last alternative chosen until the selection
reverses. For example, a respondent preferring the lower cost region on the initial question in
Figure 1 then considers the same pairwise choice, except the cost of living in that region is
raised. Continued preference for the lower cost region leads to continued increases in the cost of
living in the chosen region until the respondent faces a dominated choice in which the regions
have the same cost of living but differ only in terms of water quality. Similarly, continued
preference for the higher quality region leads to continued reductions in the water quality of the
chosen region until the regions have the same water quality but differ only in cost of living. This
series of questions permits a bounded estimate value of water quality improvements for all
respondents except for those at the corners of the decision tree. For these corner respondents, we
analyze their results in two ways. First, for those respondents who choose the non-dominated
region, we estimate the value as twice the maximum observed dollar value for water
improvements for those with very high and halved it for those with very low values of water
quality. Second, we used more appropriate econometric treatment for those respondents based
on censored regression methods, as described in Section 3.
As another check of rationality, for respondents who reach a corner boundary of the tree
indicating zero value for money or good water, the survey brings this decision to the
respondent's attention, offers a chance to reconsider, and then inquires regarding the reason for
their choice. The analysis deals with the 6% who indicated that they would still choose the
dominated alternative or had no preference by dropping them from the initial analysis and by
treating as non-respondents in the Heckman adjustment for selection bias.
60
-------
The survey also ends with a number of additional sections, such as a brief series of
demographic questions and whether the respondent had difficulty understanding any part of the
survey.
This process of elaborate training before the choice questions is one we have used a
similar formulation in a wide variety of other environmental risk contexts. We have found that
with sufficient grounding, the tradeoff against cost of living can be well understood.9 We
deliberately framed the choice as one between regions similar to but abstracted from the region
where the person now lives. This abstraction is one that we believe contributes to the stability,
validity and actionability of the results. In terms of stability, not having to focus on a particular
body of water conditioned on the location of one's home discourages inferences about one's
particular circumstance that may or may not apply to a particular change in the percent of good
water quality in a region. In terms of validity, the survey focuses on a free market choice that
has minimal social consequences—whether one buys in region A or B primarily influences one's
own utility. These market choices contrast with referenda where one's vote can affect the
welfare of others, confounding the results with an array of conflicting forces including altruism,
confidence in the efficacy government action, willingness to impose costs on others, and
attitudes about taxation to fund such referenda. Finally, the results are actionable in helping to
establish a general social metric for policy decisions across regions. The projected dollar value
for changes in water quality can be related to general citizen characteristics such as age, income
and education. These values can be applied using census data to evaluate a broad range of
options that affect the quality of water.
Experimental conditions
9 The first of these many studies is Magat, Viscusi, and Huber (1988).
61
-------
In order to test the robustness of the results to different versions of the questionnaire,
randomly identified groups received alternative versions. These tests permit an assessment of
the effects of anchoring and the initial range of the alternatives in the initial trade off.
Our study tests for anchoring influence by manipulating the presence of an external norm
for water quality. Approximately half the respondents received information that the national
average of water quality was rated 65% Good, whereas the other respondents received no
national information. Being told the US 65% value may increase the sensitivity to water quality,
since there is now an anchor that helps respondents value of the water percent amounts provided.
Second, the value of a given change in percent good may itself be affected by the range
of percent good and dollars in the initial choice. For example, if the first choice is between a gain
of 20% good in return for $400 in cost of living (e.g., $20 for one percentage point), then
respondents may reasonably use that information to assume that, say, $15 is a good price to pay
for one percentage point gain. By contrast, if the initial choice pits a 20% gain against $200,
($10 per one percentage point), then the $15 seems relatively high. This inference is
understandable if one takes the Gricean (1975) assumption that the initial choices provided in
such questionnaires are reasonable. To test the impact of the initial range we altered the initial
range in cost of living to be either $200, $300 or $500, and the range of the gain in percent good
to be either 20, 30 or 40 percentage points. This test is whether the initial choice is
appropriately sensitive to ranges, as required for appropriate sensitivity to scope.
3. Valuation of Water Quality Improvements
In reporting our results we first give the mean and distribution of our unit water quality
benefit measure, the dollar value of a one percentage point change in water quality. Then, to
validate the results, we regress these valuation measures against respondent characteristics to
62
-------
demonstrate that the kinds of respondents expected to have higher or lower valuations indeed
have them. To show that these results are meaningful for policy, we demonstrate that the initial
choice is appropriately sensitive to scope. That is, the choice of the region with better water
quality increases with its advantage in percent good, and goes down with its disadvantage in cost
of living.
Overall Benefit Values
The benefit value measures how much of an increase in the annual cost of living
respondents are willing to incur for each percentage point improvement of water rated good. For
each respondent, this value V is calculated at the point of indifference between two regions or
the average V where a finite bound can be estimated. The mean value of V for a 1 percent
improvement in water quality is $23.17 per year, with a standard error of the mean of 0.79, based
on 1,103 respondents.10 The median water quality benefit value V is $15, which indicates that
the benefit distribution is skewed with a large upper right tail. It is reassuring to note that these
summary statistics correspond well to a mean of $22.40 and median of $12 reported by Magat et
al. (2000).
There was a substantial variability in water quality values across people. Respondents at
the 25th percentile registered a value of $6.25 per unit improvement in water quality, as
compared to $15 at the median and $30 at the 75th percentile. The disparity between the
valuation at the 10th percentile value of $1.92 and the 90th percentile value of $75 indicates
substantial heterogeneity in the value respondents place on clean lakes, rivers, and streams.
Validity Tests
10 Carson and Mitchell (1993) examined willingness to pay for national water quality and estimated that people
would pay $242 in 1990 dollars (or $315 in 2003 dollars) annually to improve from a baseline of non-boatable to
nationally swimmable.
63
-------
Two validity tests provide evidence of the meaningfulness of the estimated water quality
values. The first test requires that the individual estimates of water quality value differ across
respondents in ways predicted by economic theory. The second validity assessment is an across
person test requiring respondents to be sensitive to the scope of differences in cost of living and
water quality provided.
Consider first the relationship between generated values and respondent characteristics.
The Magat et al. (2000) survey found very weak relationships between valuations and
demographic characteristics. The current results are far more substantial, perhaps due to a
sample almost three times as large and because of better survey implementation. The dependent
variable for analysis is the log of respondent's unit water quality benefit value, V. The log
transformation is used because it has the effect of making the right-skewed distribution of V
approximately normal.
Table 1 presents two sets of regression results for the log value of V, the unit value of
water quality. The first column presents the OLS estimates, while the second column of results
presents the censored Tobit regressions. Survey respondents consistently choose the low priced
or high quality option eventually reach or the corner maxima or minima in the iterative choices
shown in Figure 2. The censored regression in effect combines the information from the
respondents who hit the upper or lower limits with conventional regression results for the
bounded respondents. Thus, the censored regression coefficients makes the best prediction
taking into account the fact that the survey truncates the distribution of possible responses at both
the high and low end of the distribution of water quality values. The Tobit estimates in Table 1
are remarkably similar to the OLS estimates.
64
-------
The statistically significant explanatory variables all have coefficients that one would
expect. The coefficient of. 17 for log income indicates that water quality is a normal good, with
valuations increasing by 17% for a doubling in income. Individual education is likely to be a
proxy for lifetime wealth. Better educated respondents exhibit a higher value for good water
quality, controlling for current income levels and personal characteristics. Older respondents
likewise indicate a higher valuation of water quality that is consistent with life cycle changes in
wealth.
Two variables that should reflect whether a respondent is likely to have particularly
strong preferences for good water quality are whether the respondent is a member of an
environmental organization or has visited a lake or river in the last 12 months.11 The coefficients
of the environmental group membership and environmental activities variables were almost
identical in magnitude, with each increasing the value of water quality by around 28%. The
significant positive influence on benefit values of visits to lakes and rivers accords with previous
research by Cameron and Englin (1997) showing that respondent experience with the good being
valued raises the valuation amounts. After accounting for the influence of the environmental
variables and demographic effect such as income and education, variables pertaining to region,
race, and gender were not significant on an individual basis.
Whether the respondent was told the percentage of water in the country rated good did
not have a statistically significant effect on valuations. The sub-sample that was given
information pertaining to this possible anchor exhibited no difference in their valuation amounts.
This result indicates that the respondents focused on the difference between the alternatives in
the choice set, rather than on the presence of an external reference point.
11 The particular environmental organizations listed in the survey for possible membership were the following:
Environmental Defense Fund, Greenpeace, National Audabon Society, National Wildlife Federation, Nature
Conservancy, Natural Resources Defense Council, and Sierra Club.
65
-------
External Scope Tests
The second validity assessment is an external scope test. The scope test is important in
establishing context that the estimates of V were a meaningful quantitative, valuation metric.12 If
respondents are willing to incur the same cost of living increase for a 20 percentage point change
in water quality as a 40 percentage point change, then all one is measuring is a general attitude
towards water quality over cost of living, such as "warm glow" effects. The test we report is
across respondents, a stronger test than a within subject test.
This test is possible because we altered the initial range of water quality ranges and the
cost of living across respondents. In particular, one of the alternatives in the initial choice was
either 20, 30, or 40 percentage points in good water quality higher than the other, and the
difference in cost of living was either $200, $300 or $400 per year.13 To demonstrate appropriate
sensitivity to the scope of the choice, respondents' initial choices should favor the region with
higher water quality when its gain in water quality is greater. Similarly, respondents should
favor the region with lower cost of living when its gain in living expense is greater. Table 2
displays a logistic regression predicting initial choice as a function of initial ranges and the
demographic variables used to predict the final valuation amounts for the regressions in Table 1.
The variables pertaining to each of the scope tests are significant and in the expected direction.
Increasing the water quality difference or decreasing the cost of living difference makes one
more likely to choose the alternative with higher water quality. Further, the characteristics that
predict the initial choice for the regressions in Table 2 parallel those predicting the final tradeoff
reflected in the regressions in Table 1, with choice of the high water quality option increasing
12 For a detailed review of scope tests and the ability of contingent valuation studies to pass scope tests, see Smith
and Osborne (1996).
13 We also altered the average levels of water quality to see if response depended on these. Those analyses are
available on a working paper: "Coping with the Contingency of Valuation: Range and Anchoring Effects in Choice
Valuation Experiments," Huber, Viscusi and Bell (2004).
66
-------
with age, income, education and the environmental preference variables such as visits to lakes
and rivers or membership in environmental organizations.
4. Evaluation of the Panel Sample
Sample Characteristics
The sample used for the study came from the Knowledge Networks (Menlo Park, CA,
www.knowledgenetworks.com) panel. Researchers on environmental benefits valuations have
increased their use of internet panels, so that the performance of this survey approach has broad
implications beyond our particular study.14 The Knowledge Networks sample consists of a
national sample of households recruited by random-digit dialing, who either have been provided
internet access through their own computer or are given a WebTV console. The underlying
Knowledge Networks sample has been selected to be broadly representative of the U.S.
population.15
Table 3 compares the sample characteristics of those who completed the survey and with
the 2001 U.S. adult population. The survey population closely mirrors the U.S. Census
distribution. One might have hypothesized that people willing to be surveyed would be better
educated, underrepresented at the extremes of income, and younger than the general population.
However, there are no major discrepancies between the sample mix for our study and the
population. While some differences are statistically significant, including the percentage of
respondents age 64 and over and the representation of some income groups, these differences are
not consequential. For example, 11 percent of the sample is age 64-74 compared to a national
average 9 percent, and 21.1 percent of the sample have household income in the $50,000-
14 Other researchers using the Knowledge Networks sample have included Krupnick et al. (2002), Berrens et al.
(2004), and DeShazo and Cameron (2004).
15 Ongoing research by Trudy Cameron and J.R. De Shazo has examined the representativeness of this sample and
has developed a selection correction to account for differences from U.S. Census averages.
67
-------
$74,999 range, as compared to the national average of 18.9 percent. Differences such as these
are to be expected, both because of the stochastic nature of the sampling process as well as the
fact that there is not an exact match up for the 2001 Census time period and the more recent
sampling period. Overall, the sample tracks the U.S. population remarkably well.
Sample Validity Tests
Because the survey was administered via the internet using an existing panel of
respondents, we undertook a series of validity tests specifically determining whether their panel
membership influenced the valuation results. To the best of our knowledge, these are the first
such tests to have been undertaken for this sampling methodology. We tested the panel
influences of four variables on the regression analysis of the determinants of the value of water
quality benefits. Table 4 reports these regression results in which these panel variables first are
added to our earlier analysis shown in Table 1 and then are included without these variables.
The first variable is whether the respondent stopped the survey and then continued the
survey at a later time. Conceivably, such respondents might be less engaged in the survey task.
However, there was no significant effect of this variable on benefit values.
The second variable of interest is the time the respondent has been a member of the
Knowledge Network panel. Length of time in the panel may affect attentiveness to surveys and
potentially could be correlated with other personal characteristics that influence water quality
valuations. The estimates in Table 4 fail to indicate any significant influence of this variable
either.
Third, the number of days the respondent took to complete the survey after being offered
the opportunity to participate could reflect a lack of interest in the survey topic or in taking
68
-------
surveys generally. Nevertheless, there is no significant effect of this variable on benefit
valuations in either of the equations estimated in Table 4
The final survey methodology variable tested is whether the respondent subsequently quit
the panel either immediately after the invitation for this survey or at any later time until May
2004, when data for this variable were collected. Such respondents could be less interested in
taking surveys and might have different valuations. However, this variable was also not
statistically significant in the water quality valuation equations.
Overall, there is no indication that any of these key aspects of the panel methodology bias
the survey responses. In addition to the general match of our respondents to the U.S. population,
we also examined whether these four variables reflecting the methodology had any influence on
the probability that the respondent failed to pass the consistency test with respect to the benefit
valuations. There were no significant effects of any of the Knowledge Networks panel variables
so that there is no evidence that national performance of the survey task is importantly
influenced by any of these variables.
Selection Effects
Although the sample is nationally representative and had a high overall response rate, it is
useful to test for possible selection biases arising from panel members who were invited to
participate but did not successfully complete the survey. Of 1,587 panel members invited to take
the survey, 74% of respondents chose to participate. Of the 1,174 participants, three respondents
did not complete the portion of the survey that elicits water quality value. Finally, 6% of
participants completed the survey but were dropped because they chose the dominated
alternative and continued with that choice even after being so informed. Therefore, 1,103 of
1,587 invitees consistently completed the water quality valuation portion of the survey. For the
69
-------
selection correction for bias, we used variables for which we had the values for non-respondents
as well as survey respondents. This data is routinely collected by Knowledge Networks on its
panel members. Thus, an additional advantage of such panels is that there is information
available to analyze who chose not to take the survey after being offered the chance to do so.
To predict participation, we identified a number of variables that significantly affected
survey completion. In particular, we found that being African American or Hispanic was
negatively associated with completing the survey, as was household size. We also constructed
two health-related stress dummy variables. The first stress variable was for individuals who
reported that they had a high stress level. The high stress variable indicated respondents who
reported more "stress, strain, or pressure" than usual "during the past few months." The second
stress variable was for people who failed to respond to the stress information question. Each of
these variables was negatively related to the probability of taking the survey but not significantly
related to the water quality valuation amount V, thus achieving the appropriate identification.
Table 5 reports the selection equation and the selection-corrected regression of the log
value of water quality. The threshold empirical issue is whether there are any statistically
significant selection effects. As the chi-squared statistic reported at the bottom of Table 5
indicates, one cannot reject the hypothesis that there is no significant effect of sample selection
on our empirical estimates. Thus, the empirical estimates are not biased in any statistically
significant way by the self-selection of respondents in the Knowledge Networks sample who
chose to complete the survey and did so successfully.
Given this absence of statistically significant selection effects, it is not surprising that the
selection-corrected estimates closely parallel our earlier estimates. Water quality values increase
with income levels, age, and education, as before. The race variable has become significant, but
70
-------
this effect may have been due in part to the omission of the environmental group membership
and water recreation use variables from the equation, since they were not available for non-
respondents.
Similar stability in the results is implied by an examination of the extent to which the
estimates of the dollar value of water quality changes with the selection adjustment. Using the
parameter estimates of the selection-corrected regression, we estimated the log value of a one
percent improvement in water quality. The average log value then decreased by 4.5% and the
antilog by 11.1% compared to corresponding estimates using parameters from the ordinary least
squares regression. These differences are well within sample variability and thus are not
statistically significant. More important, these results indicate that the estimates are not
substantively different even after careful adjustment for sample selection.
5. Conclusions
The survey results presented here passed a variety of consistency tests and rationality
checks. These tests included dominance tests as part of the iterative choice process and external
scope tests across respondents. In addition, the internet-based methodology itself was tested
with respect to a variety of potential sources of bias, such as sample attrition, and these panel
characteristics had no significant effect on the results.
It is appropriate to speculate on why these results are much stronger than those reported
in Magat et al. (1988). The earlier study produced similar aggregate values, but the covariates
with water quality value were largely insignificant, and a scope test was not even attempted. The
Magat et al (1988) study had less than half the number of respondents, but the main differences
are methodological. In the current study, greater effort was placed on preparing the respondent
to make the trade-off between water quality and cost of living. Three warm up questions
71
-------
involving dominated choices provided easy ways to understand the choice task, and for the
relatively low percent of respondents who 'failed' those questions, provided a way to
communicate the importance of their answers.
Working with a panel had several advantages. First, since our survey design involved the
use of a computer-based sample, the Knowledge Networks panel yielded a more representative
sample of survey participants than other survey methods such as those used by Magat et al.
(2000) in which a group of subjects contacted by phone came to a central location to take the
survey. Second, respondents in the panel are accustomed to taking surveys, so they are not
confused by the process. Third, and most important, because there are data on those who
declined to take the survey, it is possible to estimate the impact of that self-selection on our
results. In this case, that self-selection had minimal effect on our estimates. However, that result
strictly applies only to our focal question about the value of water quality. The real value of
panels is that they contain the information that permits an assessment of the impact of respondent
selection mechanism that will certainly be an even greater problem in the future.
The practical benefit of these results is that they provide unit water quality benefit values
that can be matched to existing EPA measures of water quality to provide an assessment of
benefits of water quality programs. Good water quality has a unit value of $23 per percentage
point increase in water quality. This value is dependent on variables such as income, education,
and personal use of lakes and rivers in the expected fashion. To value water quality
improvements, one can use these values in conjunction with results that break down the benefits
in terms of benefits for the components of water quality—fishing, swimming, and health of the
aquatic environment— to gauge the economic benefit of an improvement project to the affected
local population.
72
-------
References
Berkman, Richard and W. Kip Viscusi. (1973). Damming the West. New York: Grossman
Publishers.
Berrens, Robert P., Alok K. Bohara, Hank C. Jenkins-Smith, Carol L. Silva, and David L.
Weimer. (2004). "Information and Effort in Contingent Valuation Surveys: Application
to Global Climate Change Using National Internet Samples," Journal of Environmental
Economics and Management, 47(2), 331-363.
Bishop, Richard C. and Michael P. Welsh. (1992). "Existence Values in Benefit-Cost Analysis
and Damage Assessment," Land Economics, 68(4), 405-417.
Bishop, Richard C. and Thomas A. Heberlein. (1990). "The Contingent Valuation Method." In
Rebecca L. Johnson and Gary V. Johnson (eds.), Economic Valuation of Natural
Resources: Issues, Theory, and Applications. Social Behavior and Natural Resources
Series. Boulder: Westview Press.
Cameron, Trudy, and Jeffrey Englin. (1997). "Respondent Experience and Contingent Valuation
of Environmental Goods," Journal of Environmental Economics and Management, 33(3),
296-313.
Carson, Richard T. and Mitchell, Robert Cameron. (1993). "The Value of Clean Water: The
Public's Willingness to Pay for Boatable, Fishable, and Swimmable Quality Water,"
Water Resources Research, 29(7), 2445-2454.
DeShazo, J.R., and German Fermo. (2002). "Designing Choice Sets for Stated Preference
Methods: The Effects of Complexity on Choice Consistency," Journal of Environmental
Economics and Management, 44(1), 123-143.
73
-------
DeShazo, J.R., and Trudy Cameron. (2004). "Moving Beyond the Value of a Statistical Life: A
New Theoretical and Empirical Analysis." Paper presented at 2004 AEA meetings.
Fischhoff, Baruch and Lita Furby. (1988). "Measuring Values: A Conceptual Framework for
Interpreting Transactions with Special Reference to Contingent Valuation of Visibility,"
Journal of Risk and Uncertainty, 1(2), 147-184.
Grice, H.P. (1975) "Logic and Conversation" In PCole and J.P. Forgas (eds.) Syntax and
Semantics: 3. Speech Acts (p. 41 -58/ New York, Academic Press.
Huber, Joel, W. Kip Viscusi, and Jason B. Bell. (2004). "Coping with the Contingency of
Valuation: Range and Anchoring Effects in Choice Valuation Experiments," Working
Paper, Fuqua School of Business, Duke University.
Kahneman, Daniel and Amos Tveski (1979) "Prospect Theory: An Analysis of Decision under
Risk " Econometrica, 47, 263-291.
Krupnick, Alan, Anna Alberini, Maureen Cropper, Nathalie Simon, Bernie O'Brien, Ron
Goeree, and Martin Heintzelman. (2002). "Age, Health and Willingness to Pay for
Mortality Risk Reductions: A Contingent Valuation Survey of Ontario Residents,"
Journal of Risk and Uncertainty, 24(2), 161-186.
Magat, Wesley A., Joel Huber, W. Kip Viscusi, and Jason Bell. (2000). "An Iterative Choice
Approach to Valuing Clean Lakes, Rivers and Streams." Journal of Risk and
Uncertainty, 21:1, 7-43.
Magat, Wesley A., W. Kip Viscusi, and Joel Huber. (1988). "Paired Comparison and Contingent
Valuation Approaches to Morbidity Risk Valuation," Journal of Environmental
Economics and Management, 15, 395-411.
74
-------
Mitchell, Robert Cameron and Richard T. Carson. (1989). Using Surveys to Value Public Goods:
The Contingent Valuation Method. Washington: Resources for the Future.
Schkade, David A. and John W. Payne. (1993). "Where Do the Numbers Come from? How
People Respond to Contingent Valuation Questions." In Jerry A. Hausman (ed.),
Contingent Valuation: A Critical Assessment. New York: North-Holland, 271-293.
Smith, V. Kerry, and William H. Desvousges. (1986). Measuring Water Quality Benefits.
Boston: Kluwer Academic Publishers.
Smith, V. Kerry, and Laura L. Osborne. (1996). "Do Contingent Valuation Estimates Pass a
' Scope' Test? A Meta-Analysis," Journal of Environmental Economics and
Management, 31(3), 287-301.
United States Environmental Protection Agency. (1994). National Water Quality Inventory, 1992
Report to Congress, EPA 841-R-94-001.
75
-------
Figure 1
Sample Private Water Quality Benefit Question
We would like to ask you some more questions like these. However, in these questions,
one region will have a lower annual cost of living and the other will have higher water
quality. Remember that the national average for water quality is 65% Good.
Region 1
Region 2
Increase in
Annual Cost
Of Living
$100
More
Expensive
$300
More
Expensive
Percent of Lake
Acres and River
Miles With Good
Water Quality
40%
Good
Water
Quality
60%
Good
Water
Quality
Which Region
Would you Prefer?
Region 1
*
Region 2
*
No Preference
*
76
-------
Figure 2
Survey Decision Tree
(If 1)
Region 1
Region 2
$100
$300
40%
60%
(If 2)
Regl
Reg2
Regl
Reg2
$200
$300
$100
$300
40%
60
0/
/O
40
%
55%
(If 1)
(If 1)
Regl
Reg2
$250
$300
40%
60%
Regl
Reg2
$275
$300
40
%
60%
(If 1)
Regl
Reg2
$300
$300
40%
60%
(If 2)
Regl
Reg2
$150
$300
40
0/
/o
60%
(If 1)
(If 2)
f
r
Regl
Reg2
Regl
Reg2
$100
$300
$100
$300
40%
57%
40
%
50%
(If 2)
Regl
Reg2
$100
$300
40%
45%
(If 2)
Regl
Reg2
$100
$300
40%
40%
77
-------
Table 1
Regression Estimates for Log of Unit Water Quality Benefit Value
Log (Unit Value for Good Water Quality)
OLS Censored
Variable Coefficient Standard Coefficient Standard
Error Error
Log (Income)
0.1668***
0.0480
0.1687***
0.0484
Years of education
0.0409***
0.0151
0.0423***
0.0153
Age
0.0115***
0.0023
0.0119***
0.0023
Environmental organization
0.2843
0.1734
0.3140*
0.1773
membership
Visited a lake or river, last 12
0.2822***
0.0778
0.2839***
0.0784
months
Told national water quality
0.0966
0.0728
0.0955
0.0734
Race: Black
-0.1403
0.1109
-0.1404
0.1117
Race: Non-black, Non-white
-0.0661
0.1637
-0.0844
0.1642
Hispanic
0.1415
0.1223
0.1325
0.1232
Gender: Female
0.0166
0.0727
0.0169
0.0733
Household size
-0.0093
0.0291
-0.0099
0.0293
Region: Northeast
-0.0271
0.1126
-0.0333
0.1134
Region: South
-0.0765
0.0955
-0.0814
0.0962
Region: West
-0.0997
0.1096
-0.0980
0.1107
Intercept
-0.4646
0.5243
-0.5031
0.5282
Adjusted R2
0.0614
0.0251
Notes: * significant at the .10 level, ** significant at the .05 level, *** significant at the .01 level,
all two-tailed tests.
78
-------
Table 2
Scope Test: Demonstrating the Impact of
Water Quality and Cost of Living Range on Initial Choice
Respondent Chose the Higher Water
Quality Region in First Choice
Variable Coefficient Standard Error
Logistic Regression
Initial Cost of Living Range
-0.00161**
0.00072
Initial Water Quality Range
0.0180**
0.00751
Log (Income)
0.2904***
0.0847
Years of education
0.0620**
0.0269
Age
0.0196***
0.00404
Environmental organization
0.6427**
0.3420
membership
Visited a lake or river, last 12 months
0.4445***
0.1357
Told national water quality
0.0642
0.1338
Race: Black
-0.0249
0.1933
Race: Non-black, Non-white
-0.1145
0.2846
Hispanic
0.2827
0.2154
Gender: Female
0.0574
0.1277
Household size
-0.0543
0.0508
Region: Northeast
0.0322
0.1999
Region: South
-0.1125
0.1679
Region: West
-0.1526
0.1927
Intercept
-4.6635***
0.9745
c = 0.654
Notes: * significant at the . 10 level, ** significant at the .05 level, *** significant at the .01 level,
all two-tailed tests.
79
-------
Table 3
Comparison of Knowledge Networks Sample to the National Adult Population1
Survey Participants US Adult Population
Demographic Variable Percent Percent
Employment Status (16 years or older)
Employed 65.1 66.9
Age
18-24 13.1 13.0
25-34 19.1 18.8
35-44 20.2 21.2
45-54 19.1 18.5
55-64 12.2 11.9
64-74 11.0* 8.6
75+ 5.4* 7.9
Educational Attainment
Less than HS 17.0 15.9
HS Diploma or higher 60.0 58.5
Bachelor or higher 23.0* 25.6
Race / Ethnicity
White 81.5 82.3
Black/African-American 13.1 11.8
American Indian or Alaska Native 1.0 0.9
Asian/Pacific Islander 3.1* 4.1
Other 1.3 1.0
Race / Ethnicity of Household
Hispanic 11.1 11.4
Gender
Male 51.0 48.3
Female 49.0 51.7
Marital Status (2000)
Married 61.4 59.5
Single (never married) 23.5 23.9
Divorced 9.0 9.8
Widowed 4.1* 6.8
Household Income (2000)
Less than $15,000 13.2* 16.0
$15,000 to $24,999 11.3 13.4
80
-------
$25,000 to $34,999 13.4 12.5
$35,000 to $49,999 18.9* 15.5
$50,000 to $74,999 21.1* 18.9
$75,000 or more 22.2 23.8
Statistical Abstract of the United States, 2002. 2001 adult population (18 years+), unless
otherwise noted.
* The 95% Confidence Interval for survey participants does not include mean adult US
population for this demographic variable.
81
-------
Table 4
Validity Tests Based on Censored Regression of Log of Unit Water Quality Benefit Values
Log (Unit Value for Good Water Quality)
Variable
Coefficient
Standard
Coefficient Standard
Error
Error
Log (Income)
0.1710***
0.0487
Years of education
0.0421***
0.0153
..
Age
0.0119***
0.0024
..
Environmental organization
0.3165*
0.1776
—
membership
Visited a lake or river, last 12
0.2787***
0.0787
..
months
Told national water quality
0.0966
0.0736
..
Race: Black
-0.1362
0.1129
..
Race: Non-black, Non-white
-0.0876
0.1643
..
Hispanic
0.1326
0.1237
—
Gender: Female
0.0150
0.0734
..
Household size
-0.0086
0.0295
..
Region: Northeast
-0.0381
0.11406
..
Region: South
-0.0873
0.0971
—
Region: West
-0.1024
0.1119
—
Respondent stopped and
-0.006
0.1467
0.0233 0.1517
continued survey later
Time as panel member, in
-0.0021
0.0032
0.0023 0.0032
months
Days from invitation to
-0.0013
0.0023
-0.0039 0.0024
completion
Has panel member quit panel
-0.0131
0.0789
-0.1006 0.0803
Intercept
-0.4561
0.5326
2.5538*** 0.0950
Adjusted R2
0.0254
0.0017
Notes: * significant at the .10 level, ** significant at the .05 level, *** significant at the .01 level,
all two-tailed tests.
82
-------
Table 5
Log Unit Water Quality Value Regression Results Controlling for Selection Effects
Coefficient
Standard Error
0.1701***
0.0447***
0.0122***
-0.2391**
-0.0919
0.0446
0.0106
-0.0195
-0.0561
-0.1059
-0.1297
-0.3666
0.0480
0.0150
0.0023
0.1119
0.1637
0.1241
0.0727
0.0303
0.1124
0.0951
0.1094
0.5243
-0.1929***
-1.4668***
-0.2364**
-0.3511***
-0.1178***
1.2453***
0.0749
0.1133
0.0968
0.1013
0.0246
0.0930
Variable
Regression Model for Log of Value
Log (Income)
Years of education
Age
Race: Black
Race: Non-black, Non-white
Hispanic
Gender: Female
Household size
Region: Northeast
Region: South
Region: West
Intercept
Participation Equation
High Stress level
Stress Data Unavailable
Race: Black
Hispanic
Household size
Intercept
LRtest of indep. eqns. (rho = 0): chi2(l) = 2.46 Prob > chi2 = 0.1164
Notes: * significant at the . 10 level, ** significant at the .05 level, *** significant at the .01 level,
all two-tailed tests.
83
-------
A Consistent Framework for Valuation
of Wetland Ecosystem Services
Using Discrete Choice Methods
J. Walter Milon,* David Scrogin* and John F. Weishampel"
"Department of Economics, "Department of Biology and Geospatial Analysis &
Modeling of Ecological Systems Lab
University of Central Florida
Orlando, FL
Goal and Approach
> The overall goal is to develop and test a
consistent framework to estimate wetland
services values.
> Our approach uses a joint modeling strategy to
integrate revealed preferences (RP) from a
discrete choice model of the housing market and
stated preferences (SP) from a choice model for
wetland ecosystem services.
> The analysis will be based on a comprehensive
database from a stratified sample of residential
property owners in three Metropolitan Statistical
Areas (MSAs) in Central Florida.
84
-------
Objectives
> 1) To estimate the demand for proximity to
wetlands and other water resources using
discrete choice and hedonic pricing models of
residential property values.
> 2) To estimate the demand for ecosystem
services from different types of wetlands that are
not in proximity to residential property using a
stated choice survey.
> 3) To develop and test a combined discrete
choice model from the RP and SP data to
produce a general valuation function for wetland
ecosystem services.
> 4) To estimate the implicit prices of wetland
services in mitigation banking markets.
Conceptual Framework
> Assume that housing and environmental protection are
separable so an individual n maximizes the utility
function:
Un(X,Z;S)
subject to M = PxX + PzZ
where U is utility, X is a vector of housing attributes, Z is
a vector of environmental services associated with
wetland resources, S is a vector of observed individual
characteristics, M is income, Px is a vector of prices for
housing attributes and Pz is a vector of prices for
environmental services.
> Wetland services are public goods so Pz must be
revealed through direct elicitation but some services may
be packaged with housing units.
85
-------
Conceptual Framework: Housing
> Discrete housing choice model:
Un ~ Vn (XM , XA2» ¦¦¦ > XA!, Pj) + sn
where At represents attributes of the Ath
alternative housing bundle from the choice
set K.
>The probability that individual n chooses
housing unit A is given by:
*An = eXP W VAr) /ZbsK eXP W VBr)
where X* is a scale parameter.
Conceptual Framework: Ecosystem
> Discrete stated choice model:
Un=Vn(Zc1,ZC2,...,Zcl,Pc)
where the C, attributes represent specific
wetland services such as size, type, habitat
quality, and groundwater recharge and Pc is a
cost associated with the Cth wetland alternative
from choice set E.
> The probability that individual n chooses wetland
services package Cis given by:
*fcn = exp (X2VCr) / ZDEEexp (X2VDr)
where l2 is a scale parameter.
86
-------
Consistency Tests
> To empirically investigate the consistency of
the housing and wetland services choices,
we employ the likelihood ratio test:
-2[(LA + L°) - LJoint].
> Other tests will be used to evaluate the
effects of treatments used in the stated
choice experiments.
Treatment Effects
> Evaluate the effects of information on preferences and task
complexity using a 3 x 2 block design. We contrast choice
task complexity, choices involving partial sets of wetland
attributes vs choices involving a full set of attributes, with
the format for the provision of information, text description
vs spatial description. Motivation for full/partial attribute
design is the legal context for determining mitigation.
Spatial
Description
Text
Description
Full Attributes
Partial Attributes A
Partial Attributes B
87
-------
Application
> Single family residential housing data will be collected
from county tax appraisers in three Metropolitan
Statistical Areas (MSAs) in Central Florida: Daytona
Beach, Lakeland-Winter Haven and Orlando
representing over 2.6 million people.
> From housing sales during the 2002 - 2004 period,
develop a proportionally weighted sample of 1200
purchasers across the three MSAs.
> The sample of 1200 housing buyers will be contacted to
participate in the stated choice wetlands survey. We
anticipate a 50 percent response rate (600 property
owners) to participate in interview surveys that will be
conducted at a central location within each MSA.
Application
> GIS analysis will be used to identify the neighborhood and
ecosystem attributes associated with each housing parcel.
> For the stated choice analysis, select a stratified random
sample of wetland sites based on three stratification
criteria: type of freshwater wetland, site acreage, and
whether the wetland is connected to or isolated from
surface waters. Sites will be from the land area containing
the 3 MSAs.
> Each site selected will be profiled using GIS analysis to
identify attributes of the site; these profiles will be 'ground-
truthed' with site visits and additional information from
wetland specialists in regional and state environmental
agencies.
88
-------
land Use Coverage
LAND USE Percent Cover
Orange
Polk
Volusia
URBAN & BUILT-UP
17.3%
33.2%
12.3%
AGRICULTURE
9.1%
31.8%
6.0%
RANGELAND
5.7%
5.7%
4.1%
UPLAND FOREST
6.2%
5.8%
23.3%
WATER
21.1%
4.6%
19.4%
WETLANDS
36.8%
17.9%
33.3%
BARREN LAND
1.6%
0.1%
0.3%
TRANS &UTIL
2.2%
0.8%
1.2%
89
-------
Wetlarld Coverage
WETLAND TYPES Percent Cover
Orange Polk Volusia
Mixed Hardwoods 17.2% 43.5% 37.2%
Mixed Cypress/Forest 14.8% 35.2% 39.0%
Freshwater Marshes 62.5% 16.2% 9.8%
Wet Prairies 1.2% 3.9% 3.1%
Emergent Vegetation 4.2% 1.1% 10.8%
-------
-------
Comments on:
Valuation of Natural Resource Improvements in the Adirondacks
Spencer Banzhaf, Dallas Burtraw, David Evans and Alan Krupnick
and
The Value of regional Water Quality Improvements
Kip Viscusi, Jason Bell and Joel Huber
By:
Kevin J. Boyle
Distinguished Maine professor
University of Maine
October 26, 2004
Introduction
These two papers make a nice comparison of applied studies on the benefits of improved
surface water quality. My comments will address several key features in the design and
implementation of stated-preference studies. I will discuss how each study addressed
these specific design issues. The studies hereafter will be referred to as the Banzhaf and
Viscusi studies, respectively.
Geographic Scope of Application
The Banzhaf study focuses on a specific region, the Adirondacks in New York, while the
Viscusi study is designed to develop a national water quality value. These applications
represent the two extremes of the spectrum in applied valuation studies. Regional studies
for a specific application allow the design of precise and specific valuation scenarios,
which most practitioners, I believe, would agree lead to better estimates of value in terms
of validity and reliability. National value estimates are needed by U.S. EPA for RIAs of
national policies. The question in my mind is whether the Viscusi study in the quest for a
national value for policy results in value estimates that have very little empirical
credibility. On the other hand, the Banzhaf value estimate has little relevance for national
policy.
In other words, the Banzhaf says something very specific about benefits of a particular
policy in a specific area, but has little to offer for national policy. The Viscusi study
purports to comment on a general policy, but has little to say about any specific policy or
regional application. Both types of value are needed. EPA does need estimates of value
92
-------
to address national policy initiatives. In application of policy EPA leaves states and/or
regions with considerable latitude regarding how policies will be implemented. A
national policy may overestimate or underestimate regional benefits, and more refined
value estimates are needed to consider how and to what extent a policy should be
implemented in a specific region.
The revealed-preference studies presented earlier in this session, Egan et al. and Smith et
al., present a promising approach to addressing this dichotomy of policy needs that
improves on the approaches in the Banzhaf and Viscusi studies. The Egan and Smith
studies use aquatic ecosystem attributes that define the quality of anthropocentric uses.
This type of attribute design provides flexibility in the computation of value estimates for
national policy and for regional variations in the implementation of a national policy.
Moreover, the results from these types of revealed preference studies could be used to
design choice studies similar to the Viscusi study that would provide more credible
estimates of value for both national and regional benefit calculations.
What is Being Valued?
This issue is related to my discussion above. The Banzhaf study substantially attempted
to link the valuation scenario to bio-physical information on the quality change that
would arise from a policy to improve surface water. However, the actual link is not as
clear as it could be. Two suggestions arise here. First, a clearer link could be developed
by a more formal presentation of an economic model that links the policy change to the
design of the valuation scenario, to data analyses, and to interpretation of the statistical
estimates. Second, as noted above, an attribute-based design of the scenario would have
made the bio-physical link clear in the valuation questions and would have made it
explicit in the econometric model used to analyze the valuation responses. This would
not only have improved the value estimates for application sin the Adirondacks, but
would have improved the transferability of the values estimates to other regions or for use
in national p[policy.
The Viscusi study valued "good" water quality. Admittedly these investigators were
constrained by EPA's decision to defined water quality categories. However, given the
experience of this research team one might expect a more creative design that might
allow for the estimation of more credible estimates of value while developing a mapping
that would allow the value estimates to be applied to EPA's categories for policy
analyses.
If I take the title of this workshop literally, "Improving the Valuation of Ecological
Benefitsthen it seems imperative that EPA consider the support of developing more
complex valuation approaches and empirical applications the link policy effects on
ecosystem services to changes in economic value. As stated, in the preceding section, the
Egan and Smith studies are a substantial step in this direction for revealed-preference
applications, which have important implications for the design of stated-preference
studies.
93
-------
Framing of the Valuation Question
I think it is appropriate that both studies used a total economic value approach to estimate
values. I think the value estimated in the Banzhaf study is clear and could be improve
with a more explicit model that is carried though the empirical analysis.
The Viscusi study used a unique in interesting experimental choice to elicit values; the
choice of a move to a new area. I have two concerns. First, respondents were not given a
chance to say that they would not move, i.e., they are not in the market. Second, I do not
know what economic concept of value they estimated. Some of the use value captured in
a hedonic model would be included, and perhaps some recreation value of being closer to
higher quality water bodies. I think that some nonuse value might be captured through
the use of a higher cost of living as the payment vehicle. It does not appear that this
framing of the valuation question captured all of respondents' recreation and nonuse
values. This leaves the question of how much of national benefits are captured in the
Viscusi study and how can other values estimated be included in the calculation of
aggregate benefits without encountering substantial double-counting problems.
Internet Surveys
Both studies used an internet survey mode and investigated aspects of the validity of this
mode, which is appropriate given the convenience and expanding use of internet surveys.
The finding by Banzhaf that there is no difference between the value estimate between an
internet survey and a mail survey is an important contribution to the literature.
The Viscusi study considered other the effects on aspects of respondents' actual
participation in the internet survey on value estimates. No statistically-significant effects
were identified, but the internet response features considered appear to be exogenous to
valuation responses and it is not surprising that not significant effects were identified. I
think it would be more interesting to consider data on time spent reading the valuation
scenario and answering the choice questions, which may be more likely to be indicative
of the difficulty of the exercise and effort that respondents invested in answering the
valuation questions. Having said this, it is still good that the Viscusi study took these
other internet survey response features off the list of concerns for future studies.
Educating Respondents
Both studies indicated that time was taken in the administration of the survey to educate
respondents who had difficulty with the valuation tasks. Neither study fully documented
what was done to educate respondents and how this influenced value estimates. This
leaves a number of questions in my mind. Did these efforts keep people in the sample
that might otherwise not have completed the survey? Did these efforts make these people
statistically similar to other respondents in terms of their valuation responses? If
valuation responses do differ, how so?
Econometric Analysis
94
-------
Both studies are disappointing in their econometric analysis. Neither study has an
empirical specification that is linked to a theoretical model, nor both studies have
specifications that are not consistent with utility, e.g., including both bid amounts and
income as separate linear arguments.
Tests of scope in both studies focused on valuation responses that include both use and
nonuse values. I think the literature is clear that stated preference studies demonstrate
scope for use values, while the real issue is in the estimation of nonuse values. The
question is whether the use value component of the value estimate is driving the
confirmation of scope in both studies. The Banzhaf study has the potential to address this
issue by segmenting the sample to those who are not users and testing this group of
responses for sensitivity to scope.
Usefulness of Value Estimates for Policy
The effects of public policy on aquatic ecosystems are highly uncertain. Both studies
assumed this uncertainty away in the design of their studies. The Banzhaf study claimed
to address uncertainty by using two scenarios. This split design does not address
uncertainty as it simply give values for to different policy outcomes. Valuation studies
that effectively value aquatic ecosystems need to include stronger links to ecosystem
attributes in the design of valuation scenarios, and explicitly include physical and
biological uncertainties into the scenario designs.
95
-------
Summary of the Q&A Discussion Following Session III (Part 2)
Bill Mates (New Jersey Department of Environmental Protection)
Saying that he was not an economist or statistician but "might be in the position of hiring
an economist or statistician," Mr. Mates addressed the three presenters: "All of your
approaches are very well done and very persuasive, but the question I would like to ask
each of you is "Where do you think your own approach is the best, and where might you
be willing to admit that one of the other two approaches was superior?" In other words,
what circumstances would lead to one approach versus another?"
Walter Milon (University of Central Florida)
Dr. Milon responded, "Well, if I'm in EPA's Office of Water, I would probably love
Joel's [Huber] work, and I suspect that's, hopefully, what the fundamental orientation is.
If I'm at the state level and I'm worried about wetland conversion decisions and policy
choices about how we set up conversion ratios, public buyout programs, the set of
ecosystem services that we would want to protect for the public, then I think we need the
more detailed information. As laborious as it may be, I think you have to go that route. I
personally think, as was said here earlier, you need to tailor the methods to the specific
policy question."
Spencer Banzhaf (Resources for the Future)
Dr. Banzhaf said, "I would say something similar, or at least something that" gets back to
the point made at "the beginning of these comments, which is the tradeoff between
saying a lot about nothing versus saying a little bit about everything. That's really the
tradeoff, so if it's a very specific policy question that one has or if shedding light on one
specific region is enough to address the question, then that would be the way to go, but if
you went. . . big . . . you could answer a host of questions."
Joel Huber (Duke University)
Dr. Huber added, "Basically, many of these signals are political ones, and for that we
need the details. As an analogy, some of us do vote for President on the basis of what our
party is, but most of us think about the individuals, and to assume that people do
otherwise would be wrong. So, at best, the approach that we have is a good first pass, but
it abstracts from everything that most of us hold dear, so in no sense am I saying this is
true. The utilities are a fiction anyway, but ours are true fictions—theirs are partial
fictions." (laughter)
Nancy Bockstael (University of Maryland)
Addressing her comment to Joel Huber, Dr. Bockstael asked, "Did you at any point try to
get at, through focus groups or anything else, what people are thinking when they're
answering these questions? I ask that because I can imagine that people—well, neither
cost of living nor environmental quality drops from the sky. Presumably, people (some
people, at least) think about a process by which some areas become higher in
96
-------
environmental quality, some areas become higher in cost of living or whatever. Are they
reading more into these questions and voting for more than they might get?
Joel Huber
Dr. Huber replied, "Right—there's the tendency to try to sort of make sense of it. The
reason I actually asked you all to make the choice was it's about as deep as that—and you
found you could make the choice, and if it were a real choice, could you make it?—yes,
you probably could. Would you make the same choice every time?—no, probably not.
Are you affected by anything around you?—yes. And then the question we asked: Is
there any stability to what comes out?—and the answer is yes. That was the hard part,
and it took a number of years to get it right. So, there's not much there—but it's enough.
Nancy Bockstael
Dr. Bockstael countered, "I have to ask: Is stability a good thing here? I don't know."
Joel Huber
Dr. Huber expounded, "Well, let me go back to the political issue. Part of the reason we
have this is because we need to value things, and the solution I give is not a great
solution, which is start with 50/50, but it does solve the problem of stopping anyone from
entering bias into the mix. I'm a researcher—I can make the thing very biased—and this
stops that, eliminates that."
Kerry Smith (North Carolina State University)
Dr. Smith said, "One of the issues that separates, invariably, economists and ecologists
when they try to look at problems is that ecologists typically apply the risk at a spatial
dimension. They're always grounded in a location and the characteristics of that location
and a configuration, as Geof (Heal) said, of services that come as a consequence of those
characteristics and resources and so forth. One of the advantages of your approach, Joel
(Huber), from the perspective of the EPA's Office of Water, is that it isn 7 that way—it's
much more compatible with the way in which economists like to think of things—away
from space, away from locations, and you can abstract from all of that and get to a
market, even though you don't define where it is. Do you see in the experience that
you've had any hope that we could get to the point where we could do that with
ecological services? I'm not sure, so I'm wondering—based on your experience not only
in this study but in other studies, is that something we should aspire to" or not?
Joel Huber
Dr. Huber responded, ". . . the value for life, which we use. It's a number, and there's
been some agreement on it, and it's very useful. Is it the value for life?—absolutely not.
Does it apply to each person?—absolutely not. Is it useful for policy?—yes. Policy is
way worse off if they don't have some number. What bothers me, and the reason I'm
willing to put in as much as we put into this is: You've got to have a consistent number
out there if you want to do consistent policy. And the number is pretty good. It's not
97
-------
truth, but it's pretty good, and it's stable, and it will resist other people trying to say:
Let's try a different way of getting at it. So, in that sense, it's a reasonable way to deal
with this. But, for most cases, it would be a first pass, or if you have to do it quickly, you
could use it."
SpencerBanzhaf
Dr. Banzhaf stated, "One of the other tradeoffs between the specific versus the generic
approaches is what people are bringing to the task if it's a stated preference model.
Things really are embedded in nature in various places. ... In our case, it really is true in
the Adirondacks that if you change water quality, you're going to change some other
parts of the ecosystem, so these things really are embedded. We found that if we left that
out and people were educated too well in high school, they embedded it themselves. We
could, in some ways, make a concession to that, since it was really true, and it's actually
easier to value a multiple than to try to divide things up and separate them piece by piece.
What you lose when you go to the generic approach is the ability to control for that. . ."
Stephen Swallow (University of Rhode Island)
Responding first to Dr. Bockstael's statement about "whether environmental quality
drops out of the sky," Dr. Swallow asserted, "In the Adirondacks, it does." (laughter) He
continued, "Actually, I like the validity of the checks you did on the plausibility of that,
but it still disturbs me that you're saying it's an unrealistic policy ... On one hand I'm
willing to go for that as a practical matter of the dirty work of policy analysis. On the
other hand, I wonder if we could explore it even further, although you explored it pretty
well."
Addressing a different issue, Dr. Swallow said, "This morning Nancy (Bockstael) said a
whole lot of things that were absolutely right about the income point and evaluating . .
.being careful about. . . welfare analysis." He said he was "encouraged with the sessions
today to see that we're not getting caught entirely in what could become intellectual
paralysis. We need to get some answers, and maybe some number is better than no
number, and sometimes you need to check and be careful about what that number is. I
like what Joel's doing because it is a step forward on what's really a dirty problem—
when you get into policy" on an international scale or on a small, local scale, "you find
out that when they get some information from several of these approaches, there's a lot of
value to that information ..."
"My final comment is that we've talked a bunch today about production functions and
linking production functions, and I think that we've forgotten one type of production
function . . . household production. Looking back at Joe Herriges' presentation on the
Iowa lake water quality, I think that the trip behavior that people were exhibiting would
have accounted for how the lakes interact with their household production. But, I wonder
whether in some of the stated preference studies we focus a little too heavily on the
98
-------
production from the ecology side of "How do we get from water quality to recreation
days?" Yet, there's still the respondent who has his or her own production function that
we haven't necessarily tried to start to quantify."
END OF SESSION III (Part 2) Q&A
99
-------
Valuation of Ecological Benefits: Improving the Science
Behind Policy Decisions
PROCEEDINGS OF
SESSION IV: TALES OF OTTERS, EAGLES, AND OWLS:
VALUING WILDLIFE HEALTH AND BIODIVERSITY
A WORKSHOP SPONSORED BY THE U.S. ENVIRONMENTAL PROTECTION
AGENCY'S NATIONAL CENTER FOR ENVIRONMENTAL ECONOMICS (NCEE)
AND NATIONAL CENTER FOR ENVIRONMENTAL RESEARCH (NCER)
October 26-27, 2004
Wyndham Washington Hotel
Washington, DC
Prepared by Alpha-Gamma Technologies, Inc.
4700 Falls of Neuse Road, Suite 350, Raleigh, NC 27609
-------
ACKNOWLEDGEMENTS
This report has been prepared by Alpha-Gamma Technologies, Inc. with funding from
the National Center for Environmental Economics (NCEE). Alpha-Gamma wishes to
thank NCEE's Cynthia Morgan and the Project Officer, Ronald Wiley, for their guidance
and assistance throughout this project.
DISCLAIMER
These proceedings are being distributed in the interest of increasing public understanding
and knowledge of the issues discussed at the workshop and have been prepared
independently of the workshop. Although the proceedings have been funded in part by
the United States Environmental Protection Agency under Contract No. 68-W-01-055 to
Alpha-Gamma Technologies, Inc., the contents of this document may not necessarily
reflect the views of the Agency and no official endorsement should be inferred.
11
-------
TABLE OF CONTENTS
Session IV: Tales of Otters, Eagles, and Owls: Valuing Wildlife Health and
Biodiversity
Contingent Valuation for Ecological and Noncancer Effects Within an
Integrated Human Health and Ecological Risk Assessment Model
James Hammitt, and Katherine von Stackelberg, Harvard University 1
Joint Determination in a General Equilibrium Ecology/Economy Model
David Finnoff, and John Tschirhart, University of Wyoming 25
Integrated Modeling and Ecological Valuation
David S. Brookshire, and Janie Chermak, University of New Mexico;
Juliet Stromberg, Arizona State University; Arriana Brand, Colorado State
University; David Goodrich, Arizona Research Service; Thomas Maddock
III, and Steven Stewart, University of Arizona 54
Discussant
R. David Simpson, U.S EPA, National Center for Environmental
Economics 69
Discussant
Juha Siikamaki, Resources for the Future 75
Summary of Q&A Discussion Following Session IV 79
in
-------
Contingent Valuation for Ecological and Noncancer Effects within an
Integrated Human Health and Ecological Risk Assessment Model
James Ham mitt and Katherine von Stackelberg
Harvard Center for Risk Analysis, 718 Huntington Avenue, Boston, MA 02115
Objectives:
The objective of this research is to contribute to the understanding of practical and credible
approaches for estimating the benefits and costs of environmental policies and to improve
decisionmaking regarding environmental issues. Our approach is to develop an integrated
human health and ecological risk model using data from a case study, which also incorporates
economic information from two contingent valuation surveys. The case study focuses on
potential human health and ecological receptor exposure to polychlorinated biphenyl (PCB)
compounds via fish ingestion. The risk model integrates the results of two contingent valuation
(CV) surveys to quantify the benefit of potential risk reductions under assumed exposure
conditions using a publicly available database. The integrated model uses the results of a web-
based contingent valuation questionnaire to estimate willingness-to-pay of the general public to
reduce potential risks associated with exposure to PCBs. These risks include reproductive
effects in birds and mammals and developmental effects in children exposed in utero. The
integrated risk model can be used to evaluate the economic role ecological and noncancer human
health outcomes play under specific exposure conditions. The contingent valuation surveys are
designed to inform the growing literature on the value individuals place on the ecological and
noncancer benefits of risk reductions.
During the past year, we focused on working through methodological issues related to survey
development. We also conducted an extensive literature search and evaluated this literature on
the potential effects of exposure to PCBs in order to develop dose response models for
developmental noncancer effects in humans and reproductive effects to ecological receptors.
These dose response models provide the basis for the relevant endpoints used in survey
development. Finally, we developed an Excel-based two-dimensional Monte Carlo modeling
framework to predict dietary exposures to human and ecological receptors. We integrated the
dietary exposure model with the dose response models to obtain probabilistic estimates of
potential risks.
Potential Effects of PCBs
We have conducted an extensive literature search on the potential effects of PCBs in both
humans and animals, focusing specifically on potential developmental effects in humans and
reproductive effects in wildlife. This information is used to develop the specific endpoints for
valuation within the CV survey, and to develop dose-response models for the integrated risk
assessment.
1
-------
Ecological Endpoints
Typical responses to PCB exposure in animals include wasting syndrome, hepatotoxicity,
immunotoxicity, neurotoxicity, reproductive and developmental effects, gastrointestinal effects,
respiratory effects, dermal toxicity, and mutagenic and carcinogenic effects. Some of these
effects are manifested through endocrine disruption.
PCBs are typically present in the environment as complex mixtures. These mixtures consist of
discrete PCB molecules that are individually referred to as PCB congeners. PCB congeners are
often introduced into the environment as commercial mixtures known as Aroclors. PCB toxicity
varies significantly among different congeners and is dependent on a number of factors. Two
significant factors relate to the chemical structure of the PCB congener, including the degree of
chlorination and the position of the chlorines on the biphenyl structure (Safe et al. , 1985a). In
general, higher chlorine content typically results in higher toxicity, and PCB congeners that are
chlorinated in the ortho position are typically less toxic than congeners chlorinated in the meta
and para positions. Metabolic activation is believed to be the major process contributing to PCB
toxicity.
Reproductive effects tend to be the most sensitive endpoint for animals exposed to PCBs.
Indeed, toxicity studies in vertebrates indicate a relationship between PCB exposure, as
demonstrated by AHH induction, and functions that are mediated by the endocrine system, such
as reproductive success. A possible explanation for the relationship between AHH activity and
reproductive success may be due to a potential interference from the P450-dependent MFO with
the ability of this class of P450 proteins to regulate sex steroids. In fact, the induction of
cytochrome P450 isozymes from PCB exposure has been shown to alter patterns of steroid
metabolism (Spies etal., 1990).
Historically, the most common approach for assessing the ecological impact of PCBs has
involved estimating exposure and effects in terms of totals or Aroclor mixtures. It is important
to note that, since different PCB congeners may be metabolized at different rates through various
enzymatic mechanisms, when subjected to processes of environmental degradation and mixing,
the identity of Aroclor mixtures is altered (McFarland and Clarke, 1989). Therefore, depending
on the extent of breakdown, the environmental composition of PCBs may be significantly
different from the original Aroclor mixture. Furthermore, commercial Aroclor mixtures used in
laboratory toxicity studies may not represent true environmental exposure to this Aroclor. Thus,
there are considerable uncertainties associated with estimating the ecological effects of PCBs in
terms of total PCBs or Aroclors.
Ecological risk assessments follow an established framework in which there are assessment and
measurement endpoints (EPA, 1993). Assessment endpoints represent that which is being
protected, for example, protection and sustainability of wildlife populations. There are one or
more measurement endpoints, which are the specific ways in which impacts on the assessment
endpoints will be evaluated. For example, one of the associated measurement endpoints for that
assessment endpoint might include comparing predicted doses to the selected species with doses
from the toxicological literature associated with specific effects. Assessment endpoints are
2
-------
broadly defined while measurements endpoints are the specific analyses that will quantify
potential impacts.
There are two distinct ways in which potential impacts to an ecosystem can be evaluated within a
risk assessment context. The first focuses the analysis on a single species designed to represent
high-end exposure and sensitivity. A set of representative receptors is selected that serve as
proxies for the many species that the ecosystem supports. Potential impacts are evaluated by
developing dose-response models, generally from laboratory data, to predict outcomes. The very
simplest analysis involves developing a toxicity reference value (TRV) against which to compare
exposures at the site. This deterministic analysis can be expanded to include a joint probability
model that quantifies the probability of an increasing magnitude of effect using the dose-
response model for a single species (e.g., reduction in fecundity), or one can model the
probability of exceeding a threshold value. Under this representative receptor approach, a
valuation for a single "high-profile" species will implicitly value those aspects of the ecosystem
that support the species (Loomis and White, 1996). Management actions are designed to reduce
risks for the presumed highest risk species. The corresponding valuation asks respondents about
their willingness to pay to reduce the probability of an effect on a single species.
The second approach is slightly different. Rather than relying on a single dose-response
relationship for one species, the analysis relies on species sensitivity distributions. These
distributions quantify the probability of the proportion of species that will be affected (e.g., there
is a 20% probability that 80% of the species will experience adverse reproductive effects).
Under this approach, the analysis does not focus on one particular species but rather considers
the probability of impacting multiple species. Both approaches are used in this analysis (and
both kinds of endpoints valued in the survey).
In both cases, the exposure model incorporates a probabilistic bioaccumulation model to describe
uptake of PCBs into the aquatic food web based on a model developed for the Hudson River
RI/FS (EPA, 2000a; 2000b). Both benthic and pelagic invertebrates, aquatic vegetation and fish
are consumed by ecological receptors. Risks are described across an increasing magnitude of
effect (e.g., there is a 50% chance of an 80% reduction in fecundity) or across an increasing
percentage of species (e.g., there is a 50% chance that 80% of the species will experience
adverse reproductive effects).
The individual species of concern in this analysis is the bald eagle (Haliaeetus leucocephalus).
We selected this receptor for modeling because:
• It is one of the most important receptors in the Hudson River RI/FS, which provides the
exposure estimates;
• It consumes a variety of fish and fish-eating organisms;
• It only produces one or two nestlings per year;
• It is a threatened species.
Several field studies were identified that examined the effects of PCBs in eggs of bald eagles, but
not dietary doses. Clark et al. (1998) presented information on concentrations of total PCBs
(range = 20 to 54 mg/kg egg) and TEQs in eggs from two sites in New Jersey where reproductive
3
-------
failures have occurred, but the data could not be used to develop dose-response relationships.
Studies by Wiemeyer et al. (1984, 1993) reported adverse effects on mean 5-year production in
bald eagle with egg concentrations greater than 3.0 mg PCBs/kg egg. Wiemeyer et al. (1993)
studied bald eagle production over a long time period (i.e., 5- year intervals from 1969 through
1984), and examined production rates in the field. However, significant intercorrelation of many
contaminants made it difficult to determine which contaminants caused the adverse effects
(Wiemeyer, 1993), thus, it is not possible to estimate individual dose-response relationships.
We are using a laboratory study by Dahlgren et al. (1972) as the basis for the individual dose-
response modeling in this analysis. These authors found significantly reduced (p<0.01) egg
production by hens that had been fed Aroclor 1254 for a period of 16 weeks. Egg production by
hens fed PCBs at the lowest observed adverse effect level was 32-97% that of control hens. The
Aroclor 1254 was administered weekly in capsules into the esophagus. A Generalized Linear
Modeling (GLM) framework is used with a log link function and Poisson error distribution to
estimate the dose-response relationship (Moore et al., 1999; EPA, 2000b).
The species sensitivity distribution (SSD) approach uses the distribution of effects concentrations
for all species. Just as typical dose-response curves can be used to estimate the probability of
effects for an individual species, the SSD can be used to estimate the probability of effects on a
species or the probability of effects across species. We are relying on an SSD developed by
Suter (2003), based on the survival and development of avian embryos and chicks. These effects
were chosen because of data availability, comparability among studies and the clear relevance of
reproductive success to avian populations. That report focuses on the potential effects of the
dioxin-like congeners.
SSDs may be used quantitatively to estimate the proportion of a taxon (e.g., herons), trophic
group (e.g., piscivorous birds) or community that will be affected by an exposure (Suter, 2003).
This is equivalent to using a conventional dose-response function to estimate the proportion of a
population that will be affected. It requires fitting some function to the SSD so that, as in other
exposure-response models, the response can be estimated from the exposure level. The most
common functions are the log normal or its linearized version the log probit and the log logistic
or its linearized version the log logit, depending on how the outcomes are expressed (e.g.,
continuous versus dichotomous). The use of tested species to represent communities relies on
the assumption that the tested species are an unbiased sample of the community. Test species are
not chosen randomly, but, since species sensitivities are not known prior to testing, there is no
reason to expect that the selection is biased.
Human Health Endpoints
The weight-of-evidence for a relationship between in utero polychlorinated biphenyl (PCB)
exposure and developmental outcomes has been well established and continues to grow (Schantz
et al., 2003). However, as with most epidemiological studies, discrepancies exist among
measures of exposure and the strength of the relationships between the measures of exposure and
developmental outcomes. Some of those discrepancies are attributable to differences in
analytical methods, particularly in older studies (Longnecker et al., 2003) that had higher
detection levels and less sophisticated quantitation techniques.
4
-------
The primary difficulty in quantifying the relationship between dietary levels of PCBs associated
with developmental effects based on epidemiological studies is in the lack of an association
between exposure metrics and outcomes. Ingestion of food is likely the most significant
exposure pathway (Longnecker et al., 2003; Laden et al., 2000). However, the studies that show
significant associations between PCB exposure expressed as cord blood, maternal serum or
breast milk concentrations and developmental outcomes in children (ranging in age from birth to
11 years) typically show little or no correlation between those metrics and seafood or other
dietary item consumption (Schell et al., 2001;).
It is not a goal of the analysis to develop a definitive dose response model for potential
developmental effects. Therefore, we selected data from one cohort specifically which has been
well-documented in the literature, follows a cohort whose exposure is specifically tied to fish
ingestion, and for which the evidence of in utero exposure is greatest. In addition, this cohort
has been followed for 11 years with documented effects still at this age, as opposed to other
studies which only have a few years worth of followup.
The Michigan Cohort: this cohort was recruited through four maternity hospitals in western
Michigan from 1980 to 1981. Two hundred and forty two mothers who had at consumed more
than 12 kg of Lake Michigan fish during the previous six years and 71 mothers who had not
eaten any Lake Michigan fish participated in the study. The authors assessed prenatal exposure
using umbilical cord serum collected at delivery, and maternal serum and breast milk collected
shortly after delivery. Testing thus far has been conducted on the newborns, at 4 years, and at 11
years.
Newborn testing was conducted on 242 exposed and 71 control infants. Behavioral outcomes
were assessed using the Brazelton neonatal Scale (NBAS), which showed that the most highly
exposed infants were more likely than controls to be classified as "worrisome" (Jacobson et al.,
1984). The four-year followup collected data from two separate visits - the first at four years
and the second three months later. During the first visit, 236 exposed and 87 unexposed children
were evaluated using the McCarthy Scales of Children's Abilities, while the second visit
involved a series of reaction time tests (Jacobson et al., 1990). Both visits involved extensive
discussions with the mother, including completing the Peabody Picture Vocabulary Test-Revised
and Buss and Plomin Emotionality Activity Sociability Temperament Survey for Children. The
three ratings were transformed into standard scores and summed to provide a composite measure
of activity, what was then standardized to a mean of zero and a standard deviation of one. The
11-year sample evaluated 178 children tested using the Wechsler Intelligence Scale for Children-
Revised (WISC-R).
We are relying on the data presented in Jacobson et al. (2000), which presents a linear
relationship between lipid-normalized breast milk concentration of PCBs and outcomes.
Survey Development
We completed a pre-test version of the survey (see Appendix A). The survey is divided into two
parts. In the first part, half the respondents see questions related to ecological effects first, and
5
-------
the other half see questions with human health effects. The second part asks for the total amount
that a household would be willing to pay for the combined health and ecological benefits of PCB
removal.
Environmental effects can be broadly categorized in one of two ways: a small probability of a
fairly dramatic outcome (e.g., risk of developing a serious illness or cancer food poisoning, etc.),
or a relatively large probability of a very small effect. There are a number of studies that have
evaluated willingness to pay of the first category. This study is designed to evaluate willingness
to pay for a very small effect (in humans) that occurs with a fairly large probability (50% chance
if exposed). The ecological effect (e.g., reproductive impairment) occurs with a relatively high
probability and is considered a large effect, but of course does not impact humans directly. For
that outcome, we are interested in how people perceive environmental threats to ecological
resources, what they might be willing to pay to reduce that threat, and how those results can be
incorporated into a specific regulatory context - namely, risk assessment.
The human health component of the survey asks respondents about two potential developmental
endpoints associated with PCB exposures: a probability of a 6 point reduction in IQ and a
probability of a 7-month deficit in reading comprehension. The survey asks about a 50% risk of
these endpoints decreasing to either 10% or 25%. The ecological endpoints include a probability
of reproductive effects in eagles such that the ability of the population to sustain itself would be
severely compromised, and a probability of an effect on 25% of the species (using a species
sensitivity distribution). The ecological endpoints correspond to the two different management
alternatives that are currently used in typical ecological risk assessments. The risk of adverse
effects in eagles is 20% decreasing to 10% or 5%.
Willingness to pay is elicited using a double-bounded dichotomous choice format. Respondents
are presented with an initial bid randomized from a bid vector ranging from $25 to $400. If the
respondents agree to the initial bid, they are presented with a bid that is double the first bid (if
they agree to $400 initially, then they are asked if they would be willing to pay at $800). If
respondents do not agree to the initial bid, then they are presented with a bid that is half as much
($10 if they did not agree to $25 initially). The survey is currently undergoing pretest and
depending on those results, the bid vector may be adjusted for the final survey.
We are interested in evaluating differences in willingness to pay values and predictors for
ecological versus noncancer outcomes, both in relative and absolute terms. Respondents are
asked about one set of effects (either human or ecological), and then asked the total amount they
would be willing to pay for both sets of risk reductions. Because of potential embedding and
ordering effects, there may inconsistencies in responses. Ideally, we would like to have the same
respondents provide values for both sets of endpoints. However, it is difficult for people to
disaggregate their willingness to pay. Respondents may not be willing to pay any additional
amount for an additional benefit (e.g., under the first willingness to pay question, the PCBs will
have been removed, therefore, both sets of benefits will occur regardless of any additional money
that is spent). We could split the survey and administer it to half the respondents, or take the
entire survey and administer it to all respondents. We chose the latter design, since ideally we
are interested in evaluating risks of small, noncancer effects in humans versus ecological effects
in addition to determining willingness to pay for each endpoint separately.
6
-------
We have conducted several informal focus groups and discovered that among respondents who
are not involved with environmental issues in some way (e.g., work for an environmental
company or are involved in local or national environmental groups), there was greater skepticism
about the potential effects of PCBs on ecological receptors relative to human receptors. That is,
people readily accepted the concept that PCBs could cause developmental delays in children
exposed in utero, but required greater justification for potential effects to ecological receptors,
and, additionally, required greater assurances that the proposed cleanup would actually achieve
the stated risk reduction.
Current Status
The survey is programmed and undergoing final edits for pretest. We anticipate going into
pretest the week of October 18 and administering the final survey two weeks after that. The
current survey is attached as Appendix A.
Risk Modeling
The risk models use the data presented earlier to develop dose-response relationships, which are
combined with exposure data from the Hudson River case study to predict the probability of an
increasing magnitude of effect.
Human Health Model
The dose of PCBs via fish ingestion is most simply modeled using a first-order uptake model
assuming distributions for percent absorbed, lipid content, and elimination half-life.
First order uptake is given as:
f*ADD*t1/2
ln(2)
where:
Cms = concentration in maternal serum (mg/kg)
f = fraction absorbed (0.5 - 0.9)
ADD = average daily dose (mg/kg-day)
t 2 = elimination half life (considered constant during lifetime) 7.5 years
Because PCBs are lipophilic and partition into the organic fraction of the environmental media
they are in, the concentration of PCBs in breast milk is related to the concentration in serum
through lipid content. Thus, the predictions for serum concentration are normalized for the
overall tissue lipid content in pregnant women (Kopp-Hoolihan, 1999)
7
-------
c„=%- (2)
Lb
Cln = mg PCB per kg lipid
Cms = concentration in maternal serum (mg/kg)
Lb = lipid content of all tissues (fraction body fat)
The maternal average daily dose (ADD) is given as:
IR*C *
ADD = — (30
BW
ADD = average daily dose from fish consumption (mg/kg-day)
IR = ingestion rate (kg/day)
Cfish = concentration in fish (mg/kg)
BW = body weight (kg)
TABLE 1: Modeling Parameters
Parameter
Units
Variable or
Uncertain
Distribution
Reference
Ingestion rate
kg/day
Variable and
uncertain
From Hudson
River RI/FS
EPA,
Body weight
kg
variable
Normal (64.8,
7.8)
Kopp-Hoolihan,
1999
PCB
mg/kg
uncertain
Lognormal
modeled
concentration in
(depends on
fish
scenario)
Fraction absorbed
fraction
uncertain
Uniform (0.5,
0.9)
assumption
t/2 (half-life)
1/year
uncertain
7.5
WHO
Total body fat
fraction
variable
Normal (31, 5.5)
Kopp-Hoolihan,
1999
The model uses two-dimensional Latin Hypercube to sample from each of the input distributions.
It is iterative in that it first fixes all uncertain parameters, then runs an inner "variability" loop in
which it samples from the variable distributions, and then returns to the outer loop to run another
set of uncertain parameters.
Ecological Risk Model
The first ecological risk model quantifies the probability that a particular fraction of the eagle
population will experience a particular effect (e.g., 10% probability that there will be a 50%
reduction in fecundity in the eagle population). The potential for population-level effects is
evaluated by convolving the dose-response functions (based on Dahlgren et al., 1972) with
8
-------
cumulative distributions of exposure estimated using data from the Hudson River (EPA, 2000b).
The second risk model quantifies the probability that a fraction of the species will experience a
particular effects (e.g., 10% probability that 50% of the species will experience egg death).
In both cases, the dose-response functions are obtained by fitting a probit or logit model to
appropriate toxicity data. The probit model takes the following form:
log d
P(d) =
V^r log dx
(4)
where:
d50 = median or geometric mean; dose at which 50% of the population responds
ag = geometric standard deviation; a measure of dispersion of the responses
The inverse of the cumulative probability of effect corresponding to these concentrations is
obtained by using the NORMSINV function in Excel. This function provides the inverse of the
standard normal cumulative distribution assuming a mean of 0 and a standard deviation of 1. The
log-transformed concentration is thus linearly related to the inverse of the cumulative probability
of effect. These points are then used to derive a linear relationship that follows the general form:
®-\Y) = j0o + j0iX (5)
where:
~' = inverse of the normal cumulative probability (z) function (NORMSINV)
Y = probability of response
(3o = intercept (-dso/Og)
|3i = slope of the line (l/og)
X = log (concentration in mg/kg)
The following procedure is used to compare the cumulative exposure distributions with the dose
response curves. First, Monte Carlo exposure models are used to generate the cumulative
frequency of predicted dietary doses for each receptor. Output concentrations are log-
transformed, and the associated cumulative frequencies, expressed as fractions, are transformed
by the inverse of the normal cumulative distribution. The log-transformed Monte Carlo
concentrations and the transformed cumulative frequencies yield straight lines when plotted
against each other. The parameters of these regressions are used to obtain the cumulative
frequency for the specified doses in the dose-response curves from the literature. The resulting
curves can then be compared directly by plotting the probability of exccedence on the y-axis
(obtained from the cumulative distribution of exposure) and the percent reduction in fecundity on
the x-axis (obtained from the dose-response curves from the literature).
This estimate takes the following form substituting concentration c for dose d\
poo
R = P(c)
-------
where:
(|) = the lognormal probability density function
R = risk
At each log concentration, a probability of effect is estimated from the probit model and a
frequency of occurrence is estimated from the lognormally distributed body burdens. This model
has been developed and applied in several site-specific applications (EPA, 2000b; Moore et al.,
1999).
Next Steps
We plan to derive willingness to pay estimates for the specified outcomes from the survey data.
We also plan to incorporate the willingness to pay estimates into the risk assessment model to
quantify the benefits of risk reductions using actual data from the Hudson River RI/FS.
Management decisions have already been made for this site, therefore, it is used strictly for
demonstration purposes.
We will also evaluate differences in willingness to pay between the human health and ecological
endpoints. The risk model can also be used to estimate the probability of developing cancer and
these results compared to the noncancer and ecological results. The exposure and risk models
are largely complete and were used to derive the specific risk reductions that are the focus of the
contingent valuation survey.
References
Andrle, R.F. and J.R. Carroll (Editors). 1988. The Atlas of Breeding Birds in New York State.
Cornell University Press. Ithaca, New York. 551 pp.
Clark, K. E., L. J. Niles, and W. Stansley. 1998. Environmental contaminants associated with
reproductive failure in bald eagle (Haliaeetus leucocephalus) eggs in New Jersey. Bulletin of
Environmental Contamination and Toxicology. 61:247-254.
Dunning, J.B. Jr. 1993. CRC Handbook of Avian Body Masses. CRC Press. Ann Arbor
Michigan. 371 pp.
Elliott, J.E., R.J. Norstrom, A. Lorenzen, L.E. Hart, H. Philibert, S.W. Kennedy, J.J. Stegeman,
G.D. Bellward, and K.M. Cheng. 1996. Biological effects of poly chlorinated bibenzo-p-dioxins,
dibenzofurans, and biphenyls in bald eagle (Haliaeetus leucocephalus) chicks. Environmental
Toxicology and Chemistry Vol. l(5):782-793.
(EPA) United States Environmental Protection Agency. 1993a. Wildlife Exposure Factors
Handbook. Volume I of II. USEPA Office of Research and Development. Washington, D.C.
EPA/600/R-93/187a.
10
-------
(EPA) United States Environmental Protection Agency. 1993b. Wildlife Exposure Factors
Handbook. Volume II of II. USEPA Office of Research and Development. Washington, D.C.
EPA/600/R-93/187b.
(EPA) United States Environmental Protection Agency. 2000b. Further Site Characterization
and Analysis, Volume 2E - Revised Baseline Ecological Risk Assessment (ERA), Hudson River
PCBs Reassessment RI/FS. Prepared for USEPA Region 2 and USACE, Kansas City District by
Menzie-Cura & Associates, Inc. and TAMS Consultants, Inc., August 1999.
Jacobson, J.L., Jacobson, S.W., Schwartz, P.M., Fein, G.G., and J.K. Dowler. 1984. Prenatal
exposure to an environmental toxin: A test of the multiple effects model. Developmental
Pyschology 20(4):523-532.
Jacobson, S.W., Fein, G.G., Jacobson, J.L., Schwartz, P.M., and J.K. Dowler. 1985. The effect
of intrauterine PCB exposure on visual recognition memory. ChildDev 56:853-860.
Jacobson, J.L., S.W. Jacobson and H.E.B. Humphrey. 1990a. Effects of in utero exposure to
polychlorinated-biphenyls and related contaminants on cognitive functioning in young children.
J. Pediatr 116(l):38-45.
Jacobson, J.L., S.W. Jacobson and H.E.B. Humphrey. 1990b. Effects of exposure to PCBs and
related compounds on growth and activity in children. Neurotoxicol Teratol 12:319-326.
Jacobson, J.L., S.W. Jacobson. 1996a. Intellectual impairment in children exposed to
polychlorinated biphenyls in utero. N Engl J Med 335(11):783-789.
Jacobson, J.L., S.W. Jacobson. 1996b. Sources and implications of interstudy and interindividual
variability in the developmental neurotoxicity of PCBs. Neurotoxicol Teratol 18(3):257-264.
Jacobson, J.L., S.W. Jacobson. 1997. Evidence of PCBs as neurodevelopmental toxicants in
humans. Neurotoxicology 18(2):415-424.
Jacobson, J.L., Janisse, J., Banerjee, M., Jester, J., Jacobson, S.W., and J.W. Ager. 2002. A
benchmark dose analysis of prenatal exposure to polychlorinated biphenyls. Environ Health
Perspect 110(4):393-398.
Kopp-Hoolihan, L.E., M.D. van Loan, W.W. Wong, and J.C. King. 1999. Far mass deposition
during pregnancy using a four-component model. JAppl/Tzysiol 87(1): 196-202.
Laden, F., Neas., L.M., Spiegelman, D., Hankinson, S.E., Willett, W.C., Ireland, K., Wolff, M.S.
and D.J. Hunter. 1999. Predictors of plasma concentrations of DDE and PCBs in a group of
U.S. women. Environmental Health Perspectives 109(1):
Longnecker, M.P., Wolff, M.S., Gladen, B.C., Brock, J.W., Grandjean, P., Jacobson, J.L.,
Korrick, S.A., Rogan, W.J., Weisglas-Kuperus, N., Hertz-Picciotto, I., Ayotte, P., Stewart, P.,
Winneke, G., Charles, M.J., Jacobson, S.W., Dewailly, E., Boersma, E.R., Altshul, L.M.,
11
-------
Heinzow, B., Pagano, J.J. and A.A. Jensen. 2003. Comparison of polychlorinated biphenyl
levels across studies of human neurodevelopment. Environ Health Perspect 11 l(l):65-70.
Moore, D.R.J., Sample, B.E., Suter, G.W., Parkhurst, B.R. and R.S. Teed. 1999. A probabilstic
risk assessment of the effects of methylmercury and PCBs on mink and kingfishers along East
Fork Poplar Creek, Oak Ridge, Tennessee, USA. Environ. Tox. Chem. 18(12):2941-2953.
Schantz, S.L., Wiholm, J.J. and D.C. Rice. 2003. Effects of PCB exposure on
neuropsychological function in children. Environ Health Perspect 111(3):357-376.
Schell, J.D. Jr., Budinsky, R.A., and M.J. Wernke. 2001. PCBs and neurodevelopmental effects
in Michigan children: An evaluation of exposure and dose characterization. Reg Tox Pharm
33:300-312.
Todd, C. S., Young, L. S., Owen, R.B., Jr. et al. 1982. Food habits of bald eagles in Maine. J
WildlManage 46: 636-645.
Wiemeyer, S.N., T.G. Lamont, C.M. Bunck, C.R. Sindelar, F.J. Gramlich, J.D. Fraser, andM.A.
Byrd. 1984. Organochlorine pesticide, polychlorobiphenyl, and mercury residues in bald eagle
eggs- 1969-79- and their relationships to shell thinning and reproduction. Arch Environ Contam
Toxicol 13: 529-549.
Wiemeyer, S.N., C.M. Bunck, and C.J. Stafford. 1993. Environmental contaminants in bald
eagle eggs - 1980-84- and further interpretations of relationships to productivity and shell
thickness. Arch Environ Contam Toxicol 24:213-227.
12
-------
PCB Exposure Valuation Survey
September 2004
- Study Details -
Note: This page may be removed when the questionnaire is sent to the client. However, it must
exist in the version sent to Operations.
SNO
7960
Survey Name
Willingness to Pay for Noncancer
Developmental and Ecological Effects
Pretest
Client Name
Harvard University
Great Plains Project Number
K0499
Project Director Name
Stefan Subias
Team/Area Name
TM Dennis
Sample Criteria
General population adults
Samvar
Standard demos only.
Specified Pre-coding Required
No
Timing Template Required
Yes
Multi-Media
Images
Incentive
No
Disposition Information
(Provide exact descriptions
with reference to question
numbers and answer list
responses for all groups that
daily counts are desired)
Note: The change request log can be deleted, if you do not require it.
Note:
Change Request Log
(Operations Please Disregard)
Do not change Question numbers after Version 1; to add new question, use alpha
characters (e.g., 3a, 3b, 3c)
Author
Ver-
sion
Description of Change
(Q#, plus change)
Approval
Name
Date
Apprv'd
Com-
pleted
(Y/N)
13
-------
Willingness to Pay for Noncancer Developmental and Ecological Effects Pretest
September 2004
- Questionnaire -
Note to Scripter: Randomize sections B/C and D so that half see B/C first and half see D
first.
[Display]
We are conducting this survey to get your opinion on issues such as education, crime, and the
environment facing people in your state. The study will provide information so that State policy makers
can understand how people like you feel about these issues.
A. Introductory Questions
Note to Scripter: I converted the 3-point scales in A1 and A2 into 5-point ones—this is hard
TO SEE IN THE TRACKED CHANGES.
[Grid - SP by Row]
A1. There are many issues that require resources facing residents in your State. Some of them
may be important for you personally and others may not. Please identify whether the listed
issue is not important, somewhat important, or very important to you personally:
Not
Somewhat
Very
Important
Important
Important
Reducing crime
Cleaning up the environment
Improving education
Protecting State waterways
Reducing State taxes
Reducing air pollution
Improving library services
Providing more security at public events
[Grid - SP by Row]
A2. Your State government must allocate financial resources among many different programs.
Below you will see a list of different programs. For each one, please indicate whether the
amount of money being spent should be reduced, stay the same, or increased, keeping in mind
that overall expenditures cannot be increased without an increase in revenue:
Reduced A
Reduced A
Stay the
Increased
Increased
Lot
Little
Same
A Little
A Lot
Public transportation in metropolitan areas
Providing homeless shelters
Protecting endangered wildlife
Increased funds for education
Building new prisons
Updating water treatment facilities
14
-------
Maintaining the court system
Increasing security around public buildings
[Display]
Every year, the State must decide how to allocate money for the State budget. Sometimes new
programs are proposed, and the State is interested in knowing how taxpayers feel about these
programs in order to decide whether they should be funded or not. Surveys like this are used to
explore how people like you feel about the various programs that the State can spend money on in the
coming year. Everyone feels differently, and it's important to hear from as many people as possible in
order to capture all the different points of view.
This survey is asking specifically about a program involving the potential effects of chemicals in the
environment on both humans and animals. The next part of this survey will provide some background
information on the situation and the potential effects of these chemicals. After that, the survey will ask
you whether you think anything should be done about the situation. Finally, we are interested in
knowing why you feel the way you do.
[Display]
Studies have shown that babies developing in the womb (fetuses) are affected later in life by some
chemicals found in fish and other foods that are eaten by their mothers. Developing fetuses are
exposed to the same things as their mothers - but because they are so small, and their organs are still
developing, even very small amounts of substances that have little or no effect on the mother can have
a big impact on a developing fetus. The effects, typically different kinds of developmental delays, can
be observed even in small infants. Scientists have studied the issue, and have determined that the
trouble is a result of being exposed to a specific chemical that is found in the sediment (dirt) of several
rivers, streams, and lakes in your State. Scientists representing the State, Federal government, and
academic institutions have spent years studying this issue. They agree that the known deposits of a
specific chemical in the riverbeds bears some responsibility in causing these reproductive effects. The
chemical is called polychlorinated biphenyls, or PCBs.
[SP]
A3. PCBs are chemicals that were developed in the early 1940's for electrical transformers and for
other industrial purposes. They were an ideal insulating fluid because they are not flammable.
Have you ever heard of PCBs?
Yes
No
Not Sure
[Display]
Up until the early 1970's, people didn't realize that PCBs could
affect fish, wildlife, and humans. Several companies that
manufactured electrical transformers, or provided other
industrial services, were located on different rivers in the State.
Some of these companies are out of business now, but in the
1940's, 50's and 60's, they were allowed to discharge PCBs
with other wastes from their manufacturing processes. Even
though there have been no new PCBs discharged into rivers in
at least 20 years, the amounts that were historically released continue to affect wildlife in the State.
PCBs are very oily and do not dissolve in water. Once they are in water, they fall to the bottom of the
river and remain in the sediment. Sediment, which is just sand and dirt at the bottom of the river, is
very stable, except when there is a big storm. As a result, there are layers and layers of sediment
containing PCBs.
15
-------
[Display]
Insert Foodweb graphic here with Audio.
As this graphic shows, insects and
shellfish living in the sediment
absorb PCBs and transfer them to
fish. Animals, including humans,
eat the fish and in this way PCBs
accumulate through the food web.
You may have heard of fish
consumption advisories in your
State. These are due to PCBs as
well as other chemicals.
The State is proposing to either
clean up or remove the
contaminated sediment from the
river to make sure that humans and
wildlife are no longer exposed. If
the sediments are not cleaned up,
they will continue to be a source of
PCBs to the system.
[Display]
Eventually, PCBs in the sediments will grow less and less due to natural causes. New, clean sediment
will deposit over the dirty sediment over a period of many years. The insects, as they continue to work
the sediment, will eventually release or use up much of what is there. Scientists using models
developed just for this system have shown that PCBs in the sediment will decrease to levels that aren't
expected to have effects on animals and humans in approximately 100 years. A clean up remedy, such
as dredging, is expected to take one year, will decrease these levels immediately after the clean up is
completed. It will still take a few years for the species to recover, but they will not be exposed to any
new PCBs during that time.
In order to pay for this cleanup, the State is proposing a one-time additional amount on next year's
state income tax. Only this one time payment is required and the money will go into a special fund just
for this purpose. There are lots of reasons why you might vote for or against such a program.
B. Questions Related to Human Receptor Exposure to Polychlorinated Biphenyls
Programming Notes:
Show "read at 7 months below grade level" or "experience a 6 point decrease in IQ as
MEASURED BY STANDARD IQ TESTS" RANDOMLY; 50% SEE ONE AND 50% SEE THE OTHER. PLEASE
CREATE a DATA-ONLY VARIABLE INDICATING WHICH STATEMENT was shown.
Show "25%" and "10%" randomly: 50% see one and 50% see the other. Please create a data-
only VARIABLE INDICATING WHICH PERCENTAGE WAS SHOWN.
16
-------
[Display]
Studies involving children exposed in utero to PCBs have shown that these children perform less well
on a variety of developmental tests. For a unit exposure, IQ can decrease by six points, the average
decrease in reading comprehension is 7 months, children perform less well on mathematical and
quantitative tests. The chemical doesn't cause the exact same effects in every child, but it does cause
some effect in every child. One specific effect that regulators are worried about is the evidence that
exposure to PCBs causes decreases in reading comprehension below levels considered normal in
school-age children. There is a 50% chance that children exposed to PCBs in this area will [read at 7
MONTHS BELOW GRADE LEVEL /EXPERIENCE A 6 POINT DECREASE IN IQ AS MEASURED BY STANDARD IQ
tests]. If the sediments are removed and the river is cleaned up, scientists estimate that the risk will
decrease to [25% /10%]. There will always be some small chance of effects because the sediments
can't be 100% cleaned up.
Note to Scripter: Randomly select the value for Cost_H from the following choices: $25,
$50, $100, $200, $400, $800; Each value should be assigned for approximately 16.7% of the
RESPONDENTS (I.E., 100/6). PLEASE CREATE A DATA-ONLY VARIABLE INDICATING WHICH VALUE WAS
CHOSEN.
Please make the highest bid for people assigned to the $800 category equal to $1000.
Please make the lowest bid for people assigned to the $25 category equal to $10.
[SP]
B1. The State estimates that this program will cost $[Cost_H], Your household would pay this one
time tax on next year's income tax and the money would go into a special fund set up to clean
up the river. There will be a referendum to decide whether the river will be cleaned up and how
much the one-time tax should be. If the election were being held today and the total cost would
be a one time additional tax of $[Cost_H], would you vote for or against it?
For 1
Against 2
Prompt once.
Show B2 if B1 = "For".
For B2, create a data-only variable indicating what bid higher than Cost_H was selected.
[SP]
B2. $[Cost_H] represents the best estimate of the engineering costs. It could be that the cost to
each household would be as high as $[Next bid up from [Cost_H] instead of $[Cost_H], If
this was the case, and the one time tax would be $[next bid up from [Cost_H]], would you
vote for or against it?
For 1
Against 2
Show B3 if B1 = "Against" or skipped.
For B3, create a data-only variable indicating what bid lower than Cost_H was selected.
[SP]
B3. $[Cost_H] represents the best estimate of the engineering costs. It could be that the cost to
each household would be lower and would only be $[Next bid lower than [Cost_H]] instead
17
-------
of $[Cost_H], If this was the case, and the one time tax would be $[next lower bid than
[Cost-H]], would you vote for or against it?
For 1
Against 2
Show B4 if B2 = "Against" or skipped or B3 = "Against" or skipped.
[MP]
B4. The State is interested in knowing why you would vote against the program. There are lots of
different reasons why you might vote against the program, like it just isn't worth that much
money, or it would be difficult for your household to pay that much even though you support the
program. Or there might be some other reason.
Isn't worth the money 1
Difficult for my household to pay 2
Don't believe the cleanup would work 3
Some other reason, please specify: 4
Show B5 and B6 if B1, B2 or B3 = "For".
[SP]
B5. People have lots of different reasons for voting for the program. Could you briefly describe why
you would be willing to pay for it?
I'm worried about the potential risks to
unborn babies 1
I support a cleanup no matter what 2
Some other reason, please specify: 3
[SP]
B6. Thinking back on your responses, how confident would you say you are in your willingness to
pay on a scale of 1 to 5 where 1 is "Not confident at all" and 5 is "Very confident"?
Not confident at
all
Very confident
1
2
3
4
5
C. Questions Related to Qalys
[SP]
C1. Now we're going to ask a slightly different question. Assume for a moment that your child was
exposed to PCBs and has a slight reading comprehension deficit. Further assume there is a
treatment available to remedy the impairment, but that it comes with a very small chance of
dying as a result of the treatment. Would you accept a risk of death of 10 in 100,000 for your
child to cure the deficit for the rest of the child's life (assuming all other risks remain the same)?
This is also randomized - respondents see either 10 in 100,000 or 1 in 100,000 and then half
that or double that for C2 and C3 respectively depending on whether they answered no or yes,
respectively
18
-------
Yes 1
No................... 2
[SP]
C2. If the risk of death was only 5 in 100,000, would you take the treatment?
Yes 1
No 2
[SP]
C3. If the risk of death was as high as 20 in 100,000, would you take the treatment?
Yes 1
No 2
D. Questions Related to Eagle Receptor Exposure to Polychlorinated Biphenyls
PCBs can have effects on the environment and the birds and mammals that
use the environment. This part of the survey is to find out whether you would
be willing to pay an additional tax for the additional benefit of protecting
ecological receptors like eagles. Many years ago, eagles were in danger of
becoming extinct. Now, they are successfully hatching young and
maintaining their populations in some places in the United States. But that is
not the case along several waterways in this State. Studies have shown that
eagles are sensitive to chemicals in the environment, particularly ones like
PCBs that build up in the food chain. Sensitive receptors like eagles will
show effects at lower concentrations than humans. As exposure to PCBs
increases, there is an increase in the probability of a decline in reproductive
capability. Scientists aren't sure what probability of a decline in reproductive
capability leads to extinction, but any decline is likely to have a noticeable
effect in the population of a species like an eagle, which only produce one or
two young per year and which have small populations to begin with.
Programming notes:
Please show one of the next two Display screens randomly; 50% see one version, 50% see the
other. Please create a data-only variable indicating which screen was shown.
In the first Display screen, please show "10%" and "5%" randomly; 50% see one and 50% see
the other. Please create a data-only variable indicating which percentage was shown.
In the second Display screen, please show "25%" and "10%" randomly; 50% see one and 50%
SEE THE OTHER. PLEASE CREATE A DATA-ONLY VARIABLE INDICATING WHICH PERCENTAGE WAS SHOWN.
[Display]
Because of exposure to PCBs, scientists have estimated there is a 20% chance that eagles will
experience a decline in reproductive capability that could impact the population.
If the sediments are removed and the river is cleaned up, scientists estimate that the risk decreases to
[10%/5%].
[Display]
19
-------
Each dot below represents one eagle: The red dots represent the eagles that will not be able to
reproduce.
[Show image with 20 red dots.]
If the river is cleaned, scientists predict that [10/5] eagles will have trouble reproducing. There will
always be some chance of effects because the sediments can't be 100% cleaned up.
[Show image with 10 or 5 red dots depending on condition selected.]
[Display]
Insert risk graphic here.
100%
Probability
of effect
occurring
50%
25%
Because of exposure to PCBs,
scientists have estimated that
there is a 50% chance that 20%
of the species will have trouble
producing young. If the river is
cleaned up, scientists estimate
that this risk decreases to 25%.
There will always be some small
chance of effects because the
sediments can't be 100%
cleaned up.
20%
Percent of species that will
experience reproductive effects
Note to Scripter: Randomly select the value for Cost_E from the following choices: $25,
$50, $100, $200, $400, $800; Each value should be assigned for approximately 16.7% of the
RESPONDENTS (I.E., 100/6). PLEASE CREATE A DATA-ONLY VARIABLE INDICATING WHICH VALUE WAS
CHOSEN.
Also note that followup items use the next highest/lowest bid.
IMPORTANT: THE FIRST BID THAT IS SELECTED HERE SHOULD CORRESPOND TO COST_H
THAT WAS AGREED TO EARLIER. IF THE RESPONDENT SAID YES TO BOTH BIDS IN B, THEN
COST_E = NEXT HIGHEST BID AMOUNT. IF RESPONDENT SAID YES THEN NO, COST_E = TO
20
-------
LAST (REJECTED) BID AMOUNT FROM B. IF RESPONDENT SAID NO, THEN COST_E IS
RANDOMLY SELECTED FROM THE FULL BID VECTOR.
Please make the highest bid for people assigned to the $800 category equal to $1000. Please
MAKE THE LOWEST BID FOR PEOPLE ASSIGNED TO THE $25 CATEGORY EQUAL TO $10.
[SP]
D1.
The State estimates that this program will cost each household $[Cost_E], Your household
would pay this one time tax on next year's income tax and the money would go into a special
fund set up to clean up the river. There will be a referendum to decide whether the river will be
cleaned up and how much the one-time tax should be. If the election were being held today and
the total cost would be a one time additional tax of $[Cost_E], would you vote for or against it?
For
Against.
.1
.2
Prompt once.
Show D2 if D1 = "For".
[SP]
D2.
$[Cost_E] represents the best estimate of the engineering costs. It could be that the cost to
each household would be as high as $[Next bid up from [Cost_E]] instead of $[Cost_E], If
this was the case, and the one time tax would be $[next bid up from [Cost_E]], would you
vote for or against it?
For
Against.
.1
.2
Show D3 if D1 = "Against" or skipped.
[SP]
D3.
$[Cost_E] represents the best estimate of the engineering costs. It could be that the cost to
each household would be lower and would only be $[next bid down from Cost_E] instead of
$[Cost_E], If this was the case, and the one time tax would be $[next bid down from
[Cost_E]], would you vote for or against it?
For
Against.
.1
.2
Show D4 if D2 = "Against" or skipped or D3 = "Against" or skipped.
[MP]
D4.
The State is interested in knowing why you would vote against the program. There are lots of
different reasons why you might vote against the program, like it just isn't worth that much
money, or it would be difficult for your household to pay that much even though you support the
program, or you are opposed to dredging as an alternative. Or there might be some other
reason.
Isn't worth the money
1
21
-------
Difficult for my household to pay 2
Don't believe the cleanup would work 3
Some other reason, please specify: 4
Show D5 and D6 if D1, D2 or D3 = "For".
[SP]
D5. People have lots of different reasons for voting for the program. Could you briefly describe why
you would be willing to pay for it?
I'm worried about the eagles 1
I support a cleanup no matter what 2
Some other reason, please specify: 3
[SP]
D6. Thinking back on your responses, how confident would you say you are in your willingness to
pay on a scale of 1 to 5 where 1 is "Not confident at all" and 5 is "Very confident"?
Not confident at
all
Very confident
1
2
3
4
5
E. Questions Related to Motivation
[Display]
It is important for regulators to know how you came to your decision.
[SP]
E1. How concerned are you about chemicals in the environment?
Not at all concerned 1
Somewhat concerned 2
Quite concerned 3
Very concerned 4
[SP]
E2. How concerned are you about PCBs in the environment?
Not at all concerned 1
Somewhat concerned 2
Quite concerned 3
Very concerned 4
[SP]
E3. Do you believe that PCBs could cause the reproduction problems in eagles?
22
-------
Yes 1
No 2
Not Sure 3
[SP]
E4. Do you believe that PCBs could cause developmental delays in young children exposed in the
womb?
Yes 1
No 2
Not Sure 3
[SP]
E5. Did you feel like the survey pushed you to vote a particular way or did you feel like you really
made up your own mind based on the best available information?
Pushed to vote for it 1
Pushed to vote against it 2
Made up my own mind 3
Not Sure 4
[Large Text Box]
E6. What is it about the survey that made you feel that way?
[SP]
E7. Thinking back on all the information, would you say the reproduction problems facing eagles in
this state are...
Not serious at all 1
Somewhat serious 2
Very serious 3
Extremely serious 4
Not sure 5
[SP]
E8. Thinking back on all the information, would you say the risks facing unborn babies due to
exposure to PCBs in this state are...
Not serious at all 1
Somewhat serious 2
Very serious 3
Extremely serious 4
Not sure 5
F. Questions Related to Recreational Activities
23
-------
[SP]
F1. How often do you personally watch television programs about wildlife?
Never 1
Rarely 2
Sometimes 3
Often 4
All the time 5
[SP]
F2. Do you live near a river, lake or stream?
Yes 1
No 2
Never 1
Rarely 2
Sometimes 3
Often 4
All the time 5
[SP]
F3. How often does your family spend time near a river, lake or stream?
[SP]
F4.
How often do people in your household eat fish?
Never 1
A few times a year 2
A few times a month 3
Every week 4
TGrid - SP BY Row]
G2. You receive a lot of information from a lot of different sources. In general, how much confidence
do you have in information you obtain from:
No
Some
A Lot of
Confidence
Confidence
Confidence
Federal government
Scientists who work for industry
Scientists who work for universities
Television media
Internet sources [No selection for this header
item]
Government web sites
Commercial web sites
Non profit web sites
Academic web sites
Print media (newspapers, magazines)
24
-------
Joint Determination in a General Equilibrium Ecology/Economy Model
David Finnoff
dfinnoff@bus.ucf.edu
John Tschirhart
jtsch@uwyo.edu
Department of Economics and Finance
Department 3985 • 162 Ross Hall
University of Wyoming
Laramie, WY 82071
October 2004
*To be presented at the EPA Valuation of Ecological Benefits: Improving the Science Behind
Policy Decisions - A STAR Progress Review Workshop. This research is supported by the U.S.
Environmental Protection Agency grant # RD-83081901-0.
25
-------
1
1. Introduction
This work is part of the ongoing effort by economists and ecologists to better integrate
their disciplines in order to improve policymaking. The motivation for the work is the realization
that all economic activity ultimately depends on the natural resource base and the ecosystems
contained therein, but the extent to which the base is tapped has limits (Arrow, et al., 1995). By
some accounts the limits have been reached and a depleted resource base is having negative
impacts on living standards (Norgaard 1994). Monitoring depletion and predicting future
resource limits requires a better understanding of the interplay between the ecology of natural
systems and economic activity (Nordhaus and Kokkelenberg, 1999). The objective here is to
develop a method to better capture the interplay. The method is useful for addressing the
numerous conflicts that arise when economic development and environmental conservation
appear at odds. Familiar examples include logging, harvesting wildlife, preservation of
biodiversity (Weitzman, 1993) and endangered species (Shogren and Tschirhart, 2000),
bioprospecting (Simpson, Sedjo and Reid, 1996), and, more generally, conserving the essential
human services supplied by natural environments (Daily, 1997).
In many economic papers that examine biological renewable resources, logistic growth
functions are employed to capture the resources' characteristics. Usually, a single growth
function is employed to study one species, thereby omitting the other species in the community.
Occasionally, two or three species are studied in a predator-prey relationship, or, as in Brander
and Taylor (1998), humans are the predator. The point is that in all this work entire communities
are reduced to one or two species, and a few parameters must summarize the numerous
interactions that occur in real ecosystems. Moreover, the logistic growth functions depend on
entire species' populations and as such they take a macro view in which species interactions, if
present at all, are at an aggregated level.
26
-------
2
An alternative is to model many interacting species in food webs and to do this at the
micro level so that individual organism behavior yields community populations. Of course,
modeling a community with many species is a challenge because everything depends on
everything else (Amir, 1979; Crocker and Tschirhart, 1992). Economists face a similar challenge
in modeling an economy in which everything depends on everything else. Economists address
the challenge by developing computable general equilibrium (CGE) models, and this is the tack
taken here. By exploiting the three themes fundamental to economics - rational behavior,
efficiency and equilibrium, a general equilibrium model of an ecosystem is built. The general
equilibrium ecosystem model (GEEM) is then tied to a general equilibrium of an economy to
examine the ecosystem/economy interplay.
GEEM is a new adaptive approach that appeals to the oft-made analogies between
economies and ecosystems in both the economic and ecological literatures (Tschirhart, 2000,
2002, 2004).1 Like CGE models that rely on micro foundations of individual consumer and firm
behavior to drive the macro outcomes, the individual plant and animal behavior in GEEM
appeals to the micro principle that success depends on their energy utilization, and this drives the
ecological macro outcomes (i.e., population changes). Population updating uses general
equilibrium results from individual plant and animal net energy optimization aggregated to
species levels, similar to how CGE economic models start with individual consumer and firm
demands and supplies and aggregates them to market levels.
In this paper, CGE/GEEM is applied to the Alaskan economy that is linked via its fishing
and tourism industries to an eight species marine ecosystem that includes an endangered species.
The fisheries sector is modeled as a regulated open access fishery (Homans and Wilen, 1997) but
is significantly modified to be compatible with the general equilibrium framework. Each general
1 However, the similarities only go so far and there are features in GEEM that are not found in economic models
(Tschirhart, 2003). For example, predators and prey do not engage involuntary exchange, but inbiomass transfers.
27
-------
3
equilibrium calculation corresponds to one year, but the fishing season is considerably shorter
than one year. The fishing off season is explicitly modeled by allowing for fishing factors to
receive rents in season that carry them through the off season, and by including in welfare the
off-season leisure enjoyed by unemployed fishery labor.
Results from the linked models include period-by-period gross state product, prices and
quantities for final goods and factors in the economy, and predator/prey biomass consumption,
energy prices, and species populations in the ecosystem. In addition, welfare comparisons of
alternative fishing regulations are presented. Welfare is increased with mandatory reductions in
fish harvests to protect the endangered Steller sea lions that feed on fish for two main reasons.
First, and as expected, capital and labor move from the regulated open access fishery sector to
other sectors where they both earn more on an annual basis, and second, the tourism industry
grows owing to increased numbers of marine mammals.
In what follows a brief description of GEEM for the marine ecosystem is provided. This is
followed by a presentation of how the fishery is merged into the CGE model and what welfare
measure may be appropriate for the linked systems.
2 The Ecology Model
GEEM is applied here to an oft studied marine ecosystem comprising Alaska's Aleutian
Islands (AI) and the Eastern Bering Sea (EBS). The ecosystem is represented by the food web in
Figure 1. All energy in the system originates from the sun and is turned into biomass through
plant photosynthesis. Photosynthesis is carried out in the AI by individuals of various species of
algae, or kelp, and in the EBS by individuals of various species of phytoplankton. All individual
animals in the system depend either directly or indirectly on the kelp and phytoplankton plant
species. In the EBS, zooplankton prey on phytoplankton and are prey for pollock. The pollock
are a groundfish that support a very large fishery. Steller sea lions, an endangered species, prey
28
-------
4
on pollock, while killer whales prey on the sea lions. In the AI, killer whales also prey on sea
otter that in turn prey on sea urchin that in turn prey on kelp.
In GEEM, demand and supplies are developed somewhat similarly to CGE. Species are
analogous to industries, and individual plants and animals are analogous to firms. Plants and animals
are assumed to behave as if they maximize their net energy flows. Where perfectly competitive firms
sell outputs and buy inputs taking market-determined prices as signals, plants and animals transfer
biomass from prey to predators taking 'energy prices' as signals. (Plants can be thought of as preying
on the sun.) An energy price is the energy a predator loses to the atmosphere when searching for,
capturing and handling prey. A key difference between economic markets and ecological transfers,
however, is that in the latter the prey does not receive this energy price. Therefore, the biomass
transfer is not a market because there is no exchange (Tschirhart, 2003). Nevertheless, predators'
demands and preys' supplies are functions of the energy prices.
A brief sketch of GEEM is provided here, but for details see Finnoff and Tschirhart
(2003, 2004). The three basic equations that comprise GEEM are given by (2.1) - (2.3). The first
equation is a general expression for the net energy flow through a representative animal from
species i.
i-1 m i-1
Rr = E " erj K " S ei [' + ^ " /' (Z XV ) " Pi (2' ')
j=1 k=i+l j=1
NiXtjiet) = .Yr,(.v(t')) (2.2)
(2.3)
N\+l =Nj+Nj
S: S:
Ri is in power units (e.g., Watts or kilocalories/time).2 The species in (2.1) are arranged so that
members of species i prey on organisms in lower numbered species and are preyed on by
2 According to Herendeen (1991) energy is the most frequently chosen maximand in ecological maximization
models, and energy per time maximization as adopted here originates with Hannon (1973) and expanded to multiple
29
-------
5
members of higher numbered species. The first term on the right side is the inflow of energy
from members of prey species (including plants) to the representative individual of species z. The
choice variables or demands, Xy, are the biomasses (in kilograms/time) transferred from the
member of species j to the member of species z, are the energies embodied in a unit of biomass
(e.g., in kilocalories/ kilogram) from a member of species j, and e,, are the energies the member
of species z must spend to locate, capture and handle units of biomass of species j. These latter
energies are the energy prices. There is one price for each biomass transfer between a predator
and prey species. As in economic CGE models, the prices play a central role in each individual's
maximization problem, because an individual's choice of prey will depend on the relative energy
prices it pays. Individuals are assumed to be price takers: they have no control over the energy
price paid to capture prey, because each is only one among many individuals in a predator
species capturing one of many individuals in a prey species.
The second term is the outflow of energy to animals of species & that prey on z. The e, i s
the embodied energy in a unit of biomass from the representative individual of species z, and y,i:
is the biomass supplied by z to k. The term in brackets is the energy the individual uses in
attempts to avoid being preyed upon. It is assumed to be a linear function of the energy its
predators use in capture attempts: the more energy predators expend, the more energy the
individual expends escaping. U is a tax on the individual because it loses energy above what it
loses owing to being captured. The third and fourth terms in (2.1) represent respiration energy
lost to the atmosphere which is divided into a variable component, f(), that depends on energy
intake and includes feces, reproduction, defending territory, etc., and a fixed component, that
is basal metabolism.
Time in the Alaskan model is divided into yearly reproductive periods. Each year a
species in Crocker and Tschirhart (1992) and to the individual level in Tschirhart (2000). Energy per time is also the
30
-------
6
general equilibrium is determined wherein the populations of all species are constant, each plant
and animal is maximizing its net energy (using the derivatives of (2.1) for first-order conditions),
and aggregate demand equals aggregate supply between each predator and prey species. For each
price that equates a demand and supply transfer there is an equilibrium equation given by (2.2).
Each plant and animal is assumed to be representative individuals from its species; therefore, the
demand and supply sums are obtained by multiplying the representative individual's demands
and supplies by the species populations given by the Nterms.
A representative plant or animal and its species may have positive, zero or negative net
energy in equilibrium. Positive (zero, negative) net energy is associated with greater (constant,
lesser) fitness and an increasing (constant, decreasing) population between periods. (The analogy
in a competitive economy is the number of firms in an industry changes according to the sign of
profits.) Net energies, therefore, are the source of dynamic adjustments. If the period-by-period
adjustments drive the net energies to zero, the system is moving to stable populations and a
steady state. The predator/prey responses to changing energy prices tend to move the system to
steady state.
The adjustment equation for the ith species is given by (2.3) where 11 (•) = /^(x,7; N') is
the optimum net energy obtained by substituting the optimum demands and supplies as functions
of energy prices into objective function (2.1). N' is a vector of all species' populations and it
appears in 11 (•) to indicate that net energies in time period t depend on all populations in time
period t. In the steady state, 11 (•) = 0. Also, .v, is the lifespan of the representative individual, v, is
the variable respiration, vfs is the steady-state variable respiration, and N,is the species steady-
state population. The first and second terms in brackets in (2.3) are the birth and death rates.
Expression (2.3) reduces to the steady state if 11 (•) = 0 (in which case v, = v/" and A'/ = Nfs).
individual's objective in the extensive optimum foraging literature (e.g., Stephens and Krebs 1986).
31
-------
7
Because the biomass demands depend on the period t populations of all species, the population
adjustment for species i depends on the populations of all other species. In addition, out of steady
state Rj() and v, change across periods. These changes distinguish the GEEM approach from
most all ecological dynamic population models, because the latter rely on fixed parameters in the
adjustment equations that do not respond to changing ecosystem conditions.
3 The Economy Model
The CGE model pioneered by Ballard et al. (1985) and applied in the OECD GREEN
model (Burniaux et al, 1991) is most appropriate for linking with GEEM. The approach Ballard
et al. developed may be termed "myopically dynamic," because it consists of a sequence of static
optimizations and resulting equilibria connected through the evolution of factor stocks and
household savings. Households are intertemporal optimizers whose savings decisions are based
on myopic expectations over future prices.
The economy is modeled as having three production sectors: the fishery F, recreation
and tourism R, and composite goods C.3 The fishery is modeled as a single, vertically integrated
industry consisting of catcher vessels, catcher processors and, motherships and inshore
processors. Recreation and tourism represents the Census Bureau's classification of Wildlife
Related Recreation, and composite goods are a catch all for the residual private industries in
Alaska. Profit-maximizing, price-taking firms employ harvests of pollock in the fishery, non-
consumptive use of marine mammals (Steller sea lions, killer whales and sea otter) in recreation,
and capital and labor in all sectors, to produce their outputs in a continuous, nonreversible, and
bounded process. Outputs from the fishery, recreation, and composite goods are sold in regional
markets and exported out of the region, while regional production is differentiated from imports
3 The sector and regional profiles follow the Steller Sea Lion Supplemental Environmental Impact Statement (SEIS,
U.S. Department of Commerce, 1991).
32
-------
for fish and composite goods following Armington (1969). Capital K and labor L are
homogeneous and defined in service units per period. They are also perfectly mobile between
sectors and between periods, but not within periods which is pertinent for the fishery. Sector i
factor employment levels are given by Kt and L, (i = F, R, C).
The linkage between the fishery and the ecosystem is presented in detail. The treatment
of the tourism industry that depends on the marine mammals, the households, the composite
goods, and trade and price relationships are presented in detail in Finnoff and Tschirhart (2004).
3.1 Fishery Incorporating a fishery into a CGE framework raises issues that require
two modifications to the standard fishery models. First, where most of the fishery literature
employs effort as the single human factor of production, capital and labor must be included in
CGE so that the fishery interacts with other sectors. Second, the non-fishery sectors hire capital
and labor in service units per year, but in the fishery factors are employed considerably less than
one year and may earn rents.
Expressions (3.1) - (3.4) summarize production in the fishery sector:4
TACt=a + bN°/ (3.1)
HF=dFTapNA (3.2)
m )
minimize wLF +fKF subjectto T = dFLjfKF F (3.3)
Equation (3.1) introduces government into the model in the form of a fishery manager. Homans
and Wilen (HW, 1997) developed a model of a regulated open-access fishery to reflect that
fishery managers set total allowable catch, TAC, and fishing season length, T. The heavily-
regulated Alaskan pollock fishery fits this institutional arrangement. To mesh an HW type model
with the CGE framework, the fishery manager chooses period t's TAC according to (3.1) where
N4 is the population of pollock. No harvests are allowed whenever the actual biomass is less than
33
-------
9
the minimum level set by the manager. For given TAC and technology, the season length is
determined from the aggregate harvest function in (3.2), where aF and dF are parameters and HF
is aggregate harvest. The industry is assumed to harvest up to their limit so that HF = TAC.
The season length is the time needed to land the TAC given the fish stock and is
increasing in TAC (Homans and Wilen, 1997). Following the fishery manager's choices for TAC
and T, the industry is assumed to minimize the cost of harvesting according to (3.3) by
employing capital and labor to work time T. The production function exhibits constant returns to
scale, d"F, and cf"F are parameters, and w and r are the fishery wage and rental rate of capital
that may diverge from the market wage and rental rate in other sectors. The associated cost
function is linearly homogenous in time, allowing the total costs of harvesting to be written as
C(w,r)T. This setup with the industry choosing K and L for a given season length incorporates
the two modifications defined above.
The divergence of fishery factor prices from market factor prices in the other sectors
arises from the restricted season length and is an important feature of the model. Entry is
assumed to dissipate all rents in open access models. But these are partial equilibrium models
and factors are either not defined over time or if they are defined, they are instantaneous rates or
daily rates as in Clark (1976). What these factors are doing off season is not an issue, because
there is no off season in the models. In the CGE setting where all other sectors are operating year
round, the fishery experiences an off season during which factors are either unemployed or
employed elsewhere, often outside the region. In reality, unemployment is common and it may
be either voluntary, or involuntary owing to factor immobility between seasons.5 In either case,
4 We are indebted to Robert Deacon for his invaluable input in the development of this section.
5 The Alaskan Department of Labor and Workforce Development provides information about fishing jobs in Alaska
on various websites (e.g., http://www.labor.state.ak.us/esd alaska iobs/careerstreams.htni). The job descriptions
suggest that workers can save money, and pay can be substantial if the fishing is good. College students are
encouraged to apply and then return to college in the off season. Boyce (2004) examines rents in fisheries and
assumes that fishing inputs cannot be redeployed during the off season.
34
-------
10
rational factors may demand higher than market payments in season in anticipation of being
unemployed off season. If they do, seasonal factor payments will not be driven down to market
levels in season, leaving positive seasonal rents. One might argue that these above market
payments are not really rents, because they are merely covering the opportunity costs of factors
in the off season. This is certainly not true for voluntary unemployment because the factors are
enjoying leisure. But even for involuntarily unemployment, rational factors will anticipate some
transition time before reemployment, and will enjoy rents if the transition time is equal to or less
than what they anticipate.
Let W and R be the market determined factor prices for labor and capital in other sectors.
Because labor and capital are defined in service units per year, W and R are annual payments. Let
13 G (0, 1) be the percent of the year the fishery is active so that market factor prices in the fishery
are /3 W and (3 R. If there are intra-seasonal rents in the fishery, they must be reflected in factor
prices that deviate from these market prices such that C(WP,RP) < C(w,r) . Assuming any
rents impact labor and capital uniformly and linearly, let 8 be a rent divergence term so that the
factor prices in the fishery are:
w=/3 S W and r=/3SR
where 8 = 1 => no rents and 8 > 1 => positive rents. 6 (3.4)
In developing the simulation model the available data provides estimates for f3 and 8. But
the data is inadequate to determine whether factors were voluntarily or involuntarily unemployed
or whether they were reemployed during the off season. Therefore, the assumption made here is
that labor is voluntarily unemployed, i.e., enjoying leisure, and capital is idle or employed
outside the Alaskan economy. Labor's leisure time in the off season will be accounted for in
6 Factor price distortions commonly enter the CGE literature in the form of taxes (Harberger, 1974, Shoven and
Whalley, 1976, Ballard et al., 1985, and Bovenberg and Goulder 1996). The divergences here are not distortions in
the usual sense: /j is merely an accounting adjustment to correct for a shorter work year, and a d > I may be welfare
enhancing since some positive rents are desirable.
35
-------
11
welfare measures below.
Equilibrium for the industry is given by a pseudo zero-profit condition that allows for
intra-season rents:
Tip = PfHf - C(PSW, (38R)T = 0 (3.5)
In this representation, the total factor payments over the season equal the total revenue divided
by the season length, or an average revenue per time. An exogenous increase in TAC or HF
increases season length for a given fish stock and 8falls to maintain equality in (3.5). Intuitively,
the longer season implies less off-season time for the factors, and they require less rent in season
to get through the off season. To summarize, after the fishery manager sets TAC by (3.1) and T
by (3.2), the factor demands and the rent divergence 8 are determined by (3.3) and (3.5).
3.2 Equilibrium and Dynamics The economic system is in equilibrium when
households and firms optimize, there exists a set of prices and level of output at which all firms
break-even, Walras Law holds, and all markets clear. Incomes are derived through a two-stage
process. Regional households are endowed with labor coLAK and capital cof/'k. While foreign
value added expenditures (from foreign factor employment in the fishery) accumulate elsewhere,
regional value added expenditures flow first to factor "institutions", and then redistributed to
households. We close the model through the region's current account and savings investment
balance. Economic dynamics are recursive, consistent with the evolution of species populations.
Given myopic expectations, the time path of the economy is represented by a sequence of
competitive equilibria, one for each period. The periods are linked through factor accumulation,
where savings in each period (and therefore regional investment I,) expand the capital service
endowment for the subsequent period, and the effective labor force grows at an exogenous rate.
If the capital stock grows at the same rate as the effective labor force, the economy is on a
balanced growth path; however, balanced growth is not a feature of the linked model, because
36
-------
12
species populations cannot grow continually.
3.3 Welfare Measures The welfare impacts of alternative policies are evaluated in
terms of modified Hicksian equivalent variational measures similar to those developed in
Ballard et al. Each policy change leads to changes across prices and income in relation to a
reference/benchmark sequence of business as usual. Let the Hicksian expenditure function
associated with consumption in period t be given by Vectors of prices in any period t of the
reference scenario b or policy alternate a are given by and Pat, with corresponding indirect
utility functions F61 and Vt. In the results we employ annual equivalent variations
EVt =Mt(Ptb,Vta) -Mt(Ptb,Vtb)7to calculate welfare changes for any single period across policy
scenarios. Cumulative aggregate (or multi-market) welfare measures are found using discounted
summations of EVt (PEV) which is possible as the measure is based upon a common baseline
price vector. Future welfare changes are discounted both by consumers' rate of time preference
and by the human population growth rate. Also, given the exogenous time horizon a termination
term is added to account for welfare impacts after the final period T. In this we assume that by T
the economy is close to a steady state.
4. Model Specification
4.1 Ecological Specification and Data In applying GEEM to the Alaskan ecosystem,
ecological studies of the Alaskan and other ecosystems were used. Time series of pollock
biomass estimates exists for the period 1966 through 1997, and the rest of the data are from 1966
or interpolated to that date. Data were obtained for plant and animal populations, benchmark
plant biomasses and animal biomass demands, and parameters that include embodied energies,
basal metabolisms, and plant and animal weights and lifespans. Sources include numerous
7 Defined as the difference between initial expenditure and that expenditure necessary to achieve the post-policy
level of satisfaction at initial prices.
37
-------
13
National Marine Fisheries Service publications and ecological journal articles. Details on data
sources can be found in Finnoff and Tschirhart (2003).
Using this data, calibration yielded estimates for parameters in the plant and animal
respiration and supply functions (Finnoff and Tschirhart, 2003). Calibration consists of
simultaneously solving for each species the net energy expressions set to zero, first-order
conditions or the derivatives of the net energy expressions set to zero, and the equilibrium
conditions.
4.2 Economic Specification and Data In a similar fashion as with the ecosystem
model, the economic specification is based on a chosen benchmark year, and the data were used
in calibrations to estimate parameters. The benchmark dataset constructed in the analysis is
shown in Table 1 where all values are in millions of dollars. The data sources include reports
from the U.S. Department of Commerce, Bureau of Census, Bureau of Labor Statistics, the
Alaskan Bureau of Economic Analysis, and others. Details on data sources are in Finnoff and
Tschirhart (2004).
5. Policy Analysis
The NMFS in 2001 issued a Supplemental Environmental Impact Statement (SEIS)
containing alternative management strategies that specify various pollock catch limits and no
fishing zones to protect both the sea lions and the fishery. Using the linked CGE models, the
effects of the management strategies on economic welfare are examined, and then the linked
model is compared to a business-as-usual model that does not account for economy/ecosystem
interactions.
The management strategies are differentiated here by the regulator's choice of b in the
quota function (3.1). Holding N™m constant, b is varied by 30% and 170% of its 1997 harvest
38
-------
14
levels. (Numerous other harvest levels were examined but not reported. The 30% (170%) results
are indicative of all runs below (above) the benchmark harvest.) All general equilibrium
calculations and population updates were made with the nonlinear programming software
package GAMS. The calculations consist of four steps: 1) Given current species populations, a
GEEM equilibrium is found, determining species net energies, energy prices, biomass demands
and supplies. 2) Given current species populations, the fishing manager determines the TAC (that
is adjusted in separate scenarios by the two percentages above). 3) Given current species
populations and capital and labor endowments, a CGE is found, delivering prices, fish harvests
and other outputs, incomes, investment, savings, factor employment and the rent divergence, &
4) In the ecosystem, given the findings from step 1) and the TAC from step 2), the species
populations are updated. In the economy, given current endowments and the findings form step
3), factor endowments are updated. The updated populations and endowments from steps 3) and
4) are then used to start the next period by retuning to step 1). The steps are repeated each period
of the time horizon across each trade elasticity specification.
A benchmark scenario is initiated using the 1997 benchmark dataset, then simulated for
50 and 100 years.8 Given natural resource stocks (species populations) whose growth is limited
by biological carrying capacities, balanced growth is not a feature of the benchmark scenario.
This is a departure from Ballard et al. or numerous other applications, where balanced growth is
characterized by all quantities increasing by the same rate and constant relative prices. In the
benchmark scenario, all quantities evolve at a constant rate, but the rate may vary over sectors
owing to the reliance of the fishery and recreation sectors on biological natural resource inputs.
Further, given heterogeneous growth of the natural resources, benchmark relative prices do not
remain constant.
8 Sequence lengths of 100 years and a discount rate of 4% were chosen as representative for Federal projects.
39
-------
15
5.1 Ecosystem Impacts The impacts on populations of pollock, sea lions, sea otter and
killer whales for the 30% and 170% management strategies are shown in Figure 2. Predicted
populations (given actual harvests) prior to the 1997 calibration year and projections beyond
1997 are displayed. All populations move to new steady states, in as little as 10 years for
phytoplankton (not shown) but as many as 30 years for killer whales. Phytoplankton are short-
lived (less than one year) and reproduce rapidly, whereas killer whales are long-lived (twenty
years) and reproduce slowly.9 Reduced pollock harvests (30%) result in long-term increases in
phytoplankton, sea urchins, sea lions and killer whales, and long-term decreases in zooplankton,
kelp, and sea otters. The recreation sector will benefit from more sea lions and killer whales, but
will be hurt by fewer sea otter.
To appreciate the general equilibrium nature of the population changes, consider the
170%) harvests in some detail. The immediate affect of the higher harvest is to lower the pollock
population. In the subsequent period the lower population increases the energy price sea lions
pay to capture pollock and the sea lion demand for pollock decreases. Sea lion net energy
decreases as a result and their population falls. These changes work their way up the food web as
the killer whale population reacts in the same way to the fall in sea lions as the sea lion
population reacted to the fall in pollock. The further up the food web from pollock, the less
pronounced the impact. Where pollock populations fall by about 24%, sea lion and killer whale
populations fall by about 13% and 9%, respectively.
5.2 Economic Impacts - Following changes in fishing policies, there occur many
simultaneous changes in prices, incomes and profits. We can trace the flows of outputs, capital
and labor between industries and between domestic and foreign sectors. More detail is in Finnoff
and Tschirhart (2004). Here we concentrate on welfare impacts. The welfare impacts of
9 Average lifespan enters into the population update equation, (2.9), similar to the way the less tangible species
growth rates enters into the often-used but simplistic logistic update equation; thus, the lifespans are important in
40
-------
16
alternative management strategies are quantified as discussed in Section 3.4. Welfare changes
(from the reference) presented in Table 2 are the present value of the cumulative sum of
equivalent variations Pev° over 50 and 100 year planning horizons. For both horizons, leisure
accruing to regional labor in the fishery during the off-season was valued at full, three quarters
and half the wage rate. Under both horizons and across leisure values, decreasing the quota
always results in cumulative aggregate welfare gains (Pev)- The longer the horizon and the
greater the leisure values, the smaller the gains. For brevity, in the following discussion we
focus on the30% reduced quota, noting that the 170% increased quota produces opposite results.
Figure 3 is helpful in understanding the fishery's contribution to the welfare changes.
Starting from a steady state, 7° is the season length and the average revenue per time from (3.5)
is downward sloping as shown by the solid line. At 10 factor payments are C(w°,f°) which
exceeds market-based factor payments C(W,R) owing to rents. In the next period the fishery
manager lowers the harvests and because the fish stock has not changed, the season length falls
to T1. The shorter season means less labor and capital in the fishery, but these remaining factors
enjoy higher rents (C(wl,rl)-C(W,R)) per time employed as S adjusts upward. As explained
above, for labor the shorter season results in fewer fishery workers who enjoy higher rents per
time worked and greater off-season leisure, while the workers who leave the fishery are
employed at market wages in other sectors for the full year with no leisure.
In the second period the fish population is greater and the price of fish is higher because
of the reduced harvest strategy. Both changes cause the average revenue curve to shift upward.
The fishery manager sets a greater TAC by (3.1) because of the greater fish population, and the
determining whether population oscillations occur and how quickly populations will converge to steady state.
10 In the absence of balanced growth in the reference sequence, we deflate all prices to 1997 levels using a modified
Laspeyres formula CPIt = PtQ0P0Q0 ] * 100 where CPIt is the price index in period t, Pt current price of each
commodity, Q0 is the market quantity of each commodity in the baseline period (1997) andP0is the price of each
41
-------
17
season length increases to T2 although it is less than the initial season length. 8 adjusts downward
and rents fall to C(w ,r )-C(W,R) . Some workers now return to the fishery from the other
sectors, leaving their full-year market wages for higher part-year wages and leisure. In addition,
because the fish population is greater, the fishery factors are more productive. Over the
remainder of the planning horizons, the season lengths remain between 7° and and the rents
remain between the initial low value and the second period high value.
In a demonstration exercise, we quantify those portions of welfare changes attributable
only to changes in ecosystem populations. The simulations were rerun with marine mammal
inputs to recreation held at their reference sequence levels across the two TAC strategies. The
portion of welfare change attributable to changes in marine mammals can then be inferred as the
difference in periodic equivalent variations between the simulations with and without the
impacts fishing has on the food web.11 The ecosystem valuations for alternative quota rule are
displayed in Table 3. While the magnitudes of the ecosystem valuations are small due to
assumptions made in parameterizing the model, they are consistent for increases or decreases in
ecosystem inputs. They loosely indicate the direct value to the economic system of marginal
changes in ecosystem quality. Each one percent annual improvement in ecosystem quality in
relation to the reference is worth roughly $110,000. Further, these values demonstrate that under
a reduced TAC, if these ecosystem values were to be ignored the welfare impacts will be
understated. Increased quota rules will result in overstated welfare benefits
6. Conclusion
We demonstrate that Steller sea lion recovery measures via alternative pollock quotas
have consequences throughout the ecosystem and economy owing to the joint determination of
commodity in the baseline period. This follows the same general fashion of the BLS Consumer and Producer Price
indices
42
-------
18
important variables. Quota changes cause altered levels of all ecosystem populations, economic
factor reallocation, changes in all regional prices, incomes, demands, outputs, imports, exports,
and differential rates of factor accumulation. Without a jointly- determined analysis, the benefits
from a reduced quota accruing to the ecosystem inputs would be understated as would the costs
of a slower growing capital stock. The mediating behavior of each system to shocks arising from
the other is important for policy analysis.
Of the eight species modeled, four are used directly in the economy either as
consumption goods (fish) or non consumption goods (marine mammals). Nevertheless, all
species matter for the economy because the other four species are used indirectly as support for
ecosystem functions. A portion of the regional welfare gains from reduced quotas follow from an
economy relying less on resource extraction and more on resource non extraction. This result is
consistent with a report from the Panel on Integrated Environmental and Economic Accounting
which states: "economic research indicates that many renewable resources, especially in the
public domain, are today more valuable as sources of environmental service flows than as
sources of marketed commodities." (Nordhaus and Kokkelenberg, 1999, p. 177)
Our reported welfare impacts of alternative fishing policies may be understated for three
reasons. First, all species apart from pollock are at or close to a steady state in 1997. Changing
pollock harvests, therefore, result in relatively small changes in other species. Second, the three
marine mammals are assumed to be a small fraction of Alaska's ecological systems inputs to the
economy. Third, non-use values associated with the ecosystem (e.g., existence values) are not
considered. Turcin and Giraud (2001) conducted a willingness to pay survey that asked how
much households were willing to pay for continuing the Federal Steller Sea Lion Recovery
Program. They found Alaskan households willing to pay in total $25 million, and extrapolating
11 Values attributable to ecosystem inputs were found as EV11 - EVNLt where I, refers to ecosystem impacts (or
43
-------
19
to U.S. households the figure is $8 billion. Interestingly, household in the area of Alaska that
contains critical habitat for the sea lions were willing to pay considerably less and in some cases
negative amounts. These results do not indicate the existence value for changes in the sea lion
populations, but they do suggest that the value may be substantial
Extensions of this work will include enlarging the community of species by admitting
other harvested fish species (Pacific cod and herring), whale species (blue and sperm) and
another marine mammal (Northern fur seal). This will allow for more testing of ecological
hypotheses concerning how the economy and human actions impact the ecosystem. In addition,
more economic sectors will be added by using IMPLAN data for the Alaskan economy.
CGE models are useful in judging alternative economic policies for their effects on resource
allocation and on the distribution of net benefits. The objective of linking GEEM to CGE is to account
for resource allocation in ecosystems as well so that the scope of policies that can be judged is
broadened. While the economic and ecological underpinnings of this linked approach can be
extended and improved in many ways, CGE/GEEM is a step toward integrating disciplines with
common structures and goals.
linkages) being accounted for, and NL not accounted for (not linked).
44
-------
20
References
Amir, S. 1979. "Economic Interpretations of Equilibrium Concepts in Ecological Systems," Journal of
Social Biological Structures, 2, 293-314.
Armington, P., 1969. "A Theory of Demand for Products Distinguished by Place of Production," IMF
Staff Papers, 16, 159-178.
Arrow, K., B. Bolin, R. Costanza, P. Dasgupta, C. Folke, C. S. Holling, B.-O. Jansson, S. Levin, K.-G.
Maeler, C. Perrings and D. Pimentel. 1995. "Economic Growth, Carrying Capacity and the
Environment."
-------
21
Nordhaus, William D. and Edward C. Kokkelenberg, eds. 1999. Nature's Numbers, Washington,
D.C.: National Academy Press.
SEIS (Supplemental Environmental Impact Statement). 2001. Steller Sea Lion Protection
Measures in the Federal Groundfish Fisheries off Alaska. NMFS, Alaskan Region.
Shogren, J. F. and J. Tschirhart. 2000. Protecting Endangered Species in the United States:
Biological Needs, Political Realities, Economic Choices. New York: Cambridge Un. Press.
Shoven, J.B., and J. Whalley. 1992. Applying General Equilibrium. Cambridge: Cambridge University
Press.
Simpson, R. David, Roger A. Sedjo and John W. Reid. 1996. "Valuing Biodiversity for Use in
Pharmaceutical Research." Journal of Political Economy, 104, 163-185.
Stephens, D.W. and J.R. Krebs. 1986. Foraging Theory. Princeton: Princeton University Press.
Tschirhart, J. 2000. "General Equilibrium of an Ecosystem." J. Theoretical Biology, 203: 13-32.
Tschirhart, J. 2002. "Resource Competition Among Plants: From Optimizing Individuals To
Community Structure." Ecological Modelling 148: 191-212.
Tschirhart, J. 2003. "Ecological Transfers Parallel Economic Markets in a General Equilibrium
Ecosystem Model." Journal of Bioeconomics, 5, 193-214.
Tschirhart, J. 2004. "A New Adaptive System Approach to Predator-prey Modeling" Ecological
Modelling, 176, 255-276.
Weitzman, M. L. 1993. "What to Preserve? An Application of Diversity Theory to Crane
Conservation," Quarterly Journal of Economics, CVIII, 157-183.
46
-------
22
Table 1
Value of Benchmark Variables, in Million $
Variable
Value
Kp
365.608
Lp
293.567
k
m AK
cql
^ m
COl
CTaf
CTr
CTAC
s
If
Ir
Ic
^F
^R
^C
11263.681
307.325
9625.415
190.064
24.443
737.244
19925.646
201.764
7.638
15.554
178.572
627.340
907.968
9900.331
Regional Capital
Endowment
Foreign Capital
Regional Labor
Endowment
Foreign Labor
Household Fish
Demand
Household Recreation
Demand
Household Composite
Goods Demand
Household Savings
Fishery Investment
Recreation Investment
Composite Goods
Investment
Fish Exports
Recreation Exports
Composite Goods
Exports
47
-------
Table 2
Discounted Cumulative Welfare Impacts
23
Welfare
Measure
Value
of
Leisure
Quota
Rule
50 Year Horizon
(Million 1997 $)
100 Year Horizon
(Million 1997 $)
100%
30%
$1,117.77
$1,210.54
Wage
170%
-$7,811.23
-$8,665.10
Pev
75%
30%
$1,530.77
$1,674.54
Wage
170%
-$7,334.98
-$8,129.02
50%
30%
$1,943.77
$2,138.54
Wage
170%
-$6,858.73
-$7,592.94
48
-------
24
Table 3. Ecosystem Valuation Per Percentage Change in Ecosystem Inputs:
Welfare
Measure
Value of
Leisure
Quota
Rule
Average Annual Welfare Change
Per 1 % Change in Ecosystem Inputs:
Linked Model - Non-Linked(1997 $)
EVt
100%
30%
$109,626.43
170%
$114,458.27
75%
30%
$109,677.71
170%
$114,493.98
50%
30%
$109,728.99
170%
$114,529.69
49
-------
25
Figure 1 Economy Ecosystem Interaction
Commodity
Markets
Households
Factor
Markets
>3
s:
£
>3
O
&
o
o
o
L)
O
Regulator
* AiS
•> Monetary Flows
-> Real Flows
¦ > Ecosystem Flows
Non Consumptive
Ecosystem Flows
Killer Whale
Pollock
Zooplankton
\i if W
K
\
• \
• i
Phytoplankton
Sea Otter
Sea Urchin
Kelp
50
-------
26
Figure 2 Ecosystem Populations
Pollock
3.6E + 09
-------
Killer Whale
r\
1000 —v ^
\ '
950 \
V/
900
850
1966 1976 1986 1996 2006 2016 2026 2036 2046
Year
Benchmark 30% Quota
1 70% Quota
-------
Figure 3
Fishery Intra-season Rents
28
53
-------
Integrated Modeling and Ecological Valuation
David Brookshire (PI)
Julie Stromberg (Co-PI)
Arriana Brand, Jaiiie Chermak
Bonnie Colby, Mark Dixon, David Goodrich
John Loom is. Thomas Maddock
Holly Richter, Steven Stewart (Co-PI)
Jennifer Thacher
The University of New Mexico
Demographic Changes: Population Has Grown
Fastest in the West, Particularly in the "Public
Land States"
Percent Change in Resident Population for the 48 States and the District
of Columbia: 1990 to 2000
USCiNSUSBUKEAl
- Darker areas
denote faster
growth rates.
- Nevada (66%)
and Arizona
(40%) lead the
nation.
- Intermountain
states average
about 30%.
4^
The University of New Mexico
54
-------
Average Inches of Annual Precipitation in the
United States 1961-1990
"itv
%
m
Average Annua!
Predpttalion
0-5 Inches
5- 10
Sows®: USQA-NHCS htipihmtw nwmcs owJa-gcvijjnsm neni
pt mi
I he University of New Mexici
55
-------
Study Areas
San Pedro
River Region
The University of New Mexico
Bosque Del Apache
National Wildlife Refuge
SAHRA/UA/UNM Research
CP' M|
The University of New Mexico
56
-------
Objectives
• Integrate a hydrologic, vegetation, avian and economic models into an
integrated framework
• Couple natural science information with socio-behavioral information
to better understanding of both intra-site and inter-site data transfer
functions
• Determine the value of changes in ecological systems caused by
changes in hydrological profiles
• Introduce the non-market derived demand for water into an integrated
demand management framework
Scenarios And Study Framework
• 2-anthropogenic and 2-climatic
• Use two stated preference techniques for valuation
- Contingent valuation and choice models
• 3 information gradients
-Fully integrated science models
-2 Transferable indexes
• 2 test sites through an internet-based visualization survey
• San Pedro River and Rio Grande
The University of New Mexico
57
-------
-
San Pedro
Grande
Alternative Sites to Bosque Del Apache
Alternative Sites:
Site Attributes:
• Rio Grande Nature Center -
• Riparian areas without the
Albuquerque
extensive wetland
• Randal Davey Nature
influence
Center - Santa Fe
• Gila, Animas and
• Riverside Nature Center -
McKitrrick Creek are un-
Farmington
damned
• The Nature Conservancy's
Bear Mountain Lodge -
• All have similar issues
with urban competition for
Silver City
water
• Guadalupe Canyon -
• Lower visitation
Douglas, AZ
• McKitrrick Canyon -
Carlsbad
The University of New Mexico
58
-------
Why Two Study Sites?
Issues Of Transferability
•Don't always have the ability to collect original
data.
-Timeframe
-Lacking of basic science
•Cost of a primary studies ( both economic and
scientific) may be greater than value added from
improved accuracy.
The University of New Mexico
Data Transfer Issues
• What is the information we take from other
disciplines in our economic analysis?
• Does this information itself rely on some type of
transfer?
• What predefined conditions, if any, is the policy
analyst to assume?
The University of New Mexico
59
-------
Alternative Futures Study
• Explores how urban growth and change in the rapidly
developing Upper San Pedro Basin might influence the
hydrology and biodiversity of the area.
• Evaluation of individual scenarios from the present time (1997-
2000) to 20 years in the future (2020).
• Provides information to stakeholders in the area regarding
issues and planning choices, and their possible consequences.
• Alternative Futures study conducted by Department of Defense,
Desert Research Institute, Harvard Graduate School of Design,
Distant Past Near Past Present Near Futures Distant Futures
Alternative Futures
Carl Stcinitz and Allan Shearer
Harvard Graduate School of Design. 2001
60
-------
The Alternative Futures
° PLANS — based on interpretation of the current Arizona and Sonora
plans, and a forecast population of 95,000 in 2020 in the Arizona
portion of the study area.
° CONSTRAINED - assumes lower than forecast population growth in
Arizona. Development is concentrated in existing developed
areas.
° OPEN- assumes higher than forecast population growth in Arizona,
with major reductions of development control. Sonora remains as
forecast.
The University of New Mexico
Focus Groups
Unique aspect of this study
Necessary to bridge the gap between the specialized
science of scientists and general
knowledge/perception of public
Groups will help determine what information is
collected and how that correlates with water
management changes
Groups will identify important attributes
- Those will be measured by natural scientists and include^^
in the choice model ^
The University of New Mexico
61
-------
Valuation and Visualization Projections
• Web-enabled science-based Economic Valuation and
Visualization System (EVVS)
- Geographic information technologies
• Geo-databases
- Geographic information system (GIS)
- Internet-enabled mapping technologies
- Integrating technologies
The University of New Mexico
Simulated Streamflow of the San Pedro River Flow
(1940-2020)
40000
35000
33 30000
co
— 25000
cn
| 20000
Q
3 15000
0
u.
1 10000
ca
5000
0
0 10 20 30 40 50 60 70 ' 80 90 100 110 120 130 140 150 160 170 180
upstream ^ Approximate Stream Length (km) ^ Dowrearear
——1940 ¦ 1960 1980 ——2000 —*—Plans —*—Open Constrained
The University of New Mexico
62
-------
Groundwater Groundwater Groundwater
Cons: Impact, 2000-2020 Plans: Impact, 2000-2020 Open: Impact, 2000-2020
Two Stated-Preference Methods
Choice models (CM)
Dichotomous choice contingent valuation models
(CVM)
We will examine:
- Convergent validity for single attribute and policy (multiple
attribute)
- Valuati on across methods
- Conduct traditional tests of scope and embedding
- Examine differences between on-site and Internet survey
formats
„
The University of New Mexico
63
-------
Treatments
Sample Size
In
Total
Model
Internet
Person
Surveys
San Pedro
IM CM
300
100
400
Index CM
150
50
200
Trad CM
150
50
200
San Pedro
IM CM
600
200
800
Index CM
300
100
400
Trad CM
300
100
400
Bosque Del Apache
(Or Similar Site)
IM CM
150
50
200
Index CM
150
50
200
Totals
2100
700
2800
IM - Integrated
Model
CM- Choice Model
T rad - T raditional
Model - (Not
Significantly
Anchored In
Science)
Index- "Off The
Shelf' Scientific
Information
CVM - Contingent
Valuation
Hi
The University of New Mexici
Model
fff \\\
The University of New Mexicc
64
-------
Hydrology Component
• Analysis of anthropogenic and climate changes in San
Pedro River Region
• Alternative Futures Study (AFS)
• Hydrological Models
- GIS-Based Modflow
- Water Runoff
- Penman-Monteith Riparian Evapotranspiration
The University of New Mexico
Riparian Component
• How changes in surface flow and ground water levels
effect
- Riparian vegetation distribution
- Composition
- Structure
• Use existing information and backward linkages to:
- Identify relationships
- Develop reach-scale indices
- Model finer-scale, patch level riparian vegetation change in
response to scenarios
The University of New Mexico
65
-------
Coarse-scale Vegetation Modeling
• Divided river into 14 reaches
based on physical characteristics
\
1
I
mi
• Determined relationships between
vegetation traits (e.g., cottonwood
vs. tamarisk abundance) and site
hydrology
/
&if
IP
• Classified reaches into 3
condition classes (dry,
intermediate, wet) based on
vegetation traits (bioindicators)
indicative of site hydrology
• Will project reach-scale
vegetation changes based on
changes in condition class under
V*
>JS\
"11
Condition Classes
1 ' Dry
I 2 - Intermediate
I | 3-Wet
San Pedro River 2002
/%/Wet
A/Drv
| | Reaches SI
/\J Roads ^
W-L
y
alternative futures
Bird Component
Objective
- To determine the impact of vegetation changes on bird
populations and communities for differing type of
reaches of the SPR
- Provide characteristics of
bird abundance, productivity,
richness, and diversity
The University of Now Mexici
66
-------
Coarse-Scale Bird Modeling
Classified bird sampling locations by
3 condition classes (following
Stromberg et al.) and 5 habitats
(cottonwood, mesquite, salt cedar,
grassland, and desert scrub).
Quantified avian abundance, nest
success, and species richness as a
function of habitat / condition class.
Can project
habitat/condition
class changes
affecting birds
under alternative
futures.
San Pedro River 2002
A/Wet
WDfy
I | Reaches 1
/\y Roads ^
The University of New Mexico
Major Issues
1. How do individuals value marginal changes in
indices of ecosystem health and can such indices be
used as proxies for specific benefits?
2. Which benefits contribute most directly to human
well-being, what are their relative values, and what
are the most efficient methods of valuing them?
3. What is the tradeoff between the accuracy associated
with more detailed benefit transfers and the more
costly information necessary to provide them?
4. To what extent can simpler "reduced form" transfer
functions mitigate inaccuracies?
The University of New Mexico
67
-------
Expected Results For Determining Derived
Demand For Water
• A fully integrated valuation framework using the best
science and alternative valuation methods.
• Methodological insights into non-market valuation
techniques.
• Alternative data transfer functions that rely upon
alternative information gradients.
• Non-market water demand valuation functions for
integrated modeling.
Thank You
-------
Comments on the Papers by Finoff and Tschirhart and Hammitt and von Stackelberg
R. David Simpson
NCEE
A good place to start my discussion of these innovative and stimulating
contributions may be by repeating the commonplace observation that economics and
ecology both stem from the same Greek root: oikos, meaning house or household.
Ecology is the study of "nature's household"; economics, that of the society's. Before
Ernst Hackel coined the term oecologie in 1869, though, the disciplines were even more
closely tied in their terminology. The natural historian Gilbert White wrote in 1789 that "
. . . nature, who is a great economist, converts the recreation of one animal to the support
of another!" and "The most insignificant insects and reptiles are of much more
consequence, and have much more influence in the Economy [of] nature, than the
incurious are aware of'.
Contacts between the disciplines continued. Charles Darwin credited his
reading of Thomas Malthus's lissay on Population for providing the insight that
motivates the "survival of the fittest" in The Origin of Species, while Karl Marx is said to
have intended to dedicate one of the volumes of Das Kapital to Darwin—at least until
Engels persuaded him to identify a less bourgeois inspiration.
Cross-fertilization continued into the next century, including John Maynard-
Smith's application of game theory in the spirit of John von Neuman and Oskar
Morgenstern and John Nash to animal behavior, and Richard Nelson and Sidney Winter's
explorations of evolutionary principles in economics.
So, the first of the papers I discuss has a long list of illustrious antecedents—
and in whose distinguished company, I would venture to say, it is not misplaced. David
Finoff and John Tschirhart's work is insightful and innovative. If I ask some questions
and note some possible limitations in what follows, my remarks are intended in no way to
temper my general impression that their work is original and valuable.
I must confess that while reading a paper combining fundamental principles of
general equilibrium analysis with an application to large Alaskan marine mammals my
imagination began to work overtime (see figures la and lb). As it did, though, perhaps
the couplet I composed captured some essential element of Finoff and Tschirhart's
analysis:
The time has come, the Walras said, to speak of new devices
For getting rich, by sparing fish, when Joules mark their prices
69
-------
Figure la
Marie-Esprit Leon Walras
1834-1910
Figure lb
Odobenus Romarus
The Pacific Walrus
While I do think it's ingenious to think of the ecological "prices" of fish and
other marine organisms being calibrated in Joules, I do have some nagging doubts about
the procedure. I might express the first by asking "Are we likely to be any better at doing
their stuff than they are at doing oursT The "we" and the "they" are, respectively,
economists and ecologists. While I've discussed some of the overlaps between the fields
in my opening remarks, some recent forays by ecologists into economics have not been
met with favor by many economists. A 1997 effort by Robert Costanza and numerous
coauthors was roundly criticized for confusing marginal and average notions, failing to
appreciate the limitations of ability to pay on willingness to pay, and as Michael Toman
famously remarked, offering "A serious underestimate of infinity." A generation later the
ecologist Howard Odum proposed measuring the value of ecosystem services by the
energy required to perform them. This proposal betrays, as Partha Dasgupta has noted,
an apparent want of familiarity with Paul Samuelson's nonsubstitution theorem. The
nonsubstitution theorem demonstrates that reducing values to a common metric reflecting
the contribution of any single input is, in general, impossible.
I'm also a little concerned with an attempt to describe the functioning of one
complex system—a marine ecosystem—by analogy to another—a general competitive
equilibrium—when students of the former have already developed an elaborate
description of at least some of its mechanics. I'm referring to the evolutionary paradigm
in which the fit survive and replicate themselves. I'll also note in passing that it's always
seemed to me that evolutionary arguments in economics founder on the absence of a
mechanism of inheritance. There is no, or at best a very poor, analog in economics to the
role of genetics in biology.
Having expressed my doubts about Finoff and Tschirhart's use of energy as if it
were the objective of a biological system, I should also say that they have certainly not
fabricated it from whole cloth. The authors include many citations to biologists who
have made similar assumptions, and note especially that it's a staple of the "optimal
foraging literature" that describes creatures' feeding habits and the tradeoffs they make
70
-------
between gathering food and risking predation while doing so. Nor, I should add, do
Finoff and Tschirhart fall into Odum's error of supposing that values can be represented
by equivalent energy inputs; in Finoff and Tschirhart, energy is the objective to be
maximized, not the fundamental unit of account in which all inputs and outputs are to be
measured.1
Still, I can't help hearing my mother's voice saying "Eat your vegetables!" What
I mean is that a balanced diet requires a mixture of foods; choosing our own diets to
maximize solely calories within linear budget constraints would have us subsisting solely
on lard, albeit, presumably, not for long before our arteries clogged. In short, and in
economic terms, Finoff and Tschirhart ask us to suppose that "biological production
functions" exhibit straight-line isoquants whose slopes are determined entirely by relative
calorie contents of available food sources. This may be a reasonable approximation, but I
guess I'd be more comfortable if I felt I knew the biology a little better.
I have another concern with Finoff and Tschirhart's analysis, although I suspect
that it revolves around little more than an expositional suggestion to clarify the notion in
their paper. The classic proof of general competitive equilibrium in economics
demonstrates that there exists a set of prices at which all supplies and demands balance.
The limitations of this theorem are well known. It is silent on how these prices are
determined. What happens if there is some departure from such prices? My
understanding is that the question has never adequately been resolved, but most of the
economics profession believes that the intuitive notions that prices go up when demand
exceeds supply and decline otherwise provide a workable depiction of the process of
reaching equilibrium. Is there a similar process driving convergence to ecological
equilibrium in Finoff and Tschirhart's work? "Prices" are determined by energy content,
and so it is not clear how these prices would adjust to clear the "markets" for species.
Presumably the energy cost of seeking a certain prey decreases in that species abundance,
but I'd like to see more explanation of this.
One mechanism for equilibration is apparent: species that acquire more energy
through preying on others than they expend in predation or lose by becoming prey
themselves thrive. In this respect, population growth is to positive net energy flux as
industrial entry is to supernormal profits. Processes of industrial entry may not be wholly
adequate for eliminating supernormal profits when barriers exist to entry, however. I'm
curious as to whether similar concerns might apply in the biological model.
1 Somewhat ironically, and wholly by coincidence, between the time I presented my oral comments at the
workshop and writing them up now I picked up a copy of Robert Nadeau's recent book The Wealth of
Nature. Nadeau, following Philip Mirowski, develops the thesis that neoclassical economics is modeled on
19th century physics. In Nadeau and Mirowski's view, neoclassical economists did little save relabel the
variables in models of physics, and the analogy between utility and energy in their respective disciplines is
exact. Without commenting as to the validity of his critique, I'll simply say that Nadeau might find it
ironic to see economists now structuring a model of biology in which energy is now the analog to utility.
71
-------
My remarks thus far have largely focused on the question of whether the net energy
flux Finoff and Tschirhart use to motivate their model is the "right" objective. This is a
relative matter, though. "Right" for what? I can see several potential purposes:
• Describing population dynamics.
• Calculating and explaining the relative abundance of species.
• Developing real-world policy advice.
It seems that, in order adequately to serve the last of these purposes, the model would
have to be fairly closely calibrated to biological data and demonstrate an ability closely to
explain and predict them. It is no criticism of a model of so complex a phenomenon to
suggest that the model has not yet done this. It's a very, very hard problem!
Thus, I don't think Finoff and Tschirhart's contribution is sufficiently refined as yet
to form the basis for concise policy guidance. I might also note in passing that various
elements of the problem over and above the fundamental biology make the analysis even
more difficult. Consider, for example, fisheries policy: to what combination of political
and social factors do regulators respond, and how effective are their regulations? These
considerations make the development of policy advice even more complicated.
As another example of policy interactions, Finoff and Tschirhart note that fishermen
may earn quasi-rents in that their labor is seasonal. It's difficult, then, to know how to
evaluate their earnings and opportunity costs of time. As a personal aside, I grew up in a
small fishing town on Puget Sound in Washington State. The area had been settled at the
turn of the 19th century by Serbo-Croatian immigrants whose descendents resembled
professional basketball player Vlade Divac in appearance, stature, and, in some instances,
athletic ability. One guy who was a few years older than me—and whom we all envied
greatly—spent his summers fishing in the Gulf of Alaska. This paid him well enough
that he drove an expensive sports car. He attended college, on a basketball scholarship,
through the other seasons. The rumor around town was that he was also collecting
unemployment benefits since he was unable to fish during basketball season but he was,
or so he claimed, prepared to quit school and the team if he could find employment. So,
the interaction between labor and fishery policies might affect results as well! To be fair,
Finoff and Tschirhart demonstrate that their results are not sensitive to wage rate
assumptions. Still, any number of interrelated policy interactions might affect the
analysis. The collective uncertainties introduced at all the many stages of the analysis
might be enough to propagate cumulative errors large enough to preclude precise policy
guidance.
If Finoff and Tschirhart's analysis is not (yet) well enough grounded to provide
concise policy advice, what are its chief merits? I think they lie in providing a concise
and thought-provoking paradigm for thinking about interrelated social and ecological
systems. Whether or not a biological approach modeled after economic equilibrium is
accurate, there's tremendous value simply in communicating the possibility of
approaching problems in this way and providing concrete instances of parallels and
differences between biological and social systems.
72
-------
Another virtue of Finoff and Tschirhart's approach may be that it facilitates
"informationally dense," for want of a better term, interpretations of a system's state.
Economists often make much of the informational role of prices: they tell us everything
we need to know about preferences, costs of production, future prospects, etc. The
"energy prices" of Finoff and Tschirhart's model may perform a similar role of
describing the state of the biological system. Another analogy comes to mind. Whether
or not a particular animal is eaten by another is, of course, a random event, and can't be
accurately predicted. Yet on the aggregate level we should be able to say something
about general patterns of predation and their consequences for relative abundance. Just
as the logistic model of infection generates aggregate regularities from a host of
individually stochastic events, the Finoff and Tschirhart paper may usefully reduce
complex stochastic phenomena to the compact summary statistics of "prices".
Let me conclude my discussion of Finoff and Tschirhart with one final reservation,
however. The paper is concerned with what I might describe as "most-of-the-time"
behavior on a path toward a steady-state equilibrium. Our greatest social concern with
biological systems often involves rare stochastic events, however. How vulnerable are
such systems to climate change? To invasive species? It is, of course, difficult to predict
the response to an unpredictable shock, but another useful direction for research might be
to consider the "resilience"—to borrow a loaded term from the ecological literature—of
such systems.
The paper by James Hammitt and Katarina von Stackelberg takes up some similarly
complex issues of ecological and economic interactions using a very different set of tools.
I might also note that, because Hammitt and von Stackelberg have only been working on
a difficult problem for a limited time, the paper I had to review is still preliminary and
incomplete.2 I will, then, be commenting on their procedures rather than their results.
My first comment is simply to re-emphasize that Hammitt and von Stackelberg are
discussing extremely difficult issues. What are the effects of PCB contamination on
human health and ecosystem functioning, and how do people form values regarding these
effects? If, as Hammitt and von Stackelberg propose, such values are to be elicited by
asking stated preference questions, the questions will necessarily be very complex.
The authors are off to an impressive start in formulating the questions. I've never
read a paper that discusses, within the span of a few pages, concepts as variesd as "PCB
congenors . . . chloronated in the ortho . . . meta and para positions"; "... the induction
of cytochrome P450 enzymes . . ."and "the McCarthy Scales of Children's Abilities, . . .
the Peabody Picture Vocabulary Test-Revised and Buss and Plomin Emotionality
Activity Sociability Temperament Survey for Children . . . and the Wechsler Intelligence
Scale for Children-Revised (WISC-R)"! This breadth of coverage is both impressive to
the reader and challenging for the authors. They will need to knit together the analysis in
such a way as to make it cogent to people who are not experts in all fields (who include
2 I might also note that the project reported at the workshop by David Brookshire was still too recent for
there to have been any written material for me to review.
73
-------
just about everyone except the authors), as well as to reduce these issues to terms that
will be comprehensible to their stated preference survey respondents.
While it may always be implicit, perhaps I should mention explicitly before what I say
next that "the opinions expressed here are those of the discussant and not necessarily
those of the Agency ..." It seems to me that two questions can be asked of any exercise
in valuation, and of stated preference approaches in particular.
1. Are the specific answers generated useful for making policy? And
2. Are we learning more about the validity, reliability, and transferability of the
methods employed?
It is difficult to see how one can posit a positive answer to the first without presenting
more evidence as to the second.
From this perspective, Hammitt and von Stackelberg are building in interesting and
useful tests. By approaching the same phenomenon of PCB contamination from a variety
of angles, they are building in cross-checks as to the validity of each. Of particular
interest may be their proposal to alternate the order in which they ask questions
concerning human health and ecological effects. In theory it shouldn't matter if a
respondent is asked first about one and then the other, or if the order is reversed. It will
be interesting to see if this novel twist on the "embedding" question yields the anticipated
result.3
It's interesting to see researchers grappling with such challenging questions. One
way of thinking about the types of things that Hammitt and von Stackelberg are
considering is that they're exploring public attitudes toward experience goods with which
people have little or any experience. The severe health consequences of PCB exposure
will, for most of us, be encountered once in our lifetimes and at the end of our lives at
that. The ecological consequences of PCB are even further from the realm of things with
which most people have had any experience.
I'm reminded of something I've heard Dan Bromley say—though I should caution
that I can't claim to be quoting exactly, but rather, to the best of my recollection: "If you
think prices come from markets, you probably believe milk comes from plastic bottles."
I think what he means by this is that the properties we attribute to prices—and some
economists assert attributes of foresight and rationality that border on the magical—arise
in institutional and cognitive circumstances that have often evolved over long periods of
time. What does it mean to be eliciting "prices" respondents profess to be willing to pay
absent the trappings of a market? I have no answer to propose, but suspect that we might
better inform and refine our attempts at valuation if we could incorporate into our
analysis an understanding of what motivates the formation of markets and what we can
infer from their absence.
3 I continue to be somewhat surprised that the consistency tests proposed by Peter Diamond in his 1996
Journal of Environmental Economics and Management seem to have attracted as little attention as they
have.
74
-------
EPA Valuation of Ecological Benefits: Improving the Science Behind Policy Decisions
- A STAR Progress Review Workshop.
Session IV
Juha Siikamaki
Fellow, Resources for the Future
Discussion of two presentations:
"Joint Determination in General Equilibrium Ecology/Economy Model"
by David Finnoff and John Tschirhart and
"Contingent Valuation for Ecological and Non-Cancer Effects within an Integrated Human
Health and Ecological Risk Assessment Model"
by James Hammitt and Katherine von Stackelberg.
"Joint Determination in General Equilibrium Ecology/Economy Model"
This paper by David Finnoff and John Tschirhart is part of their research program, in
which the authors have developed a joint ecology/economy general equilibrium framework. This
modeling framework suggests concepts to modeling ecology/economy interactions, which makes
the research effort particularly challenging and ambitious.
The economic component of the joint equilibrium framework by Finnoff and Tschirhart
comprises a standard computable equilibrium model of the economic sector, in this case, the
fishing industry. The ecological component (GEEM, general equilibrium ecosystem model) of the
framework extends the economic equilibrium model to modeling ecosystems. While the
economic model comprises individual agents (consumers and firms), which maximize their
objective functions (utility, profit), the ecosystem model comprises different species, which
maximize their net energy intake. Ecosystems are organized as hierarchical food webs, which are
assumed to transfer energy between different trophic levels via "energy markets," conceptually
much the same way as goods and the factors of production are transferred within the economy.
Each period, all energy markets are required to clear, bringing the ecosystem to equilibrium.
My discussion highlights the ecological modeling tradition1 and makes an attempt to
place the research by Finnoff and Tschirhart into a larger context. I will discuss different
ecological modeling approaches by focusing on their origins.
Most mathematical models in ecology relevant to the Finnoff and Tschirhart paper are
population-level models. The first such models were introduced well before ecology became an
established discipline. Malthus' prediction of the human population growth introduced an
exponential growth model in the late 1700s. A few decades later in the 1830s, Verhulst
incorporated a carrying capacity constraint in the exponential growth model and thereby
formulated the widely used logistic growth model, which has an S-shaped population growth
curve. In the 1920s and '30s, physicists Lotka and Volterra developed sophisticated mathematical
models of, among other things, species competition and predator-prey interactions. These models,
as well as the host-parasite models by Nicholson and Bailey (1935), are still some of the
fundamental mathematical models in ecology. The approach taken by Lotka and Volterra
influenced economics (for example, Samuelson notes their work in the introduction of his
1 See, for example, Begon et al. 1996, Edelstein-Keshet 1988, Real et al. 1991.
75
-------
textbook), and their models of competition were also introduced to economics. May (1976 and
elsewhere) demonstrated that simple deterministic models could drive complicated and even
chaotic behavior, with completely different outcomes driven by slight differences in initial
conditions and population parameters. Current themes in population ecology involve developing
complex dynamic models, in particular, meta-population models (Hanski 1999).
Individual-level ecological models relevant to the Finnoff-Tschirhart paper deal with
optimal foraging behavior. These models draw directly from economics and explain animals'
strategies to exploit resources most efficiently. MacArthur and Pianka (1966) first adapted the
economic model to a patch choice problem, proposing that foraging decisions are based on the
relative benefits and costs of foraging on alternative patches. A rich consequent literature on
foraging behavior has thereafter addressed time spent foraging on each patch, nutritional
constraints, learning and memory, risk and stochastic factors, as well as fitness and the genetic
base of behavior, and other factors. Central to the foraging studies is the assumed relationship
between the behavior of organisms and their net energy intake. Optimal foraging studies often
view organism's rate of energy intake as the proxy of evolutionary fitness.
The concept of ecosystem, which was introduced by Tansley (1935) and Lindeman
(1942), emphasizes that biotic community is not differentiated from its abiotic environment and
used the term "ecosystem" to denote the biological community integrated with the abiotic
environment. Ecosystem is viewed as the fundamental ecological unit, and organisms within an
ecosystem may be grouped into a series of discrete trophic levels. The key processes structuring
ecosystems across spatial and temporal scales are tested and discussed, for example, by Holling
(1992) and Holling et al. (1995).
I will now proceed to discussing the paper by Finnoff and Tschirhart.
The ecological scales of GEEM and research questions
The suitable model type and its complexity depend on the research question. In this case,
the modeling effort seeks to identify prices and quantities of final goods and the factors of
production, predator and prey biomass consumption, and species populations in the ecosystem.
The ecological questions are, therefore, population-level questions. However, the GEEM is an
individual-level model, which is aggregated to the ecosystem level by employing representative
organisms/animals (the ecological counterparts of representative consumer/firm). Alternatively,
one could combine the economic model with a mainstream dynamic population model from
ecology. Multi-species dynamic models are demanding, but using numerical simulation models
and software may facilitate complex modeling efforts. The question then becomes which
modeling approach is more useful and accurate: the joint ecology/economy equilibrium model or
a "mainstream" bioeconomic model (see, e.g., Clark 1990, Sanchirico and Wilen 2001).
Since an altogether new modeling approach is being introduced, it would be valuable to
see a comparison of alternative modeling approaches applied to the same problem. Different
models could be compared relative to their predictive power. Such a comparison would help
determine which types of modeling efforts the GEEM is suitable for, and on the other hand,
which modeling may be best carried out by using mainstream bioeconomic models.
Data needs
The data requirements of the GEEM differ from bioeconomic models. Therefore, it
would be useful to assess the GEEM also from the perspective of data requirements and quality.
Some questions used in assessing the model could be: What are the minimum data requirements
for implementing the GEEM and how reliable are these data relative to data used in bioeconomic
models? What are the possible gains or losses in modeling accuracy associated with data sources
and quality?
76
-------
Ecosystem equilibrium
The GEEM is an equilibrium model, which clears all energy markets every period. In the
long run, the ecosystem directs itself towards the steady state. Ecological systems, however, are
often perplexed by non-linearity, complex feedback loops, and multiple or no equilibrium. An
application, which would examine how to modify the GEEM to handle these issues, would be
welcome.
Spatial scale
The relevant spatial scale may vary among different trophic levels, and depend, on the
mobility of different species or other factors. For example, sea urchins, sea otters, and killer
whales each operate at different spatial scales. The GEEM may be modified to account for
varying spatial scales, perhaps by viewing different meta-populations as separate "species" in the
same trophic level. On the other hand, meta-population models in ecology have been developed
specifically for handling spatially differentiated populations.
"Contingent Valuation for Ecological and Non-Cancer Effects within an Integrated Human
Health and Ecological Risk Assessment Model"
This paper by James Hammitt and Katherine von Stackelberg describes a research project
for valuing the ecological and human health effects of PCB contamination. Their research is
currently ongoing, with the survey involved nearly ready to be launched. My comments that
follow address mostly general issues related to the valuation of ecological and human health
risks.
Hammitt and von Stackelberg use a dichotomous choice contingent valuation method for
the valuation of both ecological and health endpoints of PCB contamination. The ecological
endpoints involve two alternative valuation endpoints: the effects of PCB contamination on a
high-profile species (bald eagle) versus its effects on a group of species (species sensitivity
distribution). The human health endpoints consist the effects of PCB on developmental outcomes
(IQ, reading comprehension). The exposure pathway causing the adverse effects is the ingestion
of fish.
Non-market valuation studies typically do not address uncertainties inherent in the
evaluated policy outcomes. Several factors contribute to this tendency. First, uncertainties often
stem from very complex relationships and scientific phenomena, which are complicated to
communicate effectively to survey respondents. Second, the expression of uncertainties as
probabilities has been a long-standing challenge in stated preference surveys (Hammitt and
Graham 1999). Third, uncertainties in policy outcomes can cause respondents to question the
scientific credibility of the scenarios presented to them.
Hammitt and von Stackelberg address the uncertainty of ecological outcomes by using
the species sensitivity distribution (SSD) approach to describe the distribution of reproductive
effects of PCB contamination for all species. The ecological basis of the SSD approach is
appealing, but the involved graphs are cognitively challenging. It will be interesting to see how
well using the SSD works in the valuation context. The survey also uses graphs with colored dots
for intuitively easy illustrations of the risk probabilities.
The survey will help estimate the value per statistical life (VSL) for children, which is an
important research question for which practically no information currently exists. Children are
especially sensitive to the effects of pollution and estimating VSL for children would therefore be
an important contribution and useful in policy evaluations.
Willingness to pay for both ecological and human health risk reductions will be
determined by respondents' risks perceptions, which, in turn, will reflect information available to
the respondents both in the survey and prior to it. Therefore, it would be interesting if the survey
77
-------
could collect information on the actual risk perceptions of the respondents. Quantitative measures
of the perceived risks may be hard to attain, but the survey could collect data on at least the extent
of ecological and human health effects respondents consider in answering the valuation questions.
References
Begon, M., J.L. Harper, and C. R. Towsend, (1996). Ecology: Individual, Populations, and
Communities, Blackwell Science, Third Edition, p. 1068.
Clark, C. W. (1990), Mathematical Bioeconomics: the Optimal Management of Renewable
Resources, Second Edition, John Wiley & Sons.
Edelstein-Keshet, L. (1998). Mathematical Models in Biology, McGraw Hill, p. 586.
Hanski, I., (1999). Metapopulation Ecology, Oxford University Press, p. 313.
Hammitt, J.K. and J.D. Graham, (1999). Willingness to Pay for Health Protection: Inadequate
Sensitivity to Probability? Journal of Risk and Uncertainty 18(1): 33-62.
Holling, C.S., (1992). Cross-Scale Morphology, Geometry and Dynamics of Ecosystems.
Ecological Monographs. 62(4): 447-5 02.
Holling, C.S., D.W. Schindler, B. W. Walker, and J. Roughgarden, (1995). Biodiversity in the
Functioning of Ecosystems: An Ecological Synthesis. In: Perrings, C. K-G Maler, C.
Folke, C. S. Holling, B-O. Jansson, Biodiversity Loss, Economic and Ecological Issues,
Cambridge University Press.
Lindemann, R. L., (1942). The Trophic-Dynamic Aspect of Ecology, Ecology 23:399-418.
Lotka, A.J., (1925). Elements of Physical Biology. Baltimore: Williams and Wilkins.
MacArthur, R. H. and E. R. Pianka, (1966). On Optimal Use of a Patchy Environment, The
American Naturalist 100: 603-09.
May, R. M., (1976). Simple Mathematical Models with Very Complicated Dynamics, Nature,
269, 471-477.
Nicholson, A. J. and Bailey, V. A., (1935). The Balance of Animal Populations, Part 1,
Proceedings of Zoological Society of London, 3:551-598.
Real, L. A. and J. H. Brown, editors, (1991). Foundations of Ecology, Ecological Society of
America.
Sanchirico, J. N., and Wilen J. E. Bioeconomics of Spatial Exploitation in a Patchy Environment,
Journal of Environmental Economics and Management, 37: 129-150
Tansley, A. G., (1935). The Use and Abuse of Vegetational Concepts and Terms, Ecology 16:
284-307.
Volterra, V., (1926). Fluctuations in Abundance of a Species Considered Mathematically, Nature
118:558-60.
78
-------
Summary of the Q&A Discussion Following Session IV
Amy Ando (University of Illinois at Urbana-Champaign)
Directing her comment to David Brookshire, Dr. Ando said, "Yesterday, Nancy
Bockstael made the astute observation that individual decisions, in your case water use—
people watering their bluegrass—don't tend to feedback directly to them. It seems to me
that the main method of feedback in your case will be policy—changes in water prices or
rules that you can't have bluegrass or at least can't water it. Are you planning on having
policy be endogenous in your study?"
Turning to John Tschirhart and David Finnoff and saying that her comment related to a
few of the points made by the discussants, Dr. Ando said, "Let me make a quick analogy
to economics. It may be profit-maximizing for a farmer to adopt a new cost-saving
technology, but if all the farmers adopt the cost-saving technology, the supply curve
shifts down and the price falls. I think you had at least two such micro but then macro
feedback [options] and I can't tell from a 25-minute talk whether you're accounting for
them. One, as David mentions, is energy prices. If a seal lion avoids being eaten, that's
good for the sea lion—the population rises—but that lowers the energy cost (energy
price) of eating a sea lion, so the killer whales may move more to prey upon sea lions.
The second one I was thinking about—again, both of your discussants alluded to this, is
fitness. Why on earth did we reintroduce wolves into vast parts of the continental U.S.?
One argument was that it actually benefits prey populations and improves their fitness.
So, if a sea lion avoids being preyed upon, that's good for the sea lion but may be bad for
the population, because maybe it was a crummy, unfit sea lion that ought to have been
eaten anyways. And, do we really want to get rid of all the killer whales? One argument
against that might be that it would reduce the fitness of the two prey populations. That
may be a hard thing to have data for, but I'm wondering if you can address those kinds of
issues."
David Brookshire (University of New Mexico)
Dr. Brookshire first explained that he didn't have a paper prepared for the workshop
because his funding came in just a couple of months ago. He then responded to the
question by saying, "Nancy's point of view is very interesting. Actually, I see it a little
differently than she does. I think at the individual level we don't get feedback on our
decisions. Actually, it's not even on our radar—let me give you an example: If you take
a 10-minute shower thirty times in Albuquerque using a 2.5-gallon-per-minute shower
head (that's a water saver), as a commodity charge that would cost you $1.09. This is
spare change in the parking lot." Referring to the decision process involved in pricing
water in the West, he went on to say that this level of individual water use is "not on our
radar, so to speak. However, collectively, it is on our radar. And this is where you have
your stakeholder groups, and again, for instance, if I may use my hometown of
Albuquerque, we're very conscious of the fact (and most people actually know this,
believe it or not) that we're mining ground water at a rate of approximately 70,000 acre
feet per year and that we'll have a short fall in the year 2035. We are compounding daily
the problem with the collective implications. So, on the one hand, at the individual level
79
-------
we don't get the signals that we as economists would like people to be seeing. Actually,
we're doing some work on experimental estimation of what will happen in terms of urban
demand at higher price levels, using both a lab and the water bills for every household in
Albuquerque for the last seven years. ... So, we'll have some idea how people would
respond if the prices were higher, but if you look at the literature, you don't see that
anywhere—you see the traditional administrative cost prices."
Addressing Dr. Ando's question more directly, Dr. Brookshire continued, "In terms of
the policy being endogenous, I don't know. I don't know exactly how we're going to do
some of this. We have an upcoming meeting with the San Pedro Partnership. To some
extent, we have to work with our local folks. How they want us to bring forth the
Harvard study and other kinds of things remains to be seen at this point. It's a
possibility—it's a good thought—but that's all I can leave you with at this point."
David Finnoff (University of Wyoming)
Dr. Finnoff responded to Dr. Ando by saying, "By the example that you used, let's say
we were to cull killer whales. That would lower the predation of the sea lions, and with
less predation the population would increase," which would result in increased
intraspecies competition among sea lions for their food supply. He concluded by saying,
"So as the population goes up, the price rises and demand will eventually go down, so
you'll have a population growth, but that will then be limited by this intraspecies
competition. It's the way that we treat our ecosystem exchanges (or "transfers," as we
call them) that allows these threads to be captured."
John Tschirhart (University of Wyoming)
Dr. Tschirhart stated, "We don't make any welfare judgments about what's good or bad
with respect to the ecosystem, whether it's good to have this many more sea lions or
something like that, other than how it might affect the economy. In fact, we start off with
calibrating the ecosystem without any humans whatsoever and call that the natural,
steady state. Then we introduce the economy—the economy is not self standing; the
economy cannot survive without the ecosystem, because in this case you need the food,
whereas the ecosystem does very well without humans." (scattered laughter)
Dan Phaneuf (North Carolina State University)
Posing his question to David Finnoff and John Tschirhart, Dr. Phaneuf said, "The
economic general equilibrium model is apparently aspatial, correct?—it assumes that
there's perfect integration of markets across space and it abstracts away from the notion
that there might be differences across landscape. I thought I'd mention that... the
ecosystem general equilibrium model is also aspatial in the way you've thought about it
thus far, and I think that that's a necessary condition for what you're looking at. I'm
wondering what you lose from going in that direction. In the economic general
equilibrium model we have a good sense of what we'd lose, and we're usually willing to
assume that there's the kind of arbitrage that sort of makes prices the same across space.
Is that going to be a reasonable assumption in what you guys are looking at, and if not,
what are we losing because of that?"
80
-------
David Finnoff
Dr. Finnoff responded, "Well, we have a little of space, in the sense that we have two
legs of the food web in different regions, but within those regions, we model" those with
no concept of space. In a project I'm working on with Kerry (Smith) right now we are
adding space, and so migration" between regions .
Dan Phaneuf
Dr. Phaneuf stated that he saw similar problems between Finnoff and Tschirhart's work
and Smith's study in dealing with the space issue. He characterized Finnoff and
Tschirhart's work as "using more careful structural modeling that abstracts more from
space."
David Finnoff
Dr. Finnoff added, that this relates to the question regarding "whether we're trying to run
two kinds of modeling frameworks in a parallel fashion to answer the same kind of
question." He said it always comes down to looking at what a potential model "brings to
the table that a standard macro population dynamic model doesn't, and vice versa."
John Tschirhart
Dr. Tschirhart clarified, "Basically, you could have killer whales feeding in two different
areas, and which area are they going to go to?—They're going to go to where the prices
are lower, and when they move from one area to another, they're going to cause those
prices to increase and then there's a tendency to move back to the first place." So, there
is interaction through price with regards to space.
Nancy Bockstael (University of Maryland)
Offering a follow-up comment on Amy Ando and David Brookshire's exchange, Dr.
Bockstael stated, "Actually the system I was talking about yesterday is one in which there
isn't that strong public feedback. It's where residential development is putting a lot of
pressure along the East Coast—water quality, stream ecology, and such—but there isn't
any sense of that and it's not affecting the housing market except to the extent that it
induces policy responses in some feeble effort to reconfigure how the land use changes."
Dr. Bockstael continued, "Concerning your point about policy being endogenous, it
seems to me that in that context policy is endogenous, but there's a long lag. It is such a
long lag that it almost doesn't pay to view it as endogenous except to the extent that all of
the ecologists want to spew out these scenarios over 50 years." She went on to state that
these extended projections are senseless exercises because "the structure is going to
change in the system so much because of the induced policy changes" that we can't
anticipate all the effects and ramifications. She concluded by saying that this is a big
difference between the ecologists and economists, who "understand that there will indeed
be induced policy changes but who really aren't sure what form they would take."
81
-------
Amy Ando
Dr. Ando commented to David Brookshire: "Considering your idea that people have a
collective sense of "Yes, we're drawing down the water"—in a public choice model, that
feeds into a policy change. So, I think that these are not entirely different things."
Nancy Bockstael
Addressing John Tschirhart, Dr. Bockstael stated, "It seems to me in my encounters with
ecologists that there's a lot of resistance to belief in the equilibrium. I may be wrong, but
that's the way I feel. Loosely speaking, when we're thinking about equilibrium . . . we're
thinking about it in two ways, really: There's sort of a market equilibrium—prices adjust
to equate supply and demand . . . but then there's this steady state equilibrium in a
dynamic system. That's the more interesting part, I think, about the equilibrium in your
model—that you're thinking in terms of steady state. My sense is that ecologists would
say, "Well, we never get there, so why even talk about it?" But shocks aren't
infrequent—they're very frequent, and they're environmental shocks. I wonder if you've
encountered that criticism and whether there's a way to think about this issue in a way
that would be more pleasing to ecologists in terms of introducing random shocks so that
you don't have this steady state equilibrium that you characterize in the system."
John Tschirhart
Dr. Tschirhart answered, "Yes, we definitely have gone into that. They don't talk about
general equilibrium, of course—they talk about steady states, and they have gone away
from the steady state type idea in more recent times. But, we have another model, for
example, in which we have just plants. Modeling plants without animals is okay because
that's where it all starts. What we do, very simply, is we have temperature as the random
variable, so the system is constantly being jostled from heading toward one steady state
to another. That seems somewhat satisfying, so we are trying to bring that into account."
Stephen Swallow (University of Rhode Island)
Also addressing David Finnoff and John Tschirhart, Dr. Swallow commented, "When I
looked at your paper in the Journal of Biological Theory a while ago, I had some
questions about capturing the ////raspedfie competition for prey and I wondered whether
you were capturing wterspecific data. As an analogy, when you have a lot of firms, they
are all competing for the same labor pool, whereas in the ecology model you showed us
today it appeared that the consumers of prey were competing "within their own industry"
for that prey, but not necessarily between the two species for the same prey. As you go to
looking at a larger number of species, I imagine that an extra level of detail would enter
in." He closed by asking whether the researchers had "any insights as to whether that's
actually going to be a significant problem in your next modeling step and, if it is, how
you might handle it."
82
-------
John Tschirhart
Dr. Tschirhart replied, "That's a really good observation—we've thought about doing it
both ways. The issue would be, for instance, if you have, as we do in our expanded
model," a predator preying not just on one but on two species. He stated that this raises
the issue of deciding whether it's one market—in which the predator is paying one price
for the two different species—or two markets. Dr. Tschirhart clarified that "in
economics, it would be just one market—one price for both." He added, "We didn't do
that because it just seemed to work better in the simulations to have separate markets
between each pair of species, but it's a good question." He acknowledged, though, that it
would be reasonable to make it one price because the price is actually measured "per
kilogram of fish flesh," thereby eliminating any species size differential that would
influence species-specific prices. Dr. Tschirhart closed by saying that they are continuing
to look at both options and haven't determined yet which is preferable. He added that
they hadn't yet talked to ecologists about the details of this issue because ecologists "are
so unused to seeing this whole methodology that getting down to those kinds of details
just hasn't happened yet."
Kerry Smith (North Carolina State University)
Dr. Smith commented, "We didn't get much of a look at the utility function on the
economic part of the model and this is sort of a detailed question: Is the value derived
from recognition separable or non-separable from the quantity and type of species of
fish? In other words, what I'm looking for is a feedback effect, such that the rate of
leisure choice associated with looking is influenced by the availability of these species to
look at." He said when you put that feedback back in, "you get a couple of different
loops in the model that connect the economic structure with the biological structure."
David Finnoff
Dr. Finnoff responded, "No, it's not like that right now. . . . Essentially, this was our test
model . . . One thing I'd like to say, though, is that our leisure choice is essentially
regulator determined in this model."
Will Wheeler (U.S. EPA/NCER)
Dr. Wheeler said he didn't have a question but wanted to clarify something for the
audience, and he commented, "It's part of the STAR Grant policy that all grantees come
in once a year to present their work and we have violated that policy. That's why when
we have a conference on ecovaluation, we have all our ecovalution grantees come in.
That's why, although David's grant was just awarded this year, we still had to make him
come in so we don't violate our policy quite so bad as we have. Actually, the other two
grantees work began only last year, so that's why we're seeing work in progress—we're
still trying to tweak everything. We appreciate everyone's work.
83
-------
Bob Reilly (Virginia Commonwealth University)
Dr. Reilly stated, "I'm still a little bit troubled with these energy prices in the general
equilibrium model. It seems to me that the way things are at the moment these prices are
just market-clearing prices affecting differences in excess demand. In point of fact,
really, these energies are functional relationships. Some of them are easily movable,
possibly, and others move as a result of density interactions in the species. It's nothing
like moving the market-clearing prices in the economic market. It's hard for me to
believe that an ecologist is going to sign off on this kind of a view of moving energy
prices when they are very well aware what it takes to handle and absorb this energy price
is very different from the movement of an energy price when in fact you have to worry
about spatial issues and search costs and a lot of things that appear to be central in this
problem. . . . Those relationships are difficult to nail down and the literature is very poor
on how does energy cost vary as a result of density effects. As far as I know, there's not
a lot of good development literature there. . . .
David Finnoff
Dr. Finnoff answered by saying, "I definitely agree. One thing in our model is that our
respiration function is a physiological function that really governs the transfer of energy
into heat production and whatnot. So, a lot of what you're talking about takes place
within our respiration function itself. What we think is unique about this energy price
rate is that the system together helps determine outcomes—and not just a functional
relationship for one species to another, but there's something that's endogenous to the
interactions of all these individuals. That's really what we add to the table here, because
there are input-output kinds of models in ecology. But, we think that this system-wide
interaction is important and that's what these prices are allowing us to bring to the table."
David Brookshire
Dr. Brookshire said, "I can't let this issue on policies go. . . .This is like saying, "let the
market do it. What market? What does it look like?"—these kinds of things. We use the
phrase "heterogeneous preferences" in economics. That takes on real meaning when we
look at water issues in the West—line everybody up and you'll get 2x number of ideas.
So, when we talk about policies, we're really talking about chipping at the market, and
Nancy (Bockstael) alluded to that." He used an example of putting in low-flow toilets
only to discover that everyone then double-flushes and water consumption goes up—or
installing low-flow shower heads, and then everybody takes longer showers. Dr.
Brookshire went on to say that "we haven't really come up with a set of systematic
policies that account for the human behavior side of this issue. We would like to see
prices raised, which is a popular and unpopular thing to say, depending on who you speak
with. Our typical policies are short-term: we ration; we jawbone through conservation;
we don't raise prices, even though we should—the reason I gave you that commodity
chart is to give you a clue as to just how cheap it really is. So, it's not clear what to do in
the San Pedro area. In fact, it is very clear that if you go to the San Pedro area and you
ask, "What should be the water policy?" you will get multiple answers, and that's why
it's so difficult and why the San Pedro Partnership exists."
84
-------
Dr. Brookshire said that they are looking for "some guidance from this group" (the San
Pedro Partnership) because they "need some anchor so that we're not discredited for
"inventing" things, if you will—we're going to have to get some buy-in. But to think that
there's going to be a water policy for the San Pedro region is mythical" and he said this is
true for the West, in general.
In closing, Dr. Brookshire commented, "The one area that we are making progress in is
what we all know as "water banking"—the institution of re-allocating water on a spatial
and temporal basis, where we don't actually have to trade the property rights and so don't
run into adjudication problems. We are making progress in getting better use of what we
have through these institutions, and various states are interested. That's one of the few
kind of global policies that everybody seems to be at least willing to talk about."
END OF SESSION IV Q&A
85
-------
Valuation of Ecological Benefits: Improving the Science
Behind Policy Decisions
PROCEEDINGS OF
SESSION V: CONSERVATION AND URBAN GROWTH: FINDING THE
BALANCE
A WORKSHOP SPONSORED BY THE U.S. ENVIRONMENTAL PROTECTION
AGENCY'S NATIONAL CENTER FOR ENVIRONMENTAL ECONOMICS (NCEE)
AND NATIONAL CENTER FOR ENVIRONMENTAL RESEARCH (NCER)
October 26-27, 2004
Wyndham Washington Hotel
Washington, DC
Prepared by Alpha-Gamma Technologies, Inc.
4700 Falls of Neuse Road, Suite 350, Raleigh, NC 27609
-------
ACKNOWLEDGEMENTS
This report has been prepared by Alpha-Gamma Technologies, Inc. with funding from
the National Center for Environmental Economics (NCEE). Alpha-Gamma wishes to
thank NCEE's Cynthia Morgan and the Project Officer, Ronald Wiley, for their guidance
and assistance throughout this project.
DISCLAIMER
These proceedings are being distributed in the interest of increasing public understanding
and knowledge of the issues discussed at the workshop and have been prepared
independently of the workshop. Although the proceedings have been funded in part by
the United States Environmental Protection Agency under Contract No. 68-W-01-055 to
Alpha-Gamma Technologies, Inc., the contents of this document may not necessarily
reflect the views of the Agency and no official endorsement should be inferred.
11
-------
TABLE OF CONTENTS
Session V: Conservation and Urban Growth: Finding the Balance
Applying Economic and Ecological Principles to Identify Conservation
Reserves for Vernal Pools with Residential Development
Dana Marie Bauer, Yong Jiang, Stephen K. Swallow, Peter Paton and
Dennis Skidds, University of Rhode Island 1
Spatial Analysis of Private Land Conservation Behavior
Amy W. Ando, University of Illinois Urbana-Champaign; Heidi Albers,
Oregon State University 47
Pixels in Place of Parcels: Modeling Urban Growth Using Information
Derived From Satellite Imagery
Rich Iovanna, U.S. EPA, National Center for Environmental Economics;
Colin Vance, German Aerospace Center, Institute of Transportation
Research 81
Discussant
Sabrina Lovell, U.S. EPA, National Center for Environmental Economics 116
Discussant
Andrew J. Plantinga, Oregon State University 121
Summary of Q&A Discussion Following Session V 131
in
-------
Applying Economic and Ecological Principles to Identify Conservation Reserves for Vernal
Pools with Residential Development
By
Dana Marie Bauer,* Yong Jiang,* Stephen K. Swallow,* Peter Paton,** Dennis Skidds**
*Department of Environmental and Natural Resource Economics
"Department of Natural Resources Science
University of Rhode Island
Kingston, RI 02881
Keywords: open space, metapopulations, amphibian, reserve design, land conservation, wetland
Residential development or exurban development in rural communities is a common cause of
fragmentation of ecological habitats at a landscape scale. Existing wetlands protection
regulations commonly protect wetlands from direct development, sometimes extending
protection to include a buffer zone around delineated wetland, but without sufficient ecological
consideration of the linkage between wetland habitat patches and the matrix of uplands. Such
regulatory approaches may doom some species, such as metapopulations of amphibians, to
localized extinction, because the common approach preserves breeding habitat without
preserving sufficient habitat critical to species life-stages outside the breeding season. In a
separate literature, scientists, often involving economists, have been developing methods for
selecting lands for priority protection relative to objectives focused broadly on conservation of
biological diversity. Many of these studies have focused on identifying sites, and networks of
conservation reserves, that protect a target number of endangered or sensitive species at least
cost. However, the many of these studies have not incorporated ecological processes and
resulting spatial linkages that may influence the long-term sustainability of biological diversity in
the conservation reserves.
This presentation will consider amphibian metapopulation dynamics as a indicator of or proxy
for ecosystem integrity. (Metapopulations are populations of a species that are comprised of
many subpopulations that exist separate habitat patches, but which must exchange breeders in
order to sustain the overall population. Subpopulations in habitat patches my become extinct and
the patches may be recolonized from other subpopulations.) The analyses discussed will use
concepts from production theory (the production function) and opportunity costs of development
to identify optimal land conservation reserves, and intensity of residential development. The
research uses the framework of the Hanski-Ovaskainen spatially explicit form of Levins'
metapopulation model to identify an objective for ecological quality, which is then maximized
subject to a cost constraint. One aspect of the study places the framework in a theoretically
complete context, with applications requiring either detailed data or transfer of key parameters
from existing literature; this portion of the study may be better suited for developing intuitive
guidance for policy. An alternative portion of the study uses relatively easily obtained biological
data to develop a proxy for metapopulation production to establish spatially explicit priorities for
land conservation suitable for policy in real watersheds. Analyses include variation in the
1
-------
opportunity cost of land development, distance between habitat patches, and size of
subpopulations in habitat patches, as well as the role of the matrix of undeveloped land around
habitat patches.
OVERVIEW:
The paper consists of two parts. In the first part, we present a simple land use model constrained
by an ecological goal to protect amphibian populations in the face of development. The
empirical simulation is based on a relatively complete ecological production function relating
amphibian populations to protection of habitat patches and the matrix of land surrounding these
patches through a jurisdiction. The approach facilitates a preliminary assessment of actual
wetland protection policies, some proposed policies promoted by conservation groups in some
states, and some alternatives to these proposals. The results presented remain based on basic
assumptions and substantial simplifications from the actual heterogeneity that exists in terms of
the value of land for either ecological purposes or for development. It is indicative of the types
of analysis to be done as the project continues. The perspective of this approach is expected to
be most suitable to establishing uniform policies (regulations or incentive-based policies to be
evaluated in future work) that would apply across a jurisdiction.
The second part of the paper considers the question of balancing ecological conservation and
development from the perspective of identifying a land conservation network that would obtain
an optimum performance on ecological criteria subject to the cost of preserving land. Land
preservation could be interpreted as the purchase of land or development rights, and its cost is
therefore equivalent to the foregone value of development opportunities on land that is
preserved. This part of the paper develops a model for the ecological performance of a land
conservation reserve that can be based on relatively easily available biological data while also
recognizing some of the major biological-behavioral processes that affect the ability of a species
to survive across a landscape in the long term. The application uses data on egg-mass counts as
a proxy for population estimates of amphibians breeding in vernal pools (seasonally-flooded
wetlands), and uses this proxy to represent key features of a metapopulation model that would be
applied by ecologists to model amphibian populations. Results represent a novel approach to
incorporating bioprocess-based elements of spatial factors into the selection of a conservation
lands network. The approach could support planning by municiple or regional officials to
identify land parcels to target for conservation, while allowing residential development to
proceed elsewhere.
2
-------
PART One: Economic Consequences of Conserving Amphibian Metapopulations within
Areas of Urban Sprawl
INTRODUCTION
Development in rural fringe communities is occurring in a random and sprawling fashion,
potentially damaging healthy ecosystems (Heimlich and Anderson 2001; Daniels 1999).
Environmental impacts of development include loss, degradation, and fragmentation of wildlife
habitat, increased air and water pollution, increased soil erosion, and decreased aesthetic appeal
of the landscape (Johnson 2001). Current land use policies rarely incorporate features of
landscape-scale ecosystem health (Burke and Gibbons 1995; Miltner, White, and Yoder 2004;
Willson and Dorcas 2003). For example, wetland policies focus on protection of individual
wetlands, but at the same time provide incentives for higher-intensity development of upland
habitat (Hardie et al. 2000; Swallow 1994; Semlitsch 1998; Semlitsch and Bodie 2003). Many
wetland species, such as pond-breeding amphibians, spend much of their life histories in these
upland habitats either over-wintering or dispersing to other wetland habitats (Semlitsch 2000).
Development of upland areas decreases the long-term viability of these species by reducing the
quantity and quality of upland habitat and decreasing dispersal success (Arnold and Gibbons
1996; Lehtinen, Galatowitsch, and Tester 1999; Vos et al. 2001; Vos and Chardon 1998;
Woodford and Meyer 2003). Development affects amphibians, in particular, by destroying
upland habitat, changing the hydroperiod of the pond, adding pollutants to both wetland and
upland environments, and creating barriers to dispersal (Woodford and Meyer 2003).
This paper investigates the long-term ecological impacts of development in exurban
communities using pond-breeding amphibians as indicators of ecosystem health. Amphibians
are good indicators of ecological stress due to roads and other forms of urbanization (Lofvenhaft,
3
-------
Runborg, and Sjogren-Gulve 2003; Trombulak and Frissell 2000; Woodford and Meyer 2003).
Amphibians are also considered to be particularly good ecological indicators for wetland
ecosystems (Keddy, Lee, and Wisheu 1993; Hecnar and M'Closkey 1996). Several
characteristics of amphibians make them good indicators of ecosystem health. First, they are
extremely sensitive to changes in their environments (Vitt et al. 1990; Welsh and Droege 2001;
Welsh and Ollivier 1998). The skin of amphibians at all life-history stages is permeable to water
and thus many types of pollutants. Eggs are covered by a thin layer of gelatinous material so
they are directly exposed to the aquatic environment and adults of many species spend their lives
physically against mud, sand, or leaf litter. Second, amphibians are central to many wetland
ecosystems with their relatively high position in the food chain and biomass that often exceeds
that of all other vertebrates combined (Welsh and Droege 2001). Third, pond-breeding
amphibians are thought to exist in metapopulations, where each pond has its own local
population but dispersal of individuals between ponds occurs annually and is critical to
sustaining the overall population (Marsh and Trenham 2001). For example, although adult wood
frogs are incredibly site-faithful, approximately 18% of juvenile wood frogs leave their natal
pond and disperse across the landscape to settle in another pond (Berven and Grudzien 1990).
Fourth, many amphibian species are common and widespread (Calhoun and Klemens 2002).
The persistence of common species across the landscape are considered by some to be more
crucial to ecosystem health (King 1993).
This paper develops an empirically-based conceptual model that combines economic and
ecological principles to determine the optimal allocation of land between development and
preservation uses and then compares the optimal solution to specific land allocations resulting
4
-------
from current and alternative land use policies. Results show that current regulations protecting
only the aquatic environment ultimately lead to extinction of amphibian metapopulations in areas
of urban sprawl. Land use policies such as environmental impact fees, transferable development
rights (TDRs), or cluster developments may be better alternatives.
METAPOPTTT ATTON THEORY
Natural landscapes are patchy, with each patch supporting different types of flora and fauna.
Human activities such as building construction or timber harvesting contribute additional levels
of fragmentation to natural environments. Metapopulations consist of groups of local
subpopulations distributed throughout a patchy environment, with each subpopulation occupying
its own patch (Hanski 1999). Local subpopulations exchange individuals through a dispersal
process whereby a small number of individuals leave the patch and join a new subpopulation.
Local subpopulations can go extinct and patches can be re-colonized without threatening the
overall viability of the entire metapopulation. The status of species in a regional context may be
determined more by metapopulation dynamics than by local birth and death processes (Hecnar
and M'Closkey 1996).
Amphibian spatial dynamics resemble classical metapopulation models, in which subpopulations
in breeding ponds blink in and out of existence and extinction and colonization rates are
functions of pond size and spatial arrangement in addition to species-specific characteristics
(Marsh and Trenham 2001; Green 2003). This "ponds-as-patches" view of metapopulation
dynamics has been used in many prior amphibian studies and is used here (Carlson and
5
-------
Edenhamn 2000; Gill 1978; Pope, Fahrig, and Merriam 2000; Sjogren-Gulve 1994; Vos, Ter
Braak, and Nieuwenhuizen 2000).
The Classical Levins Model
The classical Levins metapopulation model views a metapopulation as a population of local
populations inhabiting an infinite number of identical patches (Hanski 1999; Levins 1969, 1970).
All patches are the same size, the same quality, and equally connected to all other patches.
Colonization is not affected by the distance between patches. The Levins model is an occupancy
model, based on presence or absence of the species, rather than a count model, based on number
of individuals. The model assumes that local patch dynamics can be ignored. The Levins
metapopulation model, given by
dP/dt = cP(l-P) - eP,
measures the rate of change in "metapopulation size", where P is the fraction of patches that are
occupied at time t and c and e are species-specific colonization and extinction rate parameters,
respectively. The colonization and extinction rates can be estimated with time-series occupancy
data. For example, the colonization rate can be calculated as the ratio of the number of years the
species was absent but present the next year to the total number of years the species was absent
(Gilpin and Diamond 1981). The extinction rate can be calculated in a similar manner.
Metapopulation persistence occurs when there is a balance between local extinctions and
recolonizations. The steady-state equilibrium value of patch occupancy is given by
P* = 1 - e/c.
If e/c > 1, the metapopulation goes extinct.
6
-------
A Spatially-Realistic Model
A spatially-realistic metapopulation model developed by Ilkka Hanski and others extends the
Levins model by allowing patch areas and distances between ponds to vary according to a
realistic landscape structure (Hanski 1999; Hanski and Gyllenberg 1997; Hanski and Ovaskainen
2000; Moilanen and Hanski 1998; Moilanen and Nieminen 2002; Ovaskainen 2003; Ovaskainen
and Hanski 2003, 2001; Hanski and Ovaskainen 2003). In this finite-patch metapopulation
model, the change in probability that any given patch is occupied is a function of local
colonization and extinction rates that are different for each patch. It has been observed that in
comparing different connectivity measures in their ability to predict colonization events, the best
and most consistent performance is found for a measure that takes into account the size of the
focal patch and the sizes of and distances to all potential source populations (Moilanen and
Nieminen 2002). In the spatially-realistic model, the rate of change in the probability of patch i
being occupied, dP;/dt, is given by a system of N equations for a network of N patches:
dPi/dt = C; (P)(l-Pi) - Ei(P)Pi, i=l,...,N
where colonization and extinction rates are a function of P, which is the vector of the N
occupancy probabilities. The equilibrium probability of occupancy, P;*, also called the
"incidence" of the species in patch i, depends on the probability of persistence in all other
patches (Hanski 1994, 1999):
Pi* = Ci/(Ci + Ei)
The colonization rate of patch i, C;, is function of the N patch areas, A;, and the spatial location
of patch i within the network given by the dispersal kernal, f(d;j):
C, = c A, X A| Pj f(dij) for all j^i
7
-------
where the dispersal kernal f(.) accounts for the effect that increasing distance dij between habitat
patches i and j reduces the rate of recolonization of patch i from patch j when j is occupied. The
patch-specific colonization rate can be interpreted as the sum of contributions toward
colonization from each of the other N-l patches:
Cij = c A; Aj f(dij) for all i^j.
The exponential form of the dispersal kernel, f(dy) = exp(-ady), is commonly used and indicates
that the greater the distance between two patches, the smaller the contribution to re-colonization.
The parameter, a, reflects the dispersal ability of the focal species (1/a is the average migration
distance). The probability of colonization C; increases with more patches, larger patch sizes, and
shorter distances between patches.
The extinction rate of patch i, E;, is a function of the area of patch i and the species-specific
extinction rate:
E; = e/A;.
Extinction rates vary as an inverse function of area, because larger patches usually mean larger
local populations and risks of extinction tend to decrease with larger local populations (Gilpin
and Diamond 1976).
An NxN landscape structure matrix, L, is derived from the previous colonization and extinction
equations where each element of the matrix is a function of patch areas and the dispersal kernel:
Lij = (e/c) (Cij/Ei) = A;ex • AT • A/111 • e"adlJ for i*j
Lij = 0 for i=j
8
-------
Each element gives the contribution that patch j makes to the colonization rate of patch i when
patch i is empty, multiplied by the expected lifetime of patch i when it is occupied (Ovaskainen
and Hanski 2003).
Patch areas are scaled by extinction, immigration, and emigration factors (ex, im, and em) that
are specific to the focal species (Hanski and Ovaskainen 2003). Empirical studies have shown
these parameters to vary widely, from a minimum of 0.05 to a maximum of 2.30 [(Ovaskainen
2002) pg. 428-430], For a "typical" metapopulation, the sum of the three scaling factors would
fall between 1.0 and 2.0 [(Ovaskainen 2002) p. 430], The model used in the conceptual
framework developed here expands the dispersal kernel to reflect the additional barriers to
dispersal that result from development:
L4 = A," ¦ A,'m • A™ < I-H . e"™"j
where the barrier to dispersal between any two patches, By, is a function of the percentage of
land that is developed. Thus, the greater the barrier between two patches, the smaller the
contribution of those patches towards long-term persistence of the species.
From the probability of occupancy and the landscape matrix, two constructs for comparing or
ranking different landscapes can be derived (Hanski and Ovaskainen 2000; Ovaskainen and
Hanski 2003, 2001). The metapopulation persistence capacity, or metapopulation capacity for
short, is a measure of the landscape's ability to support a viable metapopulation over the long
term. It is similar to the carrying capacity in a single-population model. It takes into account
both the quantity of habitat available and the spatial configuration of the habitat patch network.
A species is predicted to persist in a landscape if the metapopulation capacity of that landscape is
greater than a critical threshold determined by characteristics of the focal species. The larger the
metapopulation capacity the greater the long-term probability of persistence. Therefore, the
-------
metapopulation capacity can be used to rank different landscapes in terms of their capacity to
support viable metapopulations. It is possible to calculate how the metapopulation capacity is
changed by removing habitat fragments from or adding new habitat fragments to specific spatial
locations. It is also possible to calculate the effect on metapopulation capacity caused by
increasing or decreasing patch areas. Increases in the number or areas of patches results in an
increase in the metapopulation capacity, while an increase in the distances between patches
results in a decrease in the metapopulation capacity.
Mathematically, the metapopulation capacity, K, is the leading eigenvalue of the non-negative
landscape matrix, L (Hanski and Ovaskainen 2000; Ovaskainen and Hanski 2001).
Metapopulation capacities can be considered as simple sums of the contributions from individual
patches, given by the elements of the leading eigenvector. Habitat destruction, habitat
deterioration, and increased dispersal barriers all lower the metapopulation capacity of the patch
network. The effect of gradual habitat deterioration or gradual increases in dispersal barriers is
given by the derivative of K with respect to patch attributes and may be evaluated by sensitivity
analysis (Ovaskainen and Hanski 2003, 2001). In contrast, destruction of entire patches leads to
a rank modification of matrix L, the effect of which on K may be derived from eigenvector-
eigenvalue relations (Ovaskainen 2003). Metapopulation capacity defines the threshold
condition for long-term metapopulation persistence as:
K > 5=e/c.
The second theoretical construct developed from the spatially-realistic metapopulation model,
the "size" of a metapopulation, S, is a measure of the "average" patch occupancy (Ovaskainen
10
-------
and Hanski 2003). It's value reflects the rarity or commonness of the species in the given patch
network. Metapopulation size, given by
S = 1 - (8 /K),
shows a direct relationship between the metapopulation capacity of a particular habitat patch
network and the metapopulation size. The larger the metapopulation capacity, the larger the
metapopulation size. Values of metapopulation size range between 0 and 1, with values closer to
zero corresponding to rare species and values closer to 1 corresponding to common species. The
choice in a particular analysis between metapopulation capacity and metapopulation size
depends on the question being asked (Ovaskainen and Hanski 2003).
LAND ALLOCATION MODEL
A simplistic land allocation model would attempt to maximize the sum of benefits from both
development and preservation land uses. In this type of model, the optimal quantity of land
allocated to each land use is determined by equating the marginal benefits. A number of
approaches have been used to incorporate ecological "values" into economic analyses. For
example, hedonic housing studies, recreational travel-cost models, and contingent valuation
surveys have been used to estimate values of non-market public goods such as open space (Bates
and Santerre 2001; Geoghegan 2002; Irwin 2002; Johnston et al. 2001; Lutzenhiser and Netusil
2001; Rosenberger and Loomis 1999). Unfortunately, attempts to quantify the entire economic
value of ecosystem services are often difficult to acquire or, when obtained, are unreliable or met
with substantial controversy (Swallow 1996; Toman 1998; Berrens 1996). Because of the
difficulty in fully measuring all the non-market benefits of ecosystem health, an alternative "safe
minimum standard" (SMS) approach may be used (Bishop 1978; Randall and Farmer 1995;
11
-------
Ciriacy-Wantrup 1952; Farmer and Randall 1998). With the SMS approach, a government
agency, on behalf of society and based on recommendations from the EPA and other scientific
advisory boards, establishes a standard or constraint that guarantees a particular level of safety.
For example, an SMS approach is used by the Clean Water and Clean Air Acts, whereby various
pollutants are not allowed to exceed given levels. As society increases its understanding of
ecological processes and environmental conditions, standards are modified (strengthened or
relaxed) to reflect this new information.
One way of modeling a safe minimum standard is through the use of an ecological constraint.
Ecological constraints have been used in the modeling of both renewable and non-renewable
resources (Albers 1996; Marshall, Homans, and Haight 2000; Roan and Martin 1996; Yang et al.
2003). Albers presents a model for economic management of tropical forests that uses
ecological constraints to reflect the spatial interactions across forest plots and the irreversibility
of some forest land uses. The explicit recognition of the varied uses of forested land, spatial
interdependence, irreversibility, and uncertainty leads to optimal patterns that have different
structures and more forested area that those recommended by traditional models lacking an
ecological constraint. Roan and Martin model mineral production and waste reclamation as joint
products subject to the traditional ore depletion constraint and an ecosystem constraint that limits
the amount of water pollution released. Reclamation is identified as the creation of additional
"environmental slack" or expansion of the capacity of the waste pile under the ecosystem
constraint. Results indicate that the mine will lose rent on the mineral product as the shadow
price of environmental slack increases. Yang et al. developed an integrated framework of
economic, environmental and GIS modeling to study cost-effective retirement of cropland to
12
-------
reduce sediment loading of local rivers by a set amount. The analysis suggests that program
costs are minimized when the abatement standard is set for the region rather than uniformly for
individual watersheds. Marshall et al. modeled warbler population dynamics as a function of
timber rotation length to find the rotation age that attains a predetermined critical population size
at the end of the management time horizon. Management cost is calculated as the opportunity
cost of not harvesting timber at the profit maximizing rotation length. Because different
management strategies were associated with different costs and with different outcome
extinction probabilities, it was possible to construct a marginal cost curve for the probability of
species survival. Results show that the desirable combination of management tools depends on
the safety margin (SMS) selected. In each of the above studies, the ecological constraint
provided insights that weren't available from the corresponding traditional model without the
constraint.
This study uses an optimization model that maximizes the benefits of residential development
subject to a series of constraints, including an SMS-type ecological constraint:
Maximize V(Q) = R • (Qi + Z Qi)
Subject to Qi < A;o
Qi ^ Lo
and S > Smin
The benefits from development, V(Q), are calculated by multiplying the land rent associated
with residential development, R, by the sum of the number of acres developed in each patch, Qi,
plus the number of acres developed in the intervening landscape, Qi. Land is assumed to be
13
-------
homogeneous from the perspective of the developer and the subsequent home-buyer, therefore
the per-acre land rent is the same for all acres. The scale of analysis used here assumes that land
values are constant and determined exogeneously. Development is assumed to be irreversible,
thus undeveloped land can be viewed conceptually as a non-renewable resource. The first two
constraints on the system represent this finite quantity of land, and we set the total available land
to 10,000 acres (corresponding approximately to a single jurisdiction in our case study below). It
is not possible to develop more land than what is originally available, Al0 in patch i and L0 in the
intervening landscape. The third constraint is the ecological constraint. It states that the
metapopulation size, S, (described in the previous section) cannot fall below the minimum level
set by society, Smin. It is expected that this third constraint drives the system and that as long as
the minimum size is set at a level high enough to prevent metapopulation extinction, some
amount of land will be preserved. Metapopulation size, S, was chosen over metapopulation
capacity, K, because it provides a better comparison measure for common species used as
indicators of ecosystem health.
DATA AND ANALYSIS
Vernal pool and other landscape data from the Rhode Island Geographic Information System
(RIGIS) were used to establish a realistic landscape structure for the application. Analysis was
based on data from the Wood-Pawcatuck watershed in western Rhode Island which included a
GIS coverage of all vernal pools in the watershed mapped from aerial photographs with
additional field data provided from local ecologists (Paton 2004). A 10,000-acre parcel of land
was selected from the middle of the watershed because it contained areas of both high and low
densities of pond occurrences and there were no major barriers to dispersal (Figure 1). For each
14
-------
of the 123 ponds in the parcel, pond area (range 0.0065 to 10.34 acres ; mean = .5; median =.14)
and distance to each of the other 122 ponds (range <30m to >8km; mean ~= median = 2.8km)
was obtained. From this data, an initial landscape structure matrix, L0, was generated using a
pond-as-patch approach for determining habitat patches. The initial patch area, A;0, includes the
area of the pond itself as well as a 750-foot buffer of upland habitat. Habitat quality is assumed
to be homogeneous and the initial landscape is completely undeveloped.
Land value data was provided by local tax assessors for towns in the Wood-Pawcatuck
watershed. An ordinary least squares (OLS) regression was estimated using a log-log functional
form with per-acre land value as the dependent variable and size of the parcel as the dependent
variable (R2=0.96). Linear and log-linear functional forms were also tried but did not perform as
well as the log-log model. A land rent value of $640 per acre was determined by entering the
largest parcel size (150 acres) into the estimated equation.
Optimization of the model was performed using MATLAB. Appropriate species-specific
parameter values were taken from the literature: 8=1.2, a=1000m, ex=im=em=0.5 (Berven and
Grudzien 1990; Hanski and Ovaskainen 2003; Marsh and Trenham 2001; Ovaskainen 2002).1 A
series of "optimal" land use allocations was produced by varying the ecological constraint
parameter, Smin, over its entire range (0.0-1.00). The term "optimal" refers to the solution from
the optimization program and is indicative of the best we could do in a perfect world, with
perfect information, and no transaction costs. The optimization assumes that all developed
1 Amphibian-specific parameter values for area scaling factors were difficult to find in the literature, so values were
extrapolated from those of other species. Sensitivity analysis performed on these parameters did not change the
qualitative results.
15
-------
parcels are one acre and that the entire parcel gets developed. Thus, developed parcels are
removed from their respective patches or intervening landscape and added to the corresponding
barrier. The opportunity costs are calculated as the foregone benefits associated with those acres,
out of the total 10,000 acres, that cannot be developed in order to achieve the safe minimum
standard. The opportunity cost can be viewed as an indication of the level of opposition that
developers will exert on town officials if new conservation policies are put in place.
The "optimal" solution set is compared to a variety of policy alternatives (Table 1). The first
three policy alternatives reflect current regulations. In Rhode Island, for example, only the
vernal pool itself is protected. Development is allowed right up to the pool's high water mark.
Other states' regulations protect a small buffer or envelope around the pool. Note that C2 uses 2-
acre development which reduces the barrier effect on dispersal by 50%. All other policy
alternatives assume 1-acre development. The fourth policy, G for "guidelines" is based on the
best development practices put forth by the Wildlife Conservation Society (Calhoun and
Klemens 2002). According to these guidelines, the pool and envelope (100-ft buffer) are
completely protected. In addition, 75% of the critical habitat (100-750 feet from pool edge) is
protected. Policies PI, P2, and P3 are modified versions of the vernal pool guidelines that allow
more development of critical habitat, but in the case of P3 protect a portion of the intervening
landscape. Policy 0 protects all 10,000 acres in the study area. For each policy, the amount of
land that must be preserved is determined and then the opportunity cost in terms of foregone
development benefits is calculated. The corresponding metapopulation size, S, for each policy
alternative is also calculated.
16
-------
RESULTS
Results from the series of optimizations are shown in Figure 2, which plots the opportunity costs
of foregone development (in $millions) against the metapopulation size, Smin.2 The curve
represents the tradeoff between the economic benefits of residential development and the
ecological benefits of land preservation. The key result is that it is possible to achieve a relatively
high metapopulation size at a relatively minimal cost. A 90% average probability of occupancy
can be achieved at a cost in terms of foregone development of less than $0.5 million. These low
costs are possible because the optimization eliminates small, isolated patches and large "empty"
landscapes first. This is consistent with one study that showed 89% of the variability in dispersal
success can be accounted for by differences in the size and isolation of forest patches, with closer
and larger patches having significantly greater exchange of dispersing organisms (Gustafson and
Gardner 1996). In addition, the "optimal" solution does not prohibit the complete destruction of
patch. Opportunity costs rise exponentially in order to increase metapopulation size from 0.95 to
0.99 as it becomes necessary to preserve more and more habitat patches.
Figure 3 shows the opportunity costs and metapopulation size corresponding to the various
policy alternatives. Current regulations (CI, C2, and E) fall to the far left of the graph. In the
policies protecting the pool-only (CI) and the pool plus a small buffer (E), the metapopulation
size falls below zero. Thus, with current regulations, amphibian metapopulations will eventually
go extinct. The policy protecting the pool-only with 2-acre development (C2) increases the
metapopulation size to 31% average probability of occupancy because of the lower dispersal
2 The x-axis contains values of Smin from 0.65 to 1.0 in order to emphasize the policy-relevant portion of the graph,
since our goal was to keep a common species common. The curve extends to the left downwards towards zero. In
fact, it is possible to go beyond a metapopulation size of 0, which is interpreted as the entire metapopulation going
extinct.
17
-------
barrier, but still not very high and not likely to receive support from conservationists. At the
other extreme, the no development policy (0) achieves a metapopulation size of .998 or a 99.8%
average probability of occupancy. However, the opportunity cost of achieving this goal is $6.4
million.
The recommended guidelines protecting 75% of critical habitat, G, achieve a very high
metapopulation size (0.99), but at a cost of close to $2 million. Comparing this to the "optimal"
solution, we could get the same level of long-term metapopulation persistence at less cost or,
alternatively, we could achieve a higher level of persistence at the same cost. Comparing the
guidelines, G, to policy PI that allows 50% of the critical habitat to be developed, we are still
able to achieve a relatively high metapopulation size (0.97) but at a cost savings of $600,000
($1.8m - $1.2m). Policy P2, that allows 75% of the critical habitat to be developed, results in a
metapopulation size of 0.89 at an even lower opportunity cost of $680,000.
Consideration of Figure 3 can provide some insight to further policy implications. First,
environmental managers (or, say, town planners and land trust officials) could consider results
such as those in Figure 3 to identify opportunities to improve either the ecological outcomes
anticipated in the long term at a given cost of foregone development or to improve (reduce) the
cost of foregone development that leads to a particular ecological outcome. Optimizing policy
design (such as Table 1) in a manner that moves a policy outcome horizontally or vertically in
Figure 3 would improve economic effectiveness. Certainly, since current regulations in the
study area will largely fail to maintain species persistence in the long term, a change to a policy
such as P2 (Table 1, Figure 3) could provide some degree of success in ecological criteria
18
-------
without a massive increase in cost (foregone development opportunity). Also, it is clear that
policy PI dominates policy P3 in the study area, because PI is better in terms of both cost and
ecological outcome. Moving diagonally down and to the right in the figure improves economic
effectiveness of wetland policy. Second, comparisons of policies PI, P2, and P3 may shed light
on how different policy elements will influence the cost-outcome relationship. Both PI and P3
may be seen as nesting P2 within themselves. While P2 and P3 both protect 25% of critical
habitat, P3 also protects 25% of the surrounding matrix. The improved ecological outcome of P3
comes at a cost of approximately one million in additional losses of development opportunities.
Relative to P2, PI does nothing more to protect the surrounding habitat matrix but PI does
increase the protection of the critical habitat zones to 50%. These comparisons suggest that
policy involves a balance between protection of the habitat zones and protection of the
surrounding matrix, with particular outcomes for this study area. However, in general, these
factors may be affected by such considerations as the definition of the critical habitat zone and
the natural distribution of habitat sites across the landscape. For example, with policy G (75%
protection of critical habitat zones), a very high level of metapopulation size is maintained, but at
about $700,000 additional cost relative to PI. However, if habitat sites were more widely
dispersed through the study area, matrix lands between habitat sites may become more of a
limiting factor. Indeed, some preliminary analyses not reported here shows that the definition of
critical habitat, as consisting of 750 meters around the vernal pool, results in substantial overlap
of habitat sites, leaving relatively little land in the classification of matrix between habitat sites.
A reduction in the critical habitat size could cause matrix lands to play a more substantial role in
the relative merits of alternative policies, such that policies that protect more matrix lands may
19
-------
dominate some policies focused on critical habitat zones. Thus, the present results should be
taken only as illustrations.
SUMMARY AND FUTURE DIRECTIONS
The results of this study show that the current regulatory environment will eventually lead to
extinction of amphibian metapopulations that are currently common species throughout the
landscape. In addition, the recommended vernal pool guidelines (policy G, Table 1, Figure 3)
maintain a high level of amphibian metapopulation persistence, and thus ecosystem health, but
are costly to society as a whole. Future analyses may evaluate the potential for alternative
policies, including market-based or incentive-based policies, to improve outcomes in terms of
either cost or ecological results.
The land allocation framework developed here was presented in its most simplistic
implementation where land is homogeneous from both the perspective of residential land
development and the perspective of amphibian habitat quality. The model can easily be adapted
to incorporate a variety of extensions. For example, heterogeneous quality of land from the
developers perspective can be captured in the objective function by including two or more types
of benefits based on a Ricardian land rent model. Different levels of habitat quality can also be
included in a spatially realistic metapopulation model by modifying the effect of patch areas on
the metapopulation capacity and metapopulation size by incorporating a habitat quality index
(Moilanen and Hanski 1998). A metapopulation study of the butterfly Speyeria nokomis
apacheana showed that neither occupancy nor turnover patterns were best modeled as functions
20
-------
of patch area or isolation (Fleishman et al. 2002). Instead, other measures of habitat quality
explained the most variance in occupancy and turnover.
Further analysis of initial conditions, model parameters, and functional forms will provide
additional insights into the optimal allocation of land between development and preservation.
The analysis performed here starts with a pristine (i.e., totally undeveloped) landscape. Further
analysis could investigate initial conditions where development has already taken some of the
high-quality development land. With the exception of policy alternative C2, the model presented
here assumes 1-acre development where the entire acre becomes impervious surface and
development is evenly distributed throughout a given patch. An investigation of more robust
dispersal barrier functions may be warranted. The analysis presented here focused on the
establishment of a healthy ecosystem as indicated by long-term persistence of a common species.
The model could also be used to identify the cost-effective reserve network for an endangered
species.
The issue of determining the appropriate scale of analysis for guaranteeing healthy ecosystems
"across the landscape" remains open. The "optimal" solution for the 10,000-acre parcel achieves
a large metapopulation (S=0.95) by preserving two relatively small sections (areas with lots of
well connected large patches) of the entire landscape. Because of the use of amphibians as the
indicator species (short dispersal, large number of connected patches), a better method might be
to use a two-stage approach where the first stage assesses metapopulation size, S, on a smaller
21
-------
scale (e.g., 1000 acres) and the second stage assessed the connectivity between 1000-acre
sections.
Finally, this entire analysis was conducted using a static optimization model. For certain
landscape scales, that incorporate the entire regional land market, land rents will increase over
time. Thus, a dynamic analysis may be appropriate for determining the optimal allocation of
land for an entire region.
22
-------
References
Albers, Heidi J. 1996. Modeling ecological constraints on tropical forest management: spatial
interdepence, irreversibility, and uncertainty. Journal of Environmental Economics and
Management 30:73-94.
Arnold, Chester L., and C. James Gibbons. 1996. Impervious surface coverage: The emergence
of a key environmental indicator. Journal of the American Planning Association 62
(2):243-258.
Bates, Laurie J., and Rexford E. Santerre. 2001. The public demand for open space: the case of
Connecticut communities. Journal of Urban Economics 50:97-111.
Berrens, Robert P. 1996. The safe minimum standard approach: an alternative to measuring non-
use values for environmental assets. In Forestry, Economics and the Environment, edited
by W. L. Adamowicz, P. Boxall, M. K. Luckert, W. E. Phillips and W. A. White.
Wallingford, UK: CAB International.
Berven, Keith A., and Thaddeus A. Grudzien. 1990. Dispersal in the wood frog (rana sylvatica):
implications for genetic population structure. Evolution 44 (8):2047-2056.
Bishop, Richard C. 1978. Endangered species and uncertainty: the economics of a safe minimum
standard. American Journal of Agricultural Economics 60:10-18.
Burke, Vincent J., and J. Whitfield Gibbons. 1995. Terrestrial buffer zones and wetland
conservation: a case study of freshwater turtles in a Carolina bay. Conservation Biology 9
(6): 1365-1369.
Calhoun, Aram J. K, and Michael W. Klemens. 2002. Best development practices: Conserving
pool-breeding amphibians in residential and commercial developments in the
northeastern United States. Bronx, NY: Metropolitan Conservation Alliance, Wildlife
Conservation Society.
Carlson, Allan, and Per Edenhamn. 2000. Extinction dynamics and the regional persistence of a
tree frog metapopulation. Proceedings of the Royal Society fo London B 267:1311-1313.
Ciriacy-Wantrup, Siegfried V. 1952. A safe minimum standard as an objective of conservation
policy. In Resource Conservation: Economics and Policy. Berkeley: University of
California Press.
Daniels, Tom. 1999. When City and Country Collide: Managing Growth in the Metropolitan
Fringe. Washington, D.C.: Island Press.
Farmer, Michael C., and Alan Randall. 1998. The rationality of a safe minimum standard. Land
Economics 74 (3):287-302.
Fleishman, Erica, Chris Ray, Per Sjogren-Gulve, Carol L. Boggs, and Dennis D. Murphy. 2002.
Assessing the roles of patch quality, area, and isloation in predicting metapopulation
dynamics. Conservation Biology 16 (3):706-716.
Geoghegan, Jacqueline. 2002. The value of open spaces in residential land use. Land Use Policy
19:91-98.
Gill, Douglas E. 1978. The metapopulation ecology of the red-spotted newt, notophthalmus
viridescens (rafinesque). Ecological Monographs 48 (2): 145-166.
Gilpin, Michael E., and Jared M. Diamond. 1976. Calculation of immigration and extinction
curves from the species-area-distance relation. Proceedings of the National Academy of
Sciences USA 73 (11):4130-4134.
23
-------
. 1981. Immigration and extinction probabilities for individual species: relation to
incidence functions and species colonization curves. Proceedings of the National
Academy of Sciences USA 78 (l):392-396.
Green, David M. 2003. The ecology of extinction: population fluctuation and decline in
amphibians. Biological Conservation 111 (3):331-343.
Gustafson, Eric J., and Robert H. Gardner. 1996. The effect of landscape heterogeneity on the
probability of patch colonization. Ecology 77 (1):94-107.
Hanski, Ilkka. 1994. A practical model of metapopulation dynamics. Journal of Animal Ecology
63:151-162.
. 1999. Metapopulation Ecology, Oxford Series in Ecology and Evolution. New York:
Oxford University Press.
Hanski, Ilkka, and Mats Gyllenberg. 1997. Uniting two general patterns in the distribution of
species. Science 275 (5298):397-400.
Hanski, Ilkka, and Otso Ovaskainen. 2000. The metapopulation capacity of a fragmented
landscape. Nature 404:755-758.
. 2003. Metapopulation theory for fragmented landscapes. Theoretical Population
Biology 64 (1): 119-127.
Hardie, Ian W., Peter J. Parks, Peter Gottlieb, and David N. Wear. 2000. Responsiveness of rural
and urban land uses to land rent determinants in the U.S. South. Land Economics 76
(4):659-673.
Hecnar, Stephen J., and Robert T. M'Closkey. 1996. Regional dynamics and the status of
amphibians. Ecology 77 (7):2091-2097.
Heimlich, Ralph E., and William D. Anderson. 2001. Development at the Urban Fringe and
Beyond: Impacts on Agriculture and Rural Land. Washington, D.C.: Economic Research
Service, U.S. Department of Agriculture.
Irwin, Elena G. 2002. The effects of open space on residential property values. Land Economics
78 (4):465-480.
Johnson, Michael P. 2001. Environmental impacts of urban sprawl: a survey of the literature and
proposed research agenda. Environment and Planning A 33:717-735.
Johnston, Robert J., James J. Opaluch, Thomas A. Gigalunas, and Marisa J. Mazzotta. 2001.
Estimating amenity benefits of coastal farmland. Growth and Change 32:305-325.
Keddy, Paul A., Harold T. Lee, and Irene C. Wisheu. 1993. Choosing indicators of ecosystem
integrity: Wetlands as a model system. In Ecological Integrity and the Management of
Ecosystems, edited by S. Woodley, J. Kay and G. Francis. Delray Beach, FL: St. Lucie
Press.
King, Anthony W. 1993. Considerations of scale and hierarchy. In Ecological Integrity and the
Management of Ecosystems, edited by S. Woodley, J. Kay and G. Francis. Delray Beach,
FL: St. Lucie Press.
Lehtinen, Richard M., Susan M. Galatowitsch, and John R. Tester. 1999. Consequences of
habitat loss and fragmentation for wetland amphibian assemblages. Wetlands 19 (1): 1-12.
Levins, R. 1969. Some demographic and genetic consequences of environmental heterogeneity
for biological control. Bulletin of the Entomological Society of America 15 (l):237-240.
. 1970. Extinction. Lecture Notes in Mathematics 2:75-107.
Lofvenhaft, Katarina, Siv Runborg, and Per Sjogren-Gulve. 2003. Biotope patterns and
amphibian distribution as assessment tools in urban landscape planning. Landscape and
Urban Planning 68 (4):403-427.
24
-------
Lutzenhiser, Margo, and Noelwah R. Netusil. 2001. The effect of open space on a home's sale
price. Contemporary Economic Policy 19 (3):291-298.
Marsh, David M., and Peter C. Trenham. 2001. Metapopulation dynamics and amphibian
conservation. Conservation Biology 15 (l):40-49.
Marshall, Elizabeth, Frances Homans, and Robert Haight. 2000. Exploring strategies for
improving the cost effectiveness of endangered species management: the Kirtland's
Warbler as a case study. Land Economics 76 (3):462-473.
Miltner, Robert J., Dale White, and Chris Yoder. 2004. The biotic integrity of streams in urban
and suburbanizing landscapes. Landscape and Urban Planning 69 (1):87-100.
Moilanen, Atte, and IlkkaHanski. 1998. Metapopulation dynamics: Effects of habitat quality and
landscape structure. Ecology 79 (7):2503-2515.
Moilanen, Atte, and Nieminen. 2002. Simple connectivity measures in spatial ecology. Ecology
84 (4): 1131-1145.
Ovaskainen, Otso. 2002. Long-term persistence of species and the SLOSS problem. Journal of
Theoretical Biology 218 (4):419-43 3.
. 2003. Habitat destruction, habitat restoration and eigenvector-eigenvalue relations.
Mathematical Biosciences 181 (2):165-176.
Ovaskainen, Otso, and IlkkaHanski. 2001. Spatially structured metapopulation models: global
and local assessment of metapopulation capacity. Theoretical Population Biology 60
(4):281-302.
. 2003. How much does an individual habitat fragment contribute to metapopulation
dynamics and persistence? Theoretical Population Biology 64 (4):481-495.
Paton, Peter W. C. 2004. Personal communication on RIGIS vernal pool data.
Pope, Shealagh E., Lenore Fahrig, and H. Gray Merriam. 2000. Landscape complementation and
metapopulation effects on leopard frog populations. Ecology 81 (9):2498-2508.
Randall, Alan, and Michael C. Farmer. 1995. Benefits, costs, and the safe minimum standard of
conservation. In The Handbook of EnviromentalEconomics, edited by D. W. Bromley.
Oxford: Blackwell Publishers.
Roan, Philip F., and Wade E. Martin. 1996. Optimal production and reclamation at a mine site
with an ecosystem constraint. Journal of Environmental Economics and Management
30:186-198.
Rosenberger, Randall S., and John B. Loomis. 1999. The value of ranch open space to tourists:
combining observed and contingent behaviour data. Growth and Change 30:366-383.
Semlitsch, Raymond D. 1998. Biological delineation of terrestrial buffer zones for pond-
breeding salamandars. Conservation Biology 12 (5): 1113-1119.
. 2000. Principles for management of aquatic-breeding amphibians. Journal of Wildlife
Management 64 (3):615-631.
Semlitsch, Raymond D., and J. Russell Bodie. 2003. Biological criteria for buffer zones around
wetlands and riparian habitats for amphibians and reptiles. Conservation Biology 17
(5): 1219-1228.
Sjogren-Gulve, Per. 1994. Distribution and extinction patterns within a northern metapopulation
of the pool frog, rana lessonae. Ecology 75 (5): 1357-1367.
Swallow, Stephen K. 1994. Renewable and nonrenewable resource theory applied to coastal
agricultural, forest, wetland, and fishery linkages. Marine Resource Economics 9 (4):291-
310.
25
-------
. 1996. Economic issues in ecosystem management: an introduction and overview.
Agricultural and Resource Economics Review 25 (2):83-100.
Toman, Michael A. 1998. Why not to calculate the value of the world's ecosystem services and
natural capital. Ecological Economics 25:57-60.
Trombulak, Stephen C., and Christopher A. Frissell. 2000. Review of ecological effects of roads
on terrestrial and aquatic communities. Conservation Biology 14 (1): 18-30.
Vitt, Laurie J., Janalee P. Caldwell, Henry M. Wilbur, and David C. Smith. 1990. Amphibians as
harbingers of decay. Bioscience 40 (6):418.
Vos, Claire C., A. G. Antonisse-deJong, P. W. Goedhart, and M. J. M. Smulders. 2001. Genetic
similarity as a measure for connectivity between fragmented populations of the moor frog
(Rana arvalis). Heredity 86:598-608.
Vos, Claire C., and J. P. Chardon. 1998. Effects of habitat fragmentation and road density on the
distribution pattern of the moor frog Rana arvalis. Journal of Animal Ecology 35:44-56.
Vos, Claire C., Cajo J. F. Ter Braak, and Wim Nieuwenhuizen. 2000. Incidence function
modelling and conservation of the tree frog hyla arborea in the Netherlands. Ecological
Bulletins 48:165-180.
Welsh, Hartwell H., and Sam Droege. 2001. A case for using plethodontid salamanders for
monitoring biodiversity and ecosystem integrity of North American forests. Conservation
Biology 15 (3):558-569.
Welsh, Hartwell H., and LisaM. Ollivier. 1998. Stream amphibians as indicators of ecosystem
stress: A case study from California's redwoods. Ecological Applications 8 (4): 1118-
1132.
Willson, John D., and Michael E. Dorcas. 2003. Effects of habitat disturbance on stream
salamanders: implications for buffer zones and watershed management. Conservation
Biology 17 (3):763-771.
Woodford, James E., and Michael W. Meyer. 2003. Impact of lakeshore development on green
frog abundance. Biological Conservation 110:277-284.
Yang, Wanhong, Madhu Khanna, Richard Farnsworth, and Hayri Onal. 2003. Integrating
economic, environmental and GIS modeling to target cost effective land retirement in
multiple watersheds. Ecological Economics 46:249-267.
26
-------
TABLE 1. Descriptions of Policy Alternatives
Policy Alternative
Description1
Current wetland policy that protects the area of the pool only
Same as CI except developed parcels are two acres
Protect vernal pool plus 100-foot buffer (envelope)
Protect vernal pool plus 100-foot buffer plus 75% of critical
habitat;2 recommended vernal pool guidelines3
Protect vernal pool plus 100-foot buffer plus 50% of critical habitat
Protect vernal pool plus 100-foot buffer plus 25% of critical habitat
Protect vernal pool plus 100-foot buffer plus 25% of critical habitat
plus 25% of intervening landscape matrix
Protect all vernal pools plus 100-foot buffer plus 100% critical
habitat plus 100% of intervening landscape
1Unless otherwise noted, all policies assumed all developed parcels were one acre.
2Critical habitat is defined as an outer buffer 100-750 feet from pool edge.
3Vernal pool guidelines (Calhoun and Klemens 2002).
Current 1 (CI)
Current 2 (C2)
Envelope (E)
Guidelines (G)
New Policy 1 (PI)
New Policy 2 (P2)
New Policy 3 (P3)
No Development (0)
27
-------
Figure 1. Study area: 123-pond network within 10,000 acres of landscape matrix.
123 Vernal Pools — 10,000 Acres
28
-------
Figure 2. Optimization results. The foregone benefits of development (i.e., the opportunity
costs of restricting development) increase as the level of the metapopulation size is
increased.
Foregone Benefits of Developments
29
-------
Figure 3. Comparison of the opportunity costs corresponding to various policy alternatives
versus the optimal solution set.
Foregone Benefits of Development
I I
I I II
§ I
P3 I
X S
-
P1 /
x S
52
I I
i i i
0.88 0.9 0.92 0.94 0.96
Metapopulation Size (S)
0.98
x 10
4
30
-------
PART Two: Bioprocess-based Spatial Modeling of Reserve Network Design: An
Integer Programming Approach
A common, even leading, cause of biodiversity loss is habitat loss and
fragmentation. A growing literature concerns methods by which to identify priority lands
for purchase and preservation as a conservation reserve, particularly by identifying lands
that currently support species and that may be purchased or conserved at low cost (Ando
et al. 1998, Polasky et al. 2000; Possingham et al. 2000). Economists have used an index
of species diversity or species presence to identify a measure of ecological contribution
from preserving a land parcel and have minimized the cost of achieving a target level of
that index. However, many of these analyses have not considered the spatial relationship
among lands conserved, nor the ecological role that this spatial relationship implies for
land parcels preserved. As a result, many of the reserve designs show a high degree of
fragmentation or disconnection between the land parcels targeted for conservation
(Possignham et al. 2000; McDonnell et al. 2002).
Some recent analyses consider spatial relationships (Williams 1998; Williams and
ReVelle 1998; Possingham et al. 2000; Briers 2002; McDonnell et al. 2002; Nalle et al.
2002b; Onal and Briers 2002, 2003). Yet these studies have generally not emphasized
the ecological mechanisms whereby spatial linkages among land parcels affect the level
achieved relative to a measure of ecological quality. These studies have also been limited
by computational techniques, particularly when the ecological index considered is non-
linear. A combination of heuristic methods and optimization methods have been
presented in the literature (Kirkpatrick 1983; Margules et al. 1988; Nicholls and
Margules 1993; Underhill 1994; Church et al. 1996; Pressey et al. 1996, 1997; Pressey
31
-------
2002; Rodrigues and Gaston 2002). Optimization models may not be practical to solve at
a large scale using available solution methods. Heuristic methods may be used to solve
problems at a larger scale, using a measure of an ecological quality index that is logically,
but imperfectly, correlated with the actual index decision makers might target. While
optimization methods may be used to obtain a truly optimal solution over a restricted
problem, heuristic methods may not produce an optimal (e.g., minimum cost) solution but
can address a less restricted scope of the problem (often a larger geographic area).
This second portion of the research (Part Two of this paper) strives to develop a
bioprocess-based method for selecting land for conservation reserves, considering not
only the spatial relationship among land parcels but also the ecological role of land in that
spatial relationship. We are developing an optimization approach that may be solved
with available methods, but which may not be subject to the restrictions expected by
previous practitioners when optimizing a non-linear index of ecological quality. In
particular, we are developing methods that use a linearization that accurately reflects the
ecological index that one would prefer to optimize directly, and this linearization permits
use of existing integer-program optimization routines. The research is motivated by
conservation of amphibian metapopulations in the Wood-Pawcatuck Rivers watershed of
southwestern Rhode Island and southeastern Connecticut, particularly amphibians which
depend on seasonally flooded wetlands (vernal pools) for breeding sites.
Overview of Ecological Concepts of Reserve Design
Part One of this paper provided a review of a spatially explicit metapopulation
models from ecological theory. That theory notes that a population of some species may
depend on subpopulations residing within habitat patches. These subpopulations may
32
-------
occasionally go extinct and the corresponding habitat patch may be recolonized by
dispersal of individuals, from the remaining occupied patches, across a matrix of inter-
patch land. Patches with higher population size may be more likely to contribute
immigrants (or re-colonizers) to other patches, and patches are more likely to receive
immigrants from patches that are closer rather than farther away. A group of
subpopulations that interact in this way to maintain an overall population is called a
metapopulation.
Based on these fundamental relationships, the research proceeds by developing a
model of the probability that the overall population of an amphibian species goes extinct.
This model depends on the probability that a patch is occupied or, if unoccupied, the
probability that the patch is re-colonized. We then develop a simulation model that
minimizes the probability that the species goes extinct from within the lands designated
for a conservation reserve, as constrained by a budget for the cost of purchasing or
preserving land in the reserve. We omit the equations that describe this probability here,
but Table 1 is indicative of the factors that the project includes.
Three versions of this model (cf. Figure 1) are being developed and examined
with respect to the watershed in southwestern Rhode Island (Figure 2). All versions use
integer programming and a linearized version of an index for the probability that the
overall population goes extinct within a conservation reserve network. In Version One,
individual ponds (vernal pools or habitat patches) are chosen for preservation and their
contribution to the index of extinction probability is calculated. By construction, this
contribution reduces the extinction probability index only if a pond is preserved within a
biologically relevant neighborhood of another pond that is also being preserved. This
33
-------
neighborhood is defined as two kilometers, based on literature pertaining to the dispersal
distances witnessed for the study species (wood frogs or spotted salamanders). This first
version assumes matrix land between ponds will remain permeable to dispersal of
migrant individuals in the metapopulation. In Version Two, preservation of a pond only
reduces the probability of population extinction if that pond is connected to at least one
other pond by a preserved corridor of land. In this version, the budget constraint includes
the opportunity cost of purchasing land around the pond sites plus the corridor between
pond sites (habitat patches). Habitat patches are connected to at least one other patch, but
not necessarily to all other patches. The contribution that preserved patches make to
reducing the index of extinction probability is defined to be higher when the preserved
ponds are connected by a preserved corridor, as compared to version one when the
patches are not required to be connected by a preserved corridor.
Version Three of the model is an intermediate case. In Version Three, preserved
patches do reduce the index for probability of metapopulation extinction if patches are
preserved with or without preservation of the intervening corridor. This reduction in the
index requires preservation of at least one other patch within 2 kilometers but without an
intervening corridor, or it requires preservation of one other patch along with the
intervening corridor. In this third version of the model, isolated patches may be
preserved with or without corridors, but preservation of the corridor between two (or
more) patches increases the contribution of these patches to reducing the index of the
probability of metapopulation extinction within the conservation reserve relative to the
contribution of patches that are not interconnected by preserved corridors.
34
-------
In all cases, the index of the probability of metapopulation extinction is adjusted
for a measure of quality of a habitat patch, as represented by an estimate of the
subpopulation size for that patch. This measure of habitat quality or subpopulation size is
equal to the number of egg-masses estimated at the pond site for that species. This egg-
mass count is representative of the number of adult breeding females of the study species
(Crouch and Paton 2000; Paton, unpubl. data).
Overview of Empirical Application
We applied the model to a subset of vernal pool habitat patches within the
watershed of southwestern Rhode Island; the subset consisted of 39 ponds. The three
versions of the model were solved by integer programming using GAMS 20.7. Figure 3
shows the results of the network design models based on a $10,000,000 budget applied to
the land around these 39 ponds. Land values were derived from tax assessor records for
the towns in the watershed and assigned to cells within a 1 ha grid. Skidds (2004)
provides details.
Results in Figure 3 show that the Version One results in conservation of several
clusters of habitat patches dispersed across the landscape. In Version Two, the additional
requirement to preserve connecting corridors with ponds causes a reduction of the
number of habitat patches preserved and yields a smaller number of clusters of preserved
pond sites. Version Three leads to preservation of a few additional habitat patches in
exchange for not connecting all ponds to at least one other pond by a preserved corridor.
Table 2 shows that Version One preserved the most patches (29) while Version Two
preserved the fewest (12, all with at least one preserved corridor) and Version Three
preserved the next fewest (15).
35
-------
Concluding Notes
This preliminary research demonstrates that it may be feasible to explicitly
consider the ecological role of spatial connections (corridors) between preserved habitat
patches using an index of the probability that key species persist within a conservation
reserve network. The various versions of the model rely, to different degrees, on the
assumption that matrix lands intervening between habitat patches remain permeable to
migrants that could recolonize patches following extinction of the subpopulation within a
patch. Ecological research is critical to calibrating this assumption; that is, ecological
research is needed to shed light on how various types and intensities of development will
change the permeability of unpreserved matrix lands to metapopulation persistence. Such
research will aid in developing a better representation of the probability of
metapopulation persistence in an index that can be built on relatively easily available
biological data (egg-mass counts in this case). Calibrating the components of the index
of metapopulation extinction (or persistence) will allow decision-makers (watershed
managers) to better evaluate the likely impact of tradeoffs between preserving habitat
patches with and without preserving intervening corridors. From our preliminary models,
the basic models seem to offer a practical approach by which local or regional
decisionmakers could identify priority lands for conservation and then develop, purchase,
regulatory, or incentive-based policies that encourage land conservation and development
consistent with maintaining an identified conservation reserve.
References and Literature Cited
Ando, Amy, Camm, Jeffrey, Polasky, Stephen, and Andrew Solow, 1998, Species
Distributions, Land Values, and Efficient Conservation, Science 279: 2126-28.
Briers, Robert A., 2002, Incorporating Connectivity into Reserve Selection Procedures,
Biological Conservation 103: 77-83.
36
-------
Crouch, W. B. Ill, and P. W. C. Paton. 2000. Using egg mass counts to monitor wood
frog populations. Wildlife Society Bulletin 28:895-901.
Diamond, J.M., 1975, the Island Dilemma: Lessons of Modern Biogeographic Studies for
the Design of Natural Reserves, Biological Conservation 7: 129-146.
Flint, R. F. 1971. Glacial and Quaternary Geology. John Wiley and Sons, New York,
NY, USA.
Gonzalez, A., Lawton, J.H., Gilbert, F.S., Blackburn, T.M., and I. Evans-Freke, 1998,
Metapopulation Dynamics, Abundance, and Distribution in a Microecosystem,
Science 281: 2045-47.
Gosselink, J.G. et al. 1990, Landscape Conservation in a Forested Wetland Watershed,
BioScience 40: 588-600.
Hanski, I., 1994, A Practical Model of Metapopulation Dynamics, Journal of Animal
Ecology 63: 151-62.
Joly, P., Morand, C., and A. Cohas, 2003, Habitat Fragmentation and Amphibian
Conservation: Building a Tool for Assessing Landscape Matrix Connectivity,
Biologies 326: 132-139.
Kindvall, O., 1999, Dispersal in a Metapopulation of a Bush Cricket, Metrioptera Bicolor
(Orthoptera: Tettigoniidae), .Journal of Animal Ecology 68: 172-85.
Kirkpatrick, J.B., 1983, An Iterative Method for Establishing Priorities for Selection of
Nature Reserves: An Example from Tasmania, Biological Conservation 25: 127-
134.
Mann, C.C. and M.L. Plummer, 1995, Are Wildlife Corridors on the Right Path? Science
270: 1428-30.
Margules, C.R., Nicholls, A.O., and R.L. Pressey, 1988, Selecting Networks to Maximize
Biological Diversity, Biological Conservation 43: 63-76.
Margules, C.R. and R.L. Pressey, 2000, Systematic Conservation Planning, Nature 405:
243-53.
McDonnell, Mark D., Possingham, Hugh P., Ball, Ian R., and Elizabeth A. Cousins,
2002, Mathematical Methods for Spatially Cohesive Reserve Design,
Environmental Modeling and Assessment 7: 107-114.
Moilanen, A. and M. Nieminen, 2002, Simple Connectivity Measures in Spatial Ecology,
Ecology 83(4): 1131-45.
Mwalyosi, R.B., 1991, Ecological Evaluation for Wildlife Corridors and Buffer Zones for
Lake Manyara National Park, Tanzania and Its Immediate Environment,
Biological Conservation 57: 171-186.
Nalle, Darek J. Arthur, Jeffrey L., and John Sessions, 2002, Designing Compact and
Contiguous Reserve Networks with a Hybrid Heuristic Algorithm, Forest Science
48(1): 59-68.
Nicholls, A.O. and C.R. Margules, 1993, An Upgraded Reserve Selection Algorithm,
Biological Conservation 64: 165-169.
Nicholls, A.O., 1998, Integrating Population Abundance, Dynamics and Distribution into
Broad Scale Priority Setting, in: Mace, G.M., Balmford, A. and J.R. Ginsberg
(eds), Conservation in a Changing World: Integrating Processes into Priorities for
Action, Cambridge Univ. Press, 251-272.
Onal, Hayri and Robert A. Briers, 2002, Incorporating Spatial Criteria in Optimum
Reserve Network Selection, Proc. R. Soc. Lond. B 269: 2437-2441.
37
-------
Onal, Hayri and Robert A. Briers, 2003, Selection of a Minimum-boundary Reserve
Network Using Integer Programming, Proc. of Royal Society of London B 270:
1487-1491.
Onal, Hayri and R. A. Briers, 2004, Designing a Conservation Reserve Network with
Minimal Fragmentation: A Linear Integer Programming Approach, working
paper.
Pelletier, J.D., 2000, Model Assessment of the Optimal Design of Nature Reserves for
Maximizing Species Longevity, Journal of Theoretical Biology 202: 25-32.
Polasky, Stephen, Camm, Jeffrey D., and Brian Garber-Yonts, 2001, Selecting Biological
Reserves Cost-Effectively: an Application to Terrestrial Vertebrate Conservation
in Oregon, Land Economics 77(1): 68-78.
Possingham, Hugh, Ball, Ian, and Sandy Andelman, 2000, Mathematical Methods for
Identifying Representative Reserve Networks, in: Quantitative Method for
Conservation Biology, Ferson, S. and Burgman, M.(eds), Springer-Verlag, New
York.
Pressey, R.L. and R.M. Cowling, 2001, Reserve Selection Algorithms and the Real
World, Conservation Biology 15: 275-77.
Pressey, R.L., Possingham, H.P., and C.R. Margules, 1996, Optimality in Reserve
Selection Algorithms: When does It Matter and How Much? Biological
Conservation 76: 259-267.
Pressey, R.L., Possingham, H.P., and R.J. Day, 1997, Effectiveness of Alternative
Heuristic Algorithms for Identifying Indicative Minimum Requirements for
Conservation Reserves, Biological Conservation 80: 207-219.
Rodrigues, Ana S., Cerdeira, J. Orestes, and Kevin J. Gaston, 2000, Flexibility,
Efficiency, and Accountability: Adapting Reserve Selection Algorithms to More
Complex Conservation Problems, Ecography 23: 565-574.
Rodrigues, Ana S. and Kevin J. Gaston, 2002, Optimization in Reserve Selection
Procedures — Why Not? Biological Conservation 107: 123-129.
Sessions, J., 1992, Solving for Habitat Connections as a Steiner Network Problem, Forest
Science 38(1): 203-207.
Shafer, C.L., 1995, Values and Shortcomings of Small Reserves, Bioscience 45: 80-88.
Shaw, M.W., 1994a, Modeling Stochastic Process in Plant Pathology, Annual Review of
Phytopathology 32: 523-44.
Shaw, M.W., 1994b, Simulation of Population Expansion and Spatial Pattern When
Individual Dispersal Distributions Do Not Decline Exponentially with Distance.
Proceedings of the Royal Society of London, Series B 259: 243-48.
Simberloff, D.S. and L.G. Abele, 1982, Refuge Design and Island Biogeographic Theory:
Effects of Fragmentation, American Naturalist 120: 41-50.
Simberloff, D.S. and N. Gotelli, 1984, Effects of Insularization on Plant Species Richness
in the Pairie-Forest Ecotone, Biological Conservation 29: 27-46.
Skidds, D.E., 2004, Vernal Pool Amphibian Habitat Conservation in the Pawcatuck River
Watershed of Southern Rhode Island, unpublished paper.
Soule, M.E., 1991, Conservation: Tactics for a Constant Crisis, Science 253: 744-50.
Soule, M. and D. Simberloff, 1986, What Do Genetics and Ecology Tell Us about the
Design of Nature Reserves? Biological Conservation 35: 19-40.
Underhill, L.G., 1994, Optimal and Suboptimal Reserve Selection Algorithms, Biological
38
-------
Conservation 70: 85-87.
Wiens, J. A., 1996, Wildlife in Pathy Environment: Metapopulations, Mosaics, and
Management, in: D.R. McCullough (eds), Metapopulations and Wildlife
Conservation, pp. 53-84, Island Press, Washington, D.C.
Wiens, J. A., 1997, Metapopulation Dynamics and Lanscape Ecology, in: I. Hanski and
M.E. Gilpin (eds), Metapopulation Biology: Ecology, Genetics, and Evolution,
pp. 43-62, Academic Press, San Diego.
Wilcove, D.S. and R.M. May, 1986, National Park Boundaries and Ecological Realities,
Nature 324: 206-207.
Williams, Justin C. and Charles S. ReVelle, 1998, Reserve Assemblage of Critical Areas:
A Zero-one Programming Approach, European Journal of Operations Research
104: 497-509.
Williams, Justin C., ReVelle, Charles S., and Simon A. Levin, 2004, Using Mathematical
Optimization Models to Design Nature Reserves, Front. Ecol. Environ. 2(2): 98-
105.
With, K.A., 2002, Using Percolation Theory to Assess Landscape Connectivity and
Effects of Habitat Fragmentation, in: K.J. Gutzwiller (eds), Applying Landscape
Ecology in Biological Conservation, pp. 105-130, Springer-Verlag, New York.
Table 1. Definition of Model Variables and Parameters
Model Variables and Parameters
Definition
Decision variables
tij
status of vernal pool i and j, 1 indicating both
selected, 0 otherwise
Zij
status of corridor connection between pond i and j, 1
indicating being selected, 0 otherwise
Xi
status of pond i, 1 indicating being selected, 0 other
wise
Model parameters
N;
set of potential source ponds of pond i, defined by
distance
Pij
probability of migration of amphibian from pond j to
pond i with corridor connection
Pu
probability of migration of amphibian from pond j to
39
-------
pond i without corridor connection
a
dispersal parameter
dij
distance between pond i and j, km
m;
egg mass count in pond i
B
Budget line
Table 2. Comparison of Reserve Networks under Alternative Models
Model 1.
Model 2.
Model 3.
Vernal pools
29
12
15
Corridor connection
0
10
9
Extinction probability
3.11x 10"7
2.13 x 10"19
6.61 x 10"20
40
-------
o o
o o
A. conservation issue
O O
C. corridor connection only D. corridor connection and individual pond
Figure 1. Reserve Network under Alternative Conservation Strategies: Panel A presents the
conservation issue; Panel B demonstrates the reserve system by selecting individual pond; Panel
C demonstrates the reserve system by selecting habitat corridor, where the dot line and gray circle
means the budget is not enough to fully cover their cost; Panel D demonstrates the case with
individual pond and corridor.
41
-------
Figure 2. the Pawcatuck Watershed Area of Rhode Island
42
-------
Optimal Reserve Network
(without corridor connection)
Optimal Reserve Network
(with corridor connection only)
Optimal Reserve Network
(with corridor connection and single pond)
A. B. C.
Figure 3. Comparison of the Reserve Networks under Alternative Models with A Budget of 10 Million Dollars
43
-------
o o
o o
A. conservation issue
O
C. corridor connection only D. corridor connection and individual pond
Figure 1. Reserve Network under Alternative Conservation Strategies: Panel A presents the
conservation issue; Panel B demonstrates the reserve system by selecting individual pond; Panel
C demonstrates the reserve system by selecting habitat corridor, where the dot line and gray circle
means the budget is not enough to fully cover their cost; Panel D demonstrates the case with
individual pond and corridor.
44
-------
-------
Optimal Reserve Network
(without corridor connection)
Optimal Reserve Network
(with corridor connection only)
Optimal Reserve Network
(with corridor connection and single pond)
A. B. C.
Figure 3. Comparison of the Reserve Networks under Alternative Models with A Budget of 10 Million Dollars
46
-------
Spatial Analysis of Private Land Conservation Behavior
Heidi J. Albers* and Amy W. Ando**
* Associate Professor ** Assistant Professor
Dept. of Forest Resources Dept. of Ag. and Consumer Economics
College of Forestry Univ. of Illinois Urbana-Champaign
Oregon State University Urbana, IL 61801
Corvallis, OR 97331
We gratefully acknowledge major funding from the National Science Foundation. This material
is also based in part upon work supported by the Cooperative State Research, Education, and
Extension Service, U.S. Department of Agriculture, under Project No. ILLU 05-0305. We thank
our excellent research assistants, Mike Batz, Xiaoxuan Chen, and Kenny Gillingham, while
taking full responsibility for any mistakes and misstatements that lie within.
47
-------
I. Introduction
Protected lands provide important conservation benefits in the U.S. — public goods ranging
from species protection to water quality maintenance to recreation. Although state and federal
governments manage public lands to provide these benefits, private land trusts augment the
supply of benefits by conserving other land. By 1998, at least 1,211 land trusts were operating in
the U.S. (LTA, 1998), where a land trust is defined as a non-profit private organization actively
engaged in land conservation activity. The acreage of land protected by such groups tripled in the
1990s.
The existing constellation of lands protected by governments may not contain the most valuable
land for providing conservation services (Scott, Abbitt, and Groves, 2001). Gap analysis—a
technique used by conservation biologists to assess reserve systems—for various regions of the
U.S. finds glaring differences between the location of species and threatened ecosystems and the
location of conserved land (Hoctor, Carr, and Zwick, 2000; Wright, Scott, Mann, and Murray,
2001). In addition, the production of many types of conservation benefits, including both
watershed and species protection, contains thresholds of land below which only limited benefits
accrue but the tendency of governments to overdisperse their conservation spending can mean
that those thresholds are not met (Wu and Boggess, 1999). Only limited expansion of federal
protected areas is likely; hence, private land conservation agents have an important role to play
in rationalizing the nation's network of protected lands.
Despite the growing importance of private land trusts in the provision of these valuable public
goods, few economic analyses (see literature section below) have examined either individual
land trust decisions or the cumulative effect of land conservation by the uncoordinated activities
of many public organizations and private trusts. In non-economic analysis, authors describe the
failure of focused local land trusts to provide regional or cross-jurisdictional conservation
benefits (Goldsmith, 2001). This local focus and lack of coordination across trusts has become a
serious enough problem to give rise to organizations that facilitate coordination across trusts
(Maine Land Trust Network, 2002; Goldsmith, 2002; Bay Lands Center, 2002). Albers and
Ando's (2002) economic analysis of the structure of the land conservation industry revealed that
whether land trusts incorporate information about other land conservation activities into
decisions is important for the provision of a socially desirable level of conservation benefits.
However, land-trust decision making processes are complex and poorly understood.
It is difficult to control, predict, and even describe an effective state-wide conservation strategy
when the efforts that comprise that conservation are made independently by distinct
organizations. If, for example, a state government engaged in a major land-conservation
initiative, we do not know how private trusts would react. Economic theory about government
activities might suggest that such government conservation could displace some or all land
conservation by private trusts through "crowding-out." But, because conservation benefits are a
function of cumulative amounts of land and its configuration, perhaps government land
conservation could act as a "seed" and "crowd-in" private land conservation activities. This
project seeks to develop some understanding of how public and private conservation agents
interact and the consequences of different modes of interaction in order to elucidate the potential
impact of various policies to promote private land conservation. Though the discipline of
48
-------
economics is ideally suited for modeling and understanding this kind of inter-agent interaction,
little rigorous scholarly analysis exists within economics to shed light on this important
mechanism for providing valuable public goods.
To fill the gaps in our current understanding of private conservation, we are engaged in a body of
research which examines the actions and decisions of private land trusts with a focus on their
consideration of the land conservation actions of other land conservation actors. In our project,
we will characterize land trust decisions with models and data. We will explore the extent to
which land trusts make decisions based on information about the actions of other land
conservation actors, and examine the impact of information exchange and coordination among
diverse land trusts and between trusts and government. On the policy front, we will identify the
role of government protected lands in encouraging or discouraging private land conservation,
and explore policy options such as providing information and facilitating coordination across
land trusts to encourage socially-preferred patterns of land trust activity.
This paper provides a background in the literature for our work, and sets out a general modeling
framework for the research. We report on preliminary results from simulation and econometric
exercises that are being conducted as parts of this multi-faceted research project. We conclude
with observations regarding the status of the work to date, and expectations for future efforts.
II. Literature and Background
There is a now-classic literature on the impact that the public provision of public goods may
have on related levels of private provision. This literature focuses largely on the potential for and
extent of "crowding out" of private provision by public spending (e.g. Andreoni, 1989; Kingma,
1989). More recent work by Andreoni (1998) does point to the potential for government grants to
"seed in" private contributions when charitable groups have fixed costs. However, papers in this
literature have not considered the case of public goods for which spatial externalities are likely to
be important.
The social benefits of open, protected, and managed lands are public goods with complex spatial
features. While conservation has not been modeled in the public-good provision framework,
there is a sizable literature analyzing land use patterns: studies of tropical deforestation (Chomitz
and Gray, 1996; Nelson and Hellerstein, 1997; Cropper et al, 2001), participation in farmland
preservation programs (Lynch and Lovell, 2003), and land-development patterns (Irwin, 2002;
Irwin and Bockstael, 2002; Bell and Irwin, 2002). These studies provide some insight into
landowner decision making. However, they have focused on decisions made by private agents to
maximize individual utility. Any analysis of land retirement choices aimed at providing public
goods such as biodiversity must be qualitatively different, for those benefits depend differently
on the spatial pattern of land uses.
Swallow et al. (1997) and Swallow, et al. (1990) were early advocates of the notion that the
impact of land use decisions hinge critically on the precise configuration of land uses in the
landscape. Albers (1996) develops a model of tropical land management in which certain uses
create adjacency values and certain spatial patterns of uses interact with the flows of benefits
over time. MacFarlane (1998) points out that because of spatial externalities, agri-environmental
49
-------
policies could yield greater benefits if they were designed to consider the landscapes produced
by such policies rather than focusing on individual farms.
The reserve-site selection literature began to model the optimal choice of protected lands when
the benefits of protecting one parcel depend on exactly which other parcels have been protected
(Ando et al. 1998; ReVelle, Williams, and Boland, 2002). However, these papers provide
normative guidance for a single hypothetical social planner. In reality, conservation networks
emerge from land-use choices made by many agents.
Only recently have scholars begun to develop frameworks for analyzing conservation outcomes
and policy when multiple decision makers exert spatially-explicit externalities on each other.
Bergeron and Polasky (2000) show that when land-use decisions made by multiple landowners
contribute to a species' survival or downfall, conservation efforts may be inadequate or excessive
relative to the social optimum. Experimental work by Parkhurst et al. (2002) suggests that when
benefits depend on the configuration of the entire network of conserved lands, the total benefits
yielded by voluntary conservation networks may be increased by offering agglomeration bonuses
to private landowners. Several new papers allied with urban and regional economics have
modeled equilibrium urban development in settings with multiple agents and spatial externalities
(Marshall, 2004; Tajibaeva, Haight, and Polasky, 2003; Turner, forthcoming). The agents
modeled in these papers, however, do not have public-good provision as an objective.
Very few papers have studied the behavior of the collection of conservation agents which
actually exists in the U.S. Albers and Ando (2003) showed potentially desirable patterns in land-
trust proliferation. There are more trusts in states where net benefits of conservation are higher,
and fewer where the need for coordination might be high relative to the benefits of niche
diversification. The results of that paper also left behind a puzzle: at the state level, there are
more trusts where there is more public conservation. This paper seeks to explore the relationship
between public and private conservation at a finer level of spatial detail.
Work in progress by Parker and Thurman (2004) conducts a panel-data analysis of county-level
conservation acreage, and find evidence that increased government conservation in a county
tends to crowd out private conservation in that same county. Their work has the advantage of
panel data, and considers land enrolled in CRP and WRP as well as public lands which are in
permanent protected status. However, they do not have spatial data on the location of the private
protected lands, which complicates their econometric efforts.
To address the gaps in the literature, we model behavior by multiple agents who aim for public
good provision in a setting where private and total benefits depend on the spatial configuration of
all parcels protected in the landscape. This pursuit yields a framework for simulations in which
government provision may crowd private conservation either in or out, and the resulting pattern
may or may not demonstrate spatial agglomeration. Empirically, we use township-level spatial
data on the location of private and public lands in several states to explore the spatial
relationships that exist among protected lands, and to cast light on which of the scenarios we
model seems most likely to obtain in the real world.
50
-------
III. Framework
The fundamental perspective of the modeling framework employed in both the simulation and
empirical work discussed here is that land trusts make decisions about the location and amount
of land to conserve by solving a constrained optimization problem to maximize their net benefits
subject to budget constraints. At least two characteristics of land conservation differentiate this
problem from other such optimization problems. First, the benefits from conservation are a
function of the total amount of land conserved, rather than solely a function of the amount of
land the individual trust conserves. Second, the benefits from conservation are a function of the
spatial configuration of land conserved, rather than solely a function of the total land area
conserved. Both of these characteristics imply that to maximize net conservation benefits, the
trust must consider the conservation actions of other trusts and the government in choosing how
much and where to conserve.
A. Benefits
Because the benefits from conserved land are public goods, each conservation actor receives
benefits from all conserved land, not just from land they conserve themselves. The shape of the
benefit function from all conserved land can take a variety of forms. In a standard example, the
marginal value of an acre of one type of land declines as the amount of that land type that is
conserved increases. In some cases such as watershed and habitat protection, however, the
conservation benefit function contains thresholds, which implies ranges of increasing marginal
benefits to land conservation in the same area (Wu and Boggess, 1999). Furthermore, in a
wildlife example, the marginal value of a small piece of forest may be very high if that plot
provides a corridor for wildlife to travel between two larger conserved areas but that same piece
of forest may provide few benefits if the other areas are not conserved. Because conserved
parcels create "spillover" benefits when appropriately paired, the production of conservation
benefits is a function of bundles of land parcels rather than of the sum of benefits from individual
parcels (Albers 1996; Swallow, et al. 1990). In this framework, as in Albers and Ando (2003), a
land trust decides whether to purchase a particular plot based on the costs and benefits of that
parcel but the benefits generated by that parcel are a function of the amount and pattern of land
conservation overall.
B. Actors
In the framework developed here, conservation actors (government or land trusts) may differ
from each other in how they value conserved land. For example, some actors may place a high
value on creating contiguous conservation land for species protection while other actors more
highly value conservation land that creates local open space. Even though each actor may have
their own benefit function to determine the level of benefits they receive from a particular
pattern/amount of conserved land, however, each actor receives benefits from all protected land
rather than simply from the land that they conserve. The individual actor incurs the costs of the
conservation activities that it undertakes but receives benefits, according to their specific benefit
function, from those activities and all other conservation activities.
51
-------
Because of this structure to conservation benefits, and based on observations of land trust
activities, the modeling framework employed here assumes that individual land trusts do not
make their conservation decisions in a vacuum. Trusts consider what conservation activities the
government and other trusts have taken or are planning to take when they make decisions about
their own conservation activities. The individual actor's decision, then, is to maximize the net
benefits of conservation when conservation benefits are generated by their actions and the
actions of others, incurring the costs of their own conservation, and subject to a budget
constraint.
Because many actors are operating at the same time in the same general location, we posit a
game structure to the interaction of actors and their decisions. In this framework, we can
consider cases in which one actor (perhaps the government) undertakes actions first and the other
actors make decisions based on the first actor's decisions such as in a sequential move
Stackelberg equilibrium game. Similarly, all actors may make decisions at the same time with
some information about what the other actors are doing, such as in a simultaneous move game
resulting in a Nash equilibrium.
Our framework, then, is quite general and parsimonious. The shape of the benefit functions for
each conservation actor and the size of their budget constraint will contribute to the resulting
pattern of land conservation and the total conservation benefits provided.
IV. Simulation Analysis
A. Structure
The first set of simulations explore the patterns and amount of land conserved by two
conservation actors who choose discrete land parcels to conserve from a line of 7 parcels. The
two actors face the same cost function for conservation but have potentially different benefit
functions and budget constraints. The cost function is simply:
where xt is the number of parcels that that individual actor i conserves and X j is the set of
conserved plots for actor i.
For our benchmark case and for most of the analysis here, the parcels are identical in their
potential contribution to conservation benefits. In our preliminary analysis, the benefits
functions have two components. First, the trust values benefits from total land conserved
according to the simple function:
When a; is less than one, the actor i receives diminishing marginal benefits from total area of
conserved land. When a; is greater than one, the actor receives increasing marginal benefits from
total area of conserved land. The latter case may prove particularly relevant when there are
C(X1) = Px1
(1)
Bl(Xl+X_i) = (Xl+X_if-
(2)
52
-------
thresholds within the benefits function, such as occur in watershed protection and in avian
habitat protection.
Second, as a first step toward capturing benefits that derive from the spatial configuration of
plots, each trust may have a positive, negative, or zero value for adjacencies between conserved
plots. In this preliminary analysis, the pattern of conservation generates an adjacency value for
each border with conservation land on both sides. (In future work, we will include more
sophisticated configuration values such as those that differentiate between two groups of two
plots and one plot removed from a group of three plots.) The total benefits to trust i from a
pattern of conservation include both the benefits from total area conserved and the additional
adjacency value,for each of j relevant borders:
H(X ¦ X ) (X X) yj (3)
For each of a range of sets of parameters that describe the benefits and costs, we explore and
compare several types of interactions between actors: completely cooperative (social optimum);
full-information, noncooperative, sequential move and simultaneous move games; and no-
information, noncooperative, sequential move and simultaneous move outcomes. By comparing
the pattern and values that arise from the social planner's decisions to those from various sets of
independent agents, the simulations will reveal the settings in which a "free market" of
conservation agents may or may not lead to a pattern of conservation with a high level of total
social benefit. In future work, we will run simulations to mimic potential policies to increase the
provision of conservation benefits. One such policy is the use of public conservation land to
"seed" land trust activity in an area. Another policy to consider is the role of the government in
providing information and assistance in the coordination of land trust activities. The simulation
analysis will reveal the settings in which these policies are likely to encourage socially beneficial
private land trust activity in addition to settings in which the conservation by private trusts must
be augmented directly by public land conservation to achieve the socially-preferred outcome.
A MatLab program solves this game/decision model over a line of parcels for a range of
conservation actor types, interaction types, and costs/benefit parameters. The results discussed
below use two conservation actors (hereafter land trust 1, LT1, and land trust 2, LT2) and 7
parcels in a line. The model computes the equilibria for a sequential game in which LT1 is the
Stackelberg leader-trust and LT2 is the follower-trust, and then uses the same code, switching
trust parameters, to compute the equilibria for a sequential game in which LT2 leads. The Nash
equilibria for a simultaneous move game are computed from the two sets of best responses of the
follower trusts from each sequential game. The social planner problem is solved by giving one
trust all of the funding and the social benefit function and picking the highest net valued pattern
using the same code but having no reaction from a second trust.
The model first creates a set of all possible patterns of parcel conservation for both trusts. For
each pattern, for each trust, costs, benefits, and net costs are computed; if costs to either trust
exceed budget constraints, the pattern is removed from consideration as a possible equilibrium.
For each possible pattern of LT1 conservation, LT2 determines its best response pattern by
maximizing net benefits; LT2 may have multiple choices of patterns that provide the same
maximum level of benefits. LT1 then maximizes benefits over all of its possible choices and for
53
-------
each of LT2's best responses to these choices; each pattern that maximizes benefits for LT1 is an
equilibrium. The model then repeats this process, with trusts switching lead, to determine
equilibria for LT2 as leader. To compute Nash equilibria for a simultaneous move game, the
model uses LT2's best response patterns from the LT1 leader case, and LTl's best response
patterns from the LT2 leader case. If a particular pattern of LT1 and LT2 moves satisfies both
best response patterns for the individual trusts, it is a spatial Nash equilibrium to the
simultaneous move game. For all cases, the program records the resulting pattern of conservation
and the level of benefits and costs each actor incurs. In most cases we consider the government
to be LT1, with their benefit function representing that of society.
B. Preliminary Simulation Results
1. Crowding In and Out in Total Area Conserved
As a first case, when neither trust has any value for spatial adjacency, the shape of the benefit
functions from total area conserved determines whether government conservation crowds in or
crowds out private conservation (see Table 2). To determine whether there is crowding in or out
in total land conserved, we begin with the equilibrium from a game with one set of parameter
values and then compare it to the equilibrium from a game in which the government has a larger
budget and conserves more (for the same parameter values).
As expected, when both conservation actors have a benefit function that exhibits diminishing
marginal benefits, a1, demonstrates the opposite situation. In this case, at the lower
budget constraint for LT1, LT1 conserves 0 parcels and LT2 conserves 0 parcels in the
simultaneous move and sequential move Nash equilibria. The social planner, however,
conserves 2 parcels in this case. The uncoordinated actions of the two trusts create an outcome
that isn't close to the socially preferred level of conservation. In this case, the marginal benefits
of conservation do not outweigh the costs for LT2 and LT1 has too small a budget to conserve.
When LTl's budget constraint is loosened, for the simultaneous and sequential move games,
LT1 conserves 1 parcel. This extra conservation from LT1 raises the marginal benefits of
conservation enough that now LT2 spends its entire budget on conservation. This case results in
54
-------
42 possible Nash equilibria with 2.2974 net benefits generated. If LT1, the government,
provides enough conservation, it can induce crowding in of private conservation in the case of
increasing marginal benefits to conservation and the two actors together create the socially
optimal (equivalent to the social planner's) level of conservation.
In both sets of cases, no one values the spatial configuration of conservation and the result is a
large number of equilibria that all generate the same net benefits. Some of the equilibria have
scattered parcels while others have groups of contiguous plots but the actors do not distinguish
between these patterns.
2. Spatial Agglomeration
To understand the impact of benefits from spatial configuration, we begin by examining the
patterns and benefits that two conservation actors produce when at least one of them has a
positive value for each adjacency between conserved parcels.
a. Diminishing Marginal Benefits (Tables 3-5).
In the cases without spatial adjacency values described above, increased conservation activity by
one actor crowds out activity by the other actor. Similarly, in a scenario in which LT1 has a
positive adjacency value but LT2 has a zero adjacency value, increases in LTl's budget still
imply that LT2 conserves fewer parcels through crowding out (Table 3). In this case, however,
as LT1 takes on more of the conservation, the pattern of conservation becomes increasingly
contiguous; LT1 values adjacency and generates patterns of conservation with more
agglomeration. In the simultaneous move game, in the low budget case, LT1 conserves 1 of the
total 4 conserved parcels and can only guarantee one adjacency per equilibrium. At higher
budgets, LT1 conserves more parcels while LT2 conserves fewer parcels but LT1 increasingly
controls the number of adjacencies. In all simultaneous move games, the equilibrium never has
an LT1 parcel that is not adjacent to another conserved parcel. In the sequential move game with
LT1 as the leader, however, LT1 "moves" and LT2's response only leads to adjacencies some
fraction of the time. LT1 sees lower benefits on average because LT2 can generate some low
valued equilibria without adjacencies. In this parameterization, LT1 gets a particularly bad
equilibrium in which no adjacencies are created 2 percent of the time and receives a particularly
good equilibrium with 3 adjacencies in 14 percent of the possible equilibria. In contrast, when
LT2 is the leader of a sequential move game, LT1 can always guarantee at least one adjacency
and so the worst equilibria are not as low valued as in the LT1 leader case and the results are the
same as those in the simultaneous move case. Because LT1 is the only actor who values
adjacency, when its budget increases or it has a chance to react to the other actor's actions, LT1
uses that additional control to generate spatial adjacencies.
If LT1 represents the government in this case, then losses associated with the crowding out of
private conservation (by LT2) are partially offset by creating more highly valued patterns of
conservation. In this scenario, the government is better off when it is not the leader and either
follows or participates in a simultaneous move game. Still, the inability to coordinate across
actors is costly. A social planner conserves more plots as the budget increases and generates
more adjacencies and thus higher benefits.
55
-------
If both actors have a positive value for adjacency, the outcome of any game (simultaneous or
sequential with either actor leading) creates as many adjacencies as possible (Table 4). In fact, if
the adjacency value is large enough relative to the level of marginal net benefits, that value can
slow or stop crowding out of one actor's conservation when the other increases their level of
conservation. The adjacency value is potentially large enough to offset the diminishing marginal
benefits, over some range. This scenario suggests that considering the relationship between
marginal benefits and a spatial adjacency value would be useful in determining the correct size
for an agglomeration bonus (Parkhurst, et al., 2002).
If LT1 has a positive adjacency value and LT2 has a negative adjacency value, the two actors'
aims are at odds and pure strategy equilibria can be difficult to find (Table 5). For our baseline
parameters, no simultaneous move pure strategy Nash equilibria exist. When LT1 is the
Stackelberg leader, the situation is even more extreme than in the case of LT2 having no value
for adjacency. In this case, LT2 will only conserve parcels that are located away from other
parcels. At low budgets for LT1, 4 parcels are conserved in equilibrium with LT1 conserving
only 1 of those and the equilibrium patterns of conservation contain no adjacencies. When LT2
is the leader and LT1 has a low budget, LT1 conserves 1 parcel and always places it adjacent to
one of LT2's 3 parcels to generate equilibria that have one adjacency. However, when LTl's
budget increases to the point that it can conserve 2 parcels, LT2 conserves no parcels and the
equilibria contain only LTl's 2 adjacent parcels. Although the other results here suggest that
LTl's increased conservation would crowd out one parcel of LT2's conservation, in this
parameterization, the negative value of adjacency for LT2 is significant enough to reduce total
conservation by crowding out in total area and is an example of "spatial crowding out."
b. Increasing Marginal Benefits (Tables 6-8)
The case of increasing marginal benefits and no spatial adjacency values, described above,
demonstrated that for some parameter values, one actors' conservation can create a seed effect
that induces more conservation by the other actor. Adding a positive value to adjacency can
provide extra incentive for that crowding-in in total area, in addition to creating an incentive for
spatial agglomeration.
For the case of LT1 with a positive value for adjacency and LT2 with a zero value, the adjacency
value creates an incentive for LT1 to conserve more parcels, located next to each other, than
without that value (Table 6). This additional conservation by LT1 implies that there are more
cases in which LT2's conservation will be crowded in when LT1 values adjacency. Again, the
two actors generate the highest benefits when LT2 is the Stackelberg leader or in the
simultaneous move game because LT1 can locate its parcels to generate adjacencies in
equilibrium. For example, the social planner or coordinated activities of the two trusts generates
3 adjacencies while the simultaneous move and LT2 leader games generate 2.4 adjacencies on
average (forming 3 adjacencies in 42.1% of equilibria) while LT1 as the leader forms 2.1
adjacencies on average (3 adjacencies in 25 % of equilibria and 1 adjacency in 15% of
equilibria). Both the average benefits over equilibria and the "worst case scenario" equilibrium
are higher when LT1 can respond rather than lead in picking parcels.
56
-------
The situation in which both actors have a positive value for adjacency leads to the same pattern
of all parcels located adjacent to each other in the sequential move games and with a social
planner (Table 7). In our parameterization, all game structures lead to 4 conserved parcels but
the simultaneous move game has lower valued equilibria that contain gaps between 2 pairs of
adjacent parcels in 25 % of the equilibria. Benefits are higher for both trusts and for society
when the trusts' actions are coordinated or sequential. Compared to a case of the same values
with both actors having a zero value for adjacency, more conservation takes place, because it is
more highly valued with the adjacency value, and the patterns in equilibrium contain more
contiguous conserved parcels.
If LT1 has a positive value for adjacency but LT2 has a negative value for adjacency, again, pure
strategy equilibria to simultaneous move games are difficult to find (Table 8). The sequential
move games give an advantage to the follower because that actor can exercise control over
whether to generate adjacencies or not. LT2's negative value for adjacency can offset the
tendency of LTl's conservation to crowd-in LT2 activity, especially when much of the line of
parcels is conserved. Whether crowding-in in total area conserved occurs reflects the relative
value of the marginal benefits of parcels versus the adjacency "cost" and whether there are
available parcels on the line that allow for no adjacencies to be created when LT2's conservation
is crowded in.
3. Special Cases/Future Work
Many conservation organizations argue for conservation to follow priorities beginning with the
parcels that generate the highest benefits or "hotspots." Economists counter that groups of
parcels that generate the highest net benefits are preferred to a pure hotspot analysis and reserve
site selection researchers argue that the best sequence of purchases is not necessarily "hotspots
first." To think about how hotspots might affect the outcome of the uncoordinated activities of
two conservation actors, we increase the benefits from one plot (while maintaining the cost equal
to that of the other parcels) and examine the equilibrium patterns of conservation that result.
One example looks at the case of increasing marginal benefits in which the marginal benefits of
conservation do not exceed the costs for LT2 on the "normal" plots at low levels of conservation
and in which both actors have a small positive value on adjacency (Table 9). If LT1 is the
Stackelberg leader, the equilibria involve LT1 conserving a parcel adjacent to the hotspot and
leaving LT2 to conserve the hotspot itself. (At higher budgets for LT1, LT1 conserves other
parcels adjacent to each other but always leaves LT2 to conserve the hotspot.) Within the range
of values in this parameterization, LT2's costs are not outweighed by any parcel's benefits other
than those of the hotspot and LT2 does not conserve at all if the hotspot is unavailable. When
LT2 is the Stackelberg leader, LT2 leaves all the conservation to LT1, who then chooses to
conserve the hotspot (and, depending on the budget, perhaps other adjacent parcels). The
simultaneous move game results in a low valued equilibrium with only the one hotspot parcel
conserved by LT1 and no conservation by LT2 in one-third of the equilibria and a higher valued
situation in the remaining equilibria in which LT2 conserves the hotspot and LT1 conserves an
adjacent plot. Even in the case of LT2 having a negative adjacency value, the size of the hotspot
benefits can be large enough to induce LT2 to conserve the hotspot (in half of the simultaneous
move equilibria and in the LT1 as Stackelberg leader equilibria) regardless of LTl's decision to
create adjacencies with that plot.
57
-------
Expanding from this hotspot work, we plan to use this framework to examine a range of issues in
land conservation: filling spatial gaps; creating wildlife corridors; heterogeneous land; forming
decisions based on a geographic focus; and other niche conservation goals. In addition, we will
further investigate the role of information in decisions and the types of interactions amongst a
diverse set of land conservation actors. We will examine policy scenarios to determine how the
setting (both benefits/ecology and the conservation "industry" structure) contributes to policy
decisions about spatially strategic land conservation, agglomeration bonuses, and the potential
for institutions to coordinate the activities of conservation actors.
C. Discussion
The modeling and simulation work thus far has focused on developing a framework for
examining the interaction of conservation actors. As expected, the framework demonstrates that
both crowding in and crowding out of private conservation by conservation by government or
other land trusts are possible, depending on the structure of the underlying benefits function.
The crowding in case may be of particular interest in land conservation because of the many
examples in which conservation benefits exhibit increasing marginal benefits over a relevant
range.
Incorporating a simple spatial value, a bonus for adjacency between conserved plots, identifies
two ways in which spatial benefits contribute to conservation patterns. First, as long as neither
actor has a negative value for adjacency, this value leads to land conservation patterns with more
spatial agglomeration than other cases. How much spatial agglomeration we see depends on the
structure of the interaction between the trusts and whether they both value adjacency. In our
cases thus far, all parcels are equally valued and so a tiny positive spatial adjacency value creates
agglomeration. Second, the adjacency value makes conservation more valuable and can offset
crowding out and create crowding in at lower parcel-benefit levels. When one trust's actions
create opportunities for the other trust to capture spatial adjacency values, we not only see
patterns of agglomeration but we see crowding-in in total area conserved. This type of crowding
in - derived from a spatial value rather than the shape of the benefit function over total area - is
somewhat unique to land conservation and potentially useful in establishing policy.
V. Empirical Analysis
For our empirical analysis of private conservation, we will use data on the amount and patterns
of land conservation in two1 states that differ markedly in their ecological characteristics, land-
use patterns, amount of land trust activity, and amount of government protected land.
We explore these data sets to find evidence of the manner in which private conservation
decisions are affected by public land conservation. Do private conservation agents respond to the
amount of previously-protected government land in their decision-making process? Does
government conservation stimulate or stifle nearby private conservation? Is there any evidence
that private conservation is clustered (perhaps to provide benefits more cost-effectively), or are
private lands scattered across the landscape?
1 The research will soon include an analysis of California as well, bringing the number of states analyzed up to three.
58
-------
Our analyses permit a comparison across two different states of the interactions of public and
private land conservation decisions. Based on the models developed in Albers and Ando (2002)
and in this paper, we might expect to see different types and degrees of interaction between
private and public land conservation agents because of differences in the conservation benefits
provided by land in each state.
A. Model
Our empirical analysis is grounded in the conceptual framework set forth in section III of this
paper. Because the vast majority of permanently-protected government land was in conservation
status before the private conservation movement gained strength, we take government protected
areas as exogenous and fixed, and treat private conservation as the dependent variable of our
analysis. We examine the extent to which private conservation in a township is positively or
negatively correlated with the amount of government protected areas in that township and in the
areas that surround it. We also investigate whether the current network of private protected areas
is agglomerated in space.
There are multiple private agents making conservation choices in the landscape. Hence, we are
interested in the extent to which the number of acres of privately protected land in a given
township is correlated, positively or negatively, with the acres of privately protected land in the
townships that surround it. If such mutual relationships exist, we can model private conservation
acreage as a spatial reaction function (as in Bruekner (1998)):
11 =«£>/,+*,/? + «, (4)
j*i
where Y, is private conservation acreage in township and the wf/ are weights that aggregate
conservation acreages in townships other than i into a single "neighboring protected acreage"
variable which has a scalar coefficient (p. That coefficient is the slope of a township's reaction
function. The vector^ contains other important characteristics of the township that influence the
private conservation agents' choice of acreage, the vector p contains coefficients on those
variables, and 8; is the error term (assumed to be normally distributed, homoscedastic, and
independent across observations.)
This equation can be written in matrix form as
Y -
-------
A through its impact on the amount of private conservation in township B itself. In some of our
specifications, we include a variable which is "acres of government conservation in neighboring
townships."2 This permits the possibility that government conservation influences private
conservation in neighboring townships, even if private conservation in neighboring towns is
otherwise mutually uncorrelated.
B. Data and Methods
We divide each state into townships3 and calculate the amount of land protected in each
township by government and private conservation agents. Almost all townships in Massachusetts
have at least some protected land of both types, so we are able to use the standard linear spatial-
econometric techniques. Many townships in Illinois have no privately protected lands. We will
eventually use a spatial tobit to deal with the left-truncation of this dependent variable; since that
model is not readily available in pre-programmed form, we use the standard model in this
preliminary report. We estimate a spatial lag model on these data, and use the results to test two
null hypotheses: amounts of private and public land within a township are not statistically
correlated, and private conservation in a given township is not affected by conservation levels in
neighboring townships.
The spatial unit of analysis is the township. There are 351 townships in Massachusetts, and 1433
townships in Illinois. Because our data source for Illinois township boundaries had some of the
townships subdivided, we have 1691 spatial units in Illinois. "Neighbors" are formally defined as
first-order queen (in other words, all townships that touch the boundary of a given township in
any direction). Hence, the weight matrix Whas diagonal elements equal to zero, and off-diagonal
elements Wy equal to 1 if township j borders township and 0 otherwise.
Our primary sources for Illinois data are the Illinois Department of Natural Resources (IDNR)
and the Illinois State Geological Survey (ISGS). The IDNR data identifies private land registered
in protection with the Illinois Nature Preserves Commission. The ISGS data shows protected
areas owned by local, state and federal government. The two data layers are combined to create a
map of protected land in the state of Illinois. Many of the data layers for the independent
variables come from these two agencies as well.
For the state of Massachusetts, we use data developed and provided by the Massachusetts
Geographic Information System (MassGIS). The "Protected and Recreational Open Space" data
layer shows boundaries of conservation lands and outdoor recreational facilities. Relevant
information about each parcel includes ownership, level of protection, public accessibility,
assessor's map and lot numbers, and related legal interests held on the land, including
conservation restrictions. We use information about the polygons to categorize each area as
either privately or publicly protected. The MassGIS website was our source for many of the
independent variables as well.
2 This created by pre-multiplying the vector of "acres publicly preserved" by the spatial weights matrix, II'.
3 When we add California to our research, we will use 7.5 minute USGS quadrangles of similar size to the townships
in IL and MA since CA does not have townships. There are 3049 such quads in the state.
60
-------
Private protected acreage is illustrated in Figure 1 and the independent variables used in the
analyses are summarized in Tables 10 and 11. The independent variables (elements of matrix X)
include socioeconomic variables (income, education, population) to proxy for the demand for
private conservation acreage. We include characteristics of the land or township that influence
the marginal product of that land in providing benefits. The number of endangered species is
correlated with biodiversity potential. Proximity to the nearest major urban center acts as a proxy
for recreational demand. The area of surface water is correlated with the value of protected land
in providing water quality, erosion control, and aesthetic benefits. The cost of land is correlated
both with the cost of setting it in conservation status and with the probability that the land will be
developed if it is not protected. Population density is also correlated with the degree of
conversion threat faced by open lands. The coefficient of variation (standard deviation divided
by mean) of elevation is included as a proxy for recreation potential and aesthetic value. We
include variables that capture the amount of land that falls into each of a number of land-cover
categories: farmland, forest, agriculture, wetland, "open urban," and developed. Some land-cover
types might be more valued as conservation targets. Large amounts of developed land may act
both to limit the acreage of private conservation possible and to stimulate the intensity of
demand for protecting the open space that remains.
MassGIS maintains data showing which areas are characterized as being "priority habitat" for
threatened species; we use acreage of such habitat as a proxy for the value of land in providing
biodiversity benefits. We have just obtained data on the location of endangered species in
Illinois, and will be using that data to proxy for the same thing in future drafts of this work.
The current regressions do not include any variables which can control for local variation in
environmental ideology. In later drafts, we will be using precinct-level voting data to control for
that source of variation in demand for private land conservation.
C. Results
Tables 12 and 13 present preliminary results for the analyses of private conservation acreage by
township in Illinois and Massachusetts. The model fit is better in Massachusetts than in Illinois;
the fit may be improved for Illinois once we are able to employ spatial tobit methodology.4
In both states, there is a positive spatial lag in private protected areas; more acres are protected
by private agents in a township if the surrounding townships have extensive private reserves.
This finding is consistent with a story in which the equilibrium of private conservation choices
leads to spatial agglomeration in the network of private protected areas; this occurs most in our
simulations when there increasing marginal benefits to conservation, and positive values
associated with adjacency. The coefficients are similar in magnitude in the two states'
regressions, even though the states and the groups of conservation actors that work in them are
very different.
4 Non-spatial analyses were performed on the Illinois data. We ran both OLS and tobit regressions, and found that
the results were qualitatively similar. There is hope then that the spatial tobit results will not vary much from the
results reported here.
61
-------
This finding could be an artifact of spatial correlation in unobserved variables which drive
private conservation choices. However, we have controlled for many socio-economic and
physical characteristics of the land across space. We will continue to add explanatory variables
and explore the functional specification of the analyses to further reduce concerns about omitted
variables and misspecification leading to spurious findings of spatial correlation.
While private conservation may crowd other private conservation into surrounding townships,
we find that the coefficient on government protected land is negative in both regressions; there is
less private protected land in townships which have a large acreage of public reserves. The effect
is much larger in Massachusetts than in Illinois; however, in each state the coefficient is
statistically significantly smaller (in absolute magnitude) than 1. We find (in regressions not
reported here) that a variable capturing "neighbors' public protected areas" has no significant
correlation with private conservation in a township. The nature of the spatial lag model is such
that neighboring public protected areas have an indirect negative effect on private conservation
in a township by reducing private conservation in the neighboring townships. However, the
absence of a direct effect means that the impact of public protected areas on private conservation
is greatest in the township where the public reserve is located.
At first blush, then, it appears that government conservation crowds out private conservation.
However, we note that publicly owned land is, by definition, not available as a target for
acquisition by private land conservation groups. Hence, it may be that the small size of the
negative coefficients, particularly in Illinois, is actually evidence of some kind of a spatial
seeding effect. We will refine the specification of these regressions in order to clarify the impact
of public lands on private conservation.
Other variables are significant in these regressions as well. There is more private conservation in
parts of Illinois with greater variation in elevation, more forest, and more wetlands. This is
consistent with the fact that such areas are likely to yield relatively high conservation benefits in
terms of species conservation and recreational enjoyment.
Population density is negatively correlated with private conservation in Illinois, but there is more
conservation in areas where land values are high. These seem like contradictory findings,
because both variables act as proxies for the opportunity cost of the land and the degree of
conversion threat to unprotected land. We will enrich the specifications of these variables
(adding population growth to the regression, exploring nonlinear functional forms, including
interaction terms with proximity to urban areas) to improve the coherence of the story revealed
by the results.
In Massachusetts, there is more private conservation in areas with large amounts of agriculture,
forest, and priority habitat for threatened wildlife. It is interesting that agriculture was not
significantly correlated with conservation efforts in Illinois. This contrast in the results for the
two states may reflect the fact that Massachusetts is a more heavily developed state than Illinois,
and agriculture provides aesthetic landscape benefits that may be more highly valued on the
margin in the Northeast than in the Midwest. There is also more private conservation in
Massachusetts townships that have relatively large acreage of open urban land; such areas have
high marginal recreational value as protected urban open space.
62
-------
VI. Conclusion
This research is in preliminary stages, but already begins to make contributions to our
understanding of private land conservation behavior. The econometric work will be refined, but
it seems likely that we will continue to find that there is spatial agglomeration in private
conservation activity, and there are hints that we may find that government conservation has
some kind of seeding effect on the choices made by private agents. These findings are consistent
with the simulation results of scenarios in which there are increasing returns to scale and positive
benefits associated with having connected reserves. These are precisely the scenarios in which
there is the greatest potential for government policy to increase total social welfare in the
conservation arena.
Even just using the results of these simple linear simulation models, a number of important
points can be made. First, if the incentives or benefit functions for the land conservation actors
do not closely align, the conservation outcome of their uncoordinated activities creates lower
total benefits, less conservation, and less agglomeration than is socially desirable. Second, a
government conservation actor who ignores the potential activities of private conservation actors
makes decisions that lead to less conservation and less beneficial patterns of conservation than a
government who considers the actions of private actors in making its conservation choices.
Third, in land conservation when spatial adjacencies matter differently to two conservation
actors, the actor who can wait and react to the other actor's decisions has an advantage in
creating its preferred pattern of conservation. Fourth, spatial agglomeration bonus policies can
have a large impact on both the pattern and amount of conservation. Fifth, in some cases, a
government decision to conserve parcels other than a "hotspot" can induce private conservation
of that parcel and lead to higher levels of social benefits.
These findings, especially the last two, have important policy implications. As we continue our
research, we will focus on extensions that provide even more direct guidance to the government
and private officials who are responsible for choosing which lands to add to their portfolios of
protected areas, and to the decision makers in a position to set forth policies to influence private
conservation activities.
63
-------
References
Albers, H. J. 1996. "Modeling Ecological Constraints on Tropical Forest Management: Spatial
Interdependence, Irreversibility and Uncertainty." Journal of Environmental Economics
and Management 30(1): 73-94.
Albers, H. J. and A. W. Ando. 2003. "Could State-Level Variation in the Number of Land Trusts
Make Economic Sense?" Land Economics 79(3): 311-327.
Andreoni, J. 1989. "Giving with Impure Altruism: Applications to Charity and Ricardian
Equivalence." Journal of Political Economy 97: 1447-58.
Andreoni, J. 1998. "Toward a Theory of Charitable Fundraising." Journal of Political Economy
106(6): 1186-1213.
Anselin, L. 1988. Spatial Econometrics: Methods and Models. Studies in Operational Regional
Science Series. London: Kluwer Academic.
Bay Lands Center. 2002. Bay Lands Center Homepage.
http://www.savebuzzrdsbav.org/bavlandsmap.html. [3/02]
Bell, K. P. and E. G. Irwin. 2002. "Spatially Explicit Micro-level Modelling of Land Use Change
at the Rural-Urban Interface." Agricultural Economics 27(3): 217-32.
Bergeron, N. and S. Polasky. 2000. "Species Conservation with Multiple Landowners."
Unpublished manuscript, University of Minnesota.
Bruekner, J. 1998. "Testing for Strategic Interaction among Local Governments: The Case of
Growth Controls." Journal of Urban Economics 44(3): 438-67.
Chomitz, K., and D. Gray. 1996. "Roads, Land Use, and Deforestation: A Spatial Model Applied
to Belize." World Bank Economic Review 10(3): 487-512.
Cropper, M., C. Griffiths, and M. Muthukumara. 2001. "Predicting the Location of
Deforestation: The Role of Roads and Protected Areas in North Thailand." Land
Economics 77(2): 172-186.
Goldsmith, E. 2001. Beyond Municipal Boundaries: Designing a Regional Land Conservation
Strategy in Washington County, Rhode Island. Brown University: Providence, RI.
http://www.envstudies.brown.edu/thesis/2001/goldsmith/writeup/dreamweaver/index.htm
L [3/02]
Hoctor, T. S., M. H. Carr, and P. D. Zwick. 2000. "Identifying a Linked Reserve System Using a
Regional Landscape Approach: The Florida Ecological Network." Conservation Biology
14(4): 984-1000.
64
-------
Irwin, E. G. 2002. "The Effects of Open Space on Residential Property Values." Land
Economics 78: 465-80.
Irwin, E. G. and N. E. Bockstael. 2002. "Interacting Agents, Spatial Externalities, and the
Evolution of Residential Land Use Patterns." Journal of Economic Geography 2: 31-54.
Kingma, B. R. 1989. "An Accurate Measurement of the Crowd-Out Effect, Income Effect, and
Price Effect for Charitable Contributions." Journal of Political Economy 97(5): 1197-
1207.
Lynch, L. and S. J. Lovell. 2003. "Combining Spatial and Survey Data to Explain Participation
in Agricultural Land Preservation Programs." Land Economics 79(2): 259-76.
MacFarlane, R. 1998. Implementing Agri-environment Policy: A Landscape Ecology
Perspective." Journal of Environmental Planning and Management 41(5): 575-96.
Maine Land Trust Network. 2002. AboutMLTN. http://www.mltn.org/about mltn.html. [3/02]
Marshall, E. 2004. "Open-Space Amenities, Interacting Agents, and Equilibrium Landscape
Structure." Land Economics 80(2):272-293.
Nelson, G. C., and D. Hellerstein. 1997. "Do Roads Cause Deforestation? Using Satellite Images
in Econometric Analysis of Land Use." American Journal of Agricultural Economics
79(1): 80-88.
Parker, N. and W. Thurman. 2004. "Crowding out Open Space: Federal Land Programs and
their Effects on Land Trust Activity." Unpublished manuscript.
Parkhurst, G. M., J. Shogren, C. Bastian, P. Kivi, J. Donner, and R. B. W. Smith. 2002.
"Agglomeration Bonus: An Incentive Mechanism to Reunite Fragmented Habitat for
Biodiversity Conservation." Ecological Economics 41(2): 305-28.
ReVelle, C. S., J. C. Williams, and J. J. Boland. 2002. "Counterpart Models in Facility Location
Science and Reserve Selection Science." Environmental Modeling and Assessment 7:71-
80.
Scott, J. M., R. J. F. Abbitt, and C.R. Groves. 2001. "What Are We Protecting? The United
States Conservation Portfolio." Conservation Biology in Practice 2: 18-19.
Swallow, S. K., P. J. Parks and D. N. Wear. 1990. "Policy Relevant Nonconvexities in the
Production of Multiple Forest Benefits." Journal of Environmental Economics and
Management 19(3): 281-296.
Swallow, S., P. Talukdar, and D. Wear. 1997. "Spatial and Temporal Specialization in Forest
Ecosystem Management under Sole Ownership." American Journal of Agricultural
Economics 79(1): 311-327.
65
-------
Tajibaeva, L., R. Haight, and S. Polasky. 2003. "A Discrete Space Urban Model with
Environmental Amenities." Unpublished manuscript, University of Minnesota. Available
at http://www.apec.umn.edu/facultv/spolaskvAVorking%20Papers.htm.
Turner, M. Forthcoming. "Landscape Preferences and Patterns of Residential Location." Journal
of Urban Economics.
Wu, J. and W. G. Boggess. 1999. "The Optimal Allocation of Conservation Funds." Journal of
Environmental Economics and Management 38(3):302-21.
66
-------
Table 1: Baseline Trust Parameters
Increasing Marginal Benefits
Decreasing Marginal Benefits
Parameter
Low
Budget
Mid
Budget
High
Budget
Low
Budget
Mid
Budget
High
Budget
ai
3
3
3
0.96
0.96
0.96
a2
1.5
1.5
1.5
0.91
0.91
0.91
P
2.4
2.4
2.4
0.8
0.8
0.8
Budgeti (parcels)
1
2
3
1
2
3
Budget2 (parcels)
2
2
2
3
3
3
In following tables "Number of equilibria" counts the number of "nontrivial equilibria." That is,
the equilibrium of 0000000 is not counted.
67
-------
Table 2: Crowding In and Out in Total Area Conserved
Increasing Marginal Benefits
Low Mid High
Budget Budget Budget
Decreasing Marginal Benefits
Low Mid High Very High
Budget Budget Budget Budget*
LT1 Leader
Number of equilibria
0.00
210.00
210.00
140.00
140.00
140.00
140.00
Ave total parcels conserved
0.00
4.00
5.00
4.00
4.00
4.00
4.00
Ave LT1 parcels conserved
0.00
2.00
3.00
1.00
1.00
1.00
1.00
Ave total adjacencies
0.00
1.71
2.86
1.71
1.71
1.71
1.71
Ave LT1 net benefit
0.00
59.20
117.80
2.98
2.98
2.98
2.98
Ave LT2 net benefit
0.00
3.20
6.38
1.13
1.13
1.13
1.13
Ave social benefit
0.00
64.00
125.00
3.78
3.78
3.78
3.78
Max LT1 net ben (fraction)
0(0)
59.2(1)
117.8(1)
2.98(1)
2.98(1)
2.98(1)
2.98(1)
Min LT1 net ben (fract.)
0(0)
59.2(1)
117.8(1)
2.98(1)
2.98(1)
2.98 (1)
2.98 (1)
Max LT2 net ben (fract.)
0(0)
3.2(1)
6.4(1)
1.13(1)
1.13(1)
1.13(1)
1.13(1)
Min LT2 net ben (fract.)
0(0)
3.2(1)
6.4(1)
1.13(1)
1.13(1)
1.13(1)
1.13(1)
LT2 Leader
Number of equilibria
42.00
210.00
210.00
140.00
210.00
140.00
35.00
Ave total parcels conserved
2.00
4.00
5.00
4.00
4.00
4.00
4.00
Ave LT1 parcels conserved
1.00
2.00
3.00
1.00
2.00
3.00
4.00
Ave total adjacencies
0.29
1.71
2.86
1.71
1.71
1.71
1.71
Ave LT1 net benefit
5.60
59.20
117.80
2.98
2.18
1.38
0.58
Ave LT2 net benefit
0.43
3.20
6.38
1.13
1.93
2.73
3.53
Ave social benefit
8.00
64.00
125.00
3.78
3.78
3.78
3.78
Max LT1 net ben (fract.)
5.6(1)
59.2(1)
117.8(1)
2.98(1)
2.18(1)
1.38(1)
0.58 (1)
Min LT1 net ben (fract.)
5.6(1)
59.2(1)
117.8(1)
2.98(1)
2.18(1)
1.38 (1)
0.58 (1)
Max LT2 net ben (fract.)
0.43 (1)
3.2(1)
6.4(1)
1.13(1)
1.93 (1)
2.73 (1)
3.53 (1)
Min LT2 net ben (fract.)
0.43 (1)
3.2(1)
6.4(1)
1.13(1)
1.93 (1)
2.73 (1)
3.53 (1)
Simultaneous
Number of equilibria
0.00
210.00
210.00
140.00
210.00
140.00
35.00
Ave total parcels conserved
0.00
4.00
5.00
4.00
4.00
4.00
4.00
Ave LT1 parcels conserved
0.00
2.00
3.00
1.00
2.00
3.00
4.00
Ave total adjacencies
0.00
1.71
2.86
1.71
1.71
1.71
1.71
Ave LT1 net benefit
0.00
59.20
117.80
2.98
2.18
1.38
0.58
Ave LT2 net benefit
0.00
3.20
6.38
1.13
1.93
2.73
3.53
Ave social benefit
0.00
64.00
125.00
3.78
3.78
3.78
3.78
Max LT1 net ben (fract.)
0(0)
59.2(1)
117.8(1)
2.98(1)
2.18(1)
1.38 (1)
0.58 (1)
Min LT1 net ben (fract.)
0(0)
59.2(1)
117.8(1)
2.98(1)
2.18(1)
1.38 (1)
0.58 (1)
Max LT2 net ben (fract.)
0(0)
3.2(1)
6.4(1)
1.13(1)
1.93 (1)
2.73 (1)
3.53 (1)
Min LT2 net ben (fract.)
0(0)
3.2(1)
6.4(1)
1.13(1)
1.93 (1)
2.73 (1)
3.53 (1)
Social Planner
Total parcels conserved
3.00
4.00
5.00
4.00
5.00
6.00
7.00
Total adjacencies
0.86
1.71
2.86
1.71
2.86
4.29
6.00
Net benefit
19.80
54.40
113.00
0.58
0.69
0.79
0.88
Social benefit
27.00
64.00
125.00
3.78
4.69
5.59
6.48
* Budgeti =4 parcels, Budget2 = 3 parcels
Note: for increasing marginal benefits, crowding in happens between budget low and mid, but not between budgets
mid and high. For decreasing marginal benefits, crowding out happens at each budget level.
68
-------
Table 3: DECREASING Stackelberg: LT1 (+) adjacency, LT2 0 adjacency value
Small Adjacency Value
(yi = 0.1, y2= 0)
Low Budget Mid Budget High Budget
Large Adjacency Value
(Yi= 1, Y2=0)
Low Budget Mid Budget High Budget
LT1 Leader
Number of equilibria
100.00
100.00
100.00
100.00
100.00
100.00
Ave total parcels conserved
4.00
4.00
4.00
4.00
4.00
4.00
Ave LT1 parcels conserved
1.00
1.00
1.00
1.00
1.00
1.00
Ave total adjacencies
1.80
1.80
1.80
1.80
1.80
1.80
Ave LT1 net benefit
3.16
3.16
3.16
4.78
4.78
4.78
Ave LT2 net benefit
1.13
1.13
1.13
1.13
1.13
1.13
Ave social benefit
3.96
3.96
3.96
5.58
5.58
5.58
Max LT1 net ben (fract.)
3.28 (0.14)
3.28 (0.14)
3.28 (0.14)
5.98 (0.14)
5.98 (0.14)
5.98 (0.14)
Min LT1 net ben (fract.)
2.98 (0.02)
2.98 (0.02)
2.98 (0.02)
2.98 (0.02)
2.98 (0.02)
2.98 (0.02)
Max LT2 net ben (fract.)
1.13(1)
1.13(1)
1.13(1)
1.13(1)
1.13(1)
1.13(1)
Min LT2 net ben (fract.)
1.13(1)
1.13(1)
1.13(1)
1.13(1)
1.13(1)
1.13(1)
LT2 Leader
Number of equilibria
68.00
57.00
16.00
68.00
57.00
16.00
Ave total parcels conserved
4.00
4.00
4.00
4.00
4.00
4.00
Ave LT1 parcels conserved
1.00
2.00
3.00
1.00
2.00
3.00
Ave total adjacencies
2.18
2.42
3.00
2.18
2.42
3.00
Ave LT1 net benefit
3.20
2.43
1.68
5.16
4.61
4.38
Ave LT2 net benefit
1.13
1.93
2.73
1.13
1.93
2.73
Ave social benefit
4.00
4.03
4.08
5.96
6.21
6.78
Max LT1 net ben (fract.)
3.28 (0.24)
2.48 (0.42)
1.68(1)
5.98 (0.24)
5.18 (0.42)
4.38 (1)
Min LT1 net ben (fract.)
3.08 (0.06)
2.38 (0.58)
1.68(1)
3.98 (0.06)
4.18 (0.58)
4.38 (1)
Max LT2 net ben (fract.)
1.13(1)
1.93 (1)
2.73 (1)
1.13(1)
1.93 (1)
2.73 (1)
Min LT2 net ben (fract.)
1.13(1)
1.93 (1)
2.73 (1)
1.13(1)
1.93 (1)
2.73 (1)
Simultaneous
Number of equilibria
68.00
57.00
16.00
68.00
57.00
16.00
Ave total parcels conserved
4.00
4.00
4.00
4.00
4.00
4.00
Ave LT1 parcels conserved
1.00
2.00
3.00
1.00
2.00
3.00
Ave total adjacencies
2.18
2.42
3.00
2.18
2.42
3.00
Ave LT1 net benefit
3.20
2.43
1.68
5.16
4.61
4.38
Ave LT2 net benefit
1.13
1.93
2.73
1.13
1.93
2.73
Ave social benefit
4.00
4.03
4.08
5.96
6.21
6.78
Max LT1 net ben (fract.)
3.28 (0.24)
2.48 (0.42)
1.68(1)
5.98 (0.24)
5.18 (0.42)
4.38 (1)
Min LT1 net ben (fract.)
3.08 (0.06)
2.38 (0.58)
1.68(1)
3.98 (0.06)
4.18 (0.58)
4.38 (1)
Max LT2 net ben (fract.)
1.13(1)
1.93 (1)
2.73 (1)
1.13(1)
1.93 (1)
2.73 (1)
Min LT2 net ben (fract.)
1.13(1)
1.93 (1)
2.73 (1)
1.13(1)
1.93 (1)
2.73 (1)
Social Planner
Total parcels conserved
4.00
5.00
6.00
4.00
5.00
6.00
Total adjacencies
3.00
4.00
5.00
3.00
4.00
5.00
Net benefit
0.88
1.09
1.29
3.58
4.69
5.79
Social benefit
4.08
5.09
6.09
6.78
8.69
10.59
Notes: When LT1 leads, it chooses not to crowd LT2. Since LT2 wants 3 parcels and doesn't care which location,
LT1 has such high expected adjacencies even if LT2 is random that the extra certainty of choosing adjacencies isn't
worth the additional cost to LT1. Increasing adjacency value has no effect on patterns, but does increase benefits.
69
-------
Table 4: DECREASING Marginal Benefits: LT1 (+) adj, LT2 (+) adj = 0.1, j2 = 0.1)
Baseline Parameters
Low Budget Mid Budget High Budget
Steeper LT2 (a2 = 0.85)
Low Budget Mid Budget High Budget
LT1 Leader
Number of equilibria
16.00
30.00
40.00
16.00
3.00
8.00
Ave total parcels conserved
4.00
5.00
6.00
4.00
5.00
6.00
Ave LT1 parcels conserved
1.00
2.00
3.00
1.00
2.00
3.00
Ave total adjacencies
3.00
4.00
5.00
3.00
4.00
5.00
Ave LT1 net benefit
3.28
3.49
3.69
3.28
3.49
3.69
Ave LT2 net benefit
1.43
2.33
3.21
1.15
1.93
2.69
Ave social benefit
4.08
5.09
6.09
4.08
5.09
6.09
Max LT1 net ben (fract.)
3.28(1)
3.49(1)
3.69(1)
3.28(1)
3.49(1)
3.69(1)
Min LT1 net ben (fract.)
3.28(1)
3.49(1)
3.69(1)
3.28(1)
3.49(1)
3.69(1)
Max LT2 net ben (fract.)
1.43(1)
2.33 (1)
3.21 (1)
1.15(1)
1.93 (1)
2.69(1)
Min LT2 net ben (fract.)
1.43(1)
2.33 (1)
3.21 (1)
1.15(1)
1.93 (1)
2.69(1)
LT2 Leader
Number of equilibria
16.00
30.00
40.00
16.00
24.00
16.00
Ave total parcels conserved
4.00
5.00
6.00
4.00
4.00
4.00
Ave LT1 parcels conserved
1.00
2.00
3.00
1.00
2.00
3.00
Ave total adjacencies
3.00
4.00
5.00
3.00
3.00
3.00
Ave LT1 net benefit
3.28
3.49
3.69
3.28
2.48
1.68
Ave LT2 net benefit
1.43
2.33
3.21
1.15
1.95
2.75
Ave social benefit
4.08
5.09
6.09
4.08
4.08
4.08
Max LT1 net ben (fract.)
3.28(1)
3.49(1)
3.69(1)
3.28(1)
2.48(1)
1.68(1)
Min LT1 net ben (fract.)
3.28(1)
3.49(1)
3.69(1)
3.28(1)
2.48(1)
1.68(1)
Max LT2 net ben (fract.)
1.43(1)
2.33 (1)
3.21 (1)
1.15(1)
1.95(1)
2.75 (1)
Min LT2 net ben (fract.)
1.43(1)
2.33 (1)
3.21 (1)
1.15(1)
1.95(1)
2.75 (1)
Simultaneous
Number of equilibria
16.00
34.00
40.00
16.00
30.00
33.00
Ave total parcels conserved
4.00
5.00
6.00
4.00
4.10
4.76
Ave LT1 parcels conserved
1.00
2.00
3.00
1.00
2.01
3.09
Ave total adjacencies
3.00
3.88
5.00
3.00
3.00
3.76
Ave LT1 net benefit
3.28
3.48
3.69
3.28
2.57
2.44
Ave LT2 net benefit
1.43
2.31
3.21
1.15
1.94
2.73
Ave social benefit
4.08
5.08
6.09
4.08
4.17
4.84
Max LT1 net ben (fract.)
3.28(1)
3.49 (0.88)
3.69(1)
3.28(1)
3.49 (0.10)
3.69 (0.24)
Min LT1 net ben (fract.)
3.28(1)
3.39 (0.12)
3.69(1)
3.28(1)
2.38(0.10)
1.68 (0.48)
Max LT2 net ben (fract.)
1.43(1)
2.33 (0.88)
3.21 (1)
1.15(1)
1.95 (0.80)
2.75 (0.48)
Min LT2 net ben (fract.)
1.43(1)
2.23 (0.12)
3.21 (1)
1.15(1)
1.85 (0.10)
2.69 (0.24)
Social Planner
Total parcels conserved
4.00
5.00
6.00
4.00
5.00
6.00
Total adjacencies
3.00
4.00
5.00
3.00
4.00
5.00
Net benefit
0.88
1.09
1.29
0.88
1.09
1.29
Social benefit
4.08
5.09
6.09
4.08
5.09
6.09
Notes: Using the baseline parameters, there is no crowding out of LT2. The extra bonus of adjacencies to LT2
overcomes its decreasing marginal benefits. The runs to the right show that for steeper decreasing marginal benefits
for LT2, the adjacency value no longer prevents crowding out. That is, (+) adjacency values offset decreasing
marginal benefits.
70
-------
Table 5: DECREASING
G Marginal Benefits: LTl (+) adj, LT2 (-) adj (yi
©
II
o
II
Baseline Parameters
Shallower LT2 (a2 =
0.96)
Low Budget
Mid Budget High Budget
Low Budget Mid Budget High Budget
LTl Leader
Number of equilibria
4.00
4.00
4.00
20.00
35.00
35.00
Ave total parcels conserved
4.00
4.00
4.00
4.00
5.00
5.00
Ave LTl parcels conserved
1.00
1.00
1.00
1.00
2.00
2.00
Ave total adjacencies
0.00
0.00
0.00
1.00
2.00
2.00
Ave LTl net benefit
2.98
2.98
2.98
3.08
3.29
3.29
Ave LT2 net benefit
1.13
1.13
1.13
1.28
2.09
2.09
Ave social benefit
3.78
3.78
3.78
3.88
4.89
4.89
Max LTl net ben (fract.)
2.98 (1)
2.98(1)
2.98 (1)
3.08(1)
3.29 (1)
3.29(1)
Min LTl net ben (fract.)
2.98 (1)
2.98(1)
2.98 (1)
3.08(1)
3.29 (1)
3.29(1)
Max LT2 net ben (fract.)
1.13(1)
1.13(1)
1.13(1)
1.28(1)
2.09 (1)
2.09(1)
Min LT2 net ben (fract.)
1.13(1)
1.13(1)
1.13(1)
1.28(1)
2.09 (1)
2.09(1)
LT2 Leader
Number of equilibria
4.00
6.00
5.00
4.00
48.00
18.00
Ave total parcels conserved
4.00
2.00
3.00
4.00
5.00
5.00
Ave LTl parcels conserved
1.00
2.00
3.00
1.00
2.00
3.00
Ave total adjacencies
1.00
1.00
2.00
1.00
3.00
3.00
Ave LTl net benefit
3.08
0.45
0.67
3.08
3.39
2.59
Ave LT2 net benefit
1.03
1.78
2.52
1.28
1.99
2.79
Ave social benefit
3.88
2.05
3.07
3.88
4.99
4.99
Max LTl net ben (fract.)
3.08 (1)
0.45 (1)
0.67 (1)
3.08(1)
3.39 (1)
2.59(1)
Min LTl net ben (fract.)
3.08 (1)
0.45 (1)
0.67 (1)
3.08(1)
3.39 (1)
2.59(1)
Max LT2 net ben (fract.)
1.03 (1)
1.78(1)
2.52 (1)
1.28(1)
1.99 (1)
2.79(1)
Min LT2 net ben (fract.)
1.03 (1)
1.78(1)
2.52 (1)
1.28(1)
1.99 (1)
2.79(1)
Simultaneous
Number of equilibria
0.00
0.00
0.00
2.00
0.00
6.00
Ave total parcels conserved
0.00
0.00
0.00
4.00
0.00
5.00
Ave LTl parcels conserved
0.00
0.00
0.00
1.00
0.00
3.00
Ave total adjacencies
0.00
0.00
0.00
1.00
0.00
3.00
Ave LTl net benefit
0.00
0.00
0.00
3.08
0.00
2.59
Ave LT2 net benefit
0.00
0.00
0.00
1.28
0.00
2.79
Ave social benefit
0.00
0.00
0.00
3.88
0.00
4.99
Max LTl net ben (fract.)
0(0)
0(0)
0(0)
3.08(1)
0(0)
2.59 (1)
Min LTl net ben (fract.)
0(0)
0(0)
0(0)
3.08(1)
0(0)
2.59 (1)
Max LT2 net ben (fract.)
0(0)
0(0)
0(0)
1.28(1)
0(0)
2.79 (1)
Min LT2 net ben (fract.)
0(0)
0(0)
0(0)
1.28(1)
0(0)
2.79 (1)
Social Planner
Total parcels conserved
4.00
5.00
6.00
4.00
5.00
6.00
Total adjacencies
3.00
4.00
5.00
3.00
4.00
5.00
Net benefit
0.88
1.09
1.29
0.88
1.09
1.29
Social benefit
4.08
5.09
6.09
4.08
5.09
6.09
Notes: With baseline, no Nash equilibrium. The negative adjacency value for LT2 drops its benefit curve too low to
conserve at all. Competing adjacency preferences make it difficult to find equilibria for simultaneous game. Runs to the
right increase the benefit curve for LT2, offsetting negative adjacency values on total parcels, but strange, erratic
equilibria result. As budgetl goes from 1 to 2 to 3, parcels conserved in the simultaneous game are (1,3), (0,0), and (3,2).
71
-------
Table 6: INCREASING Marginal Benefits: LT1 (+) ad.j, LT2 0 ad.j
Small Adjacency Value
(yi =0.1,y2 = 0)
Low Budget Mid Budget High Budget
Large Adjacency Value
(Yi = 1, Y2 = 0)
Low Budget Mid Budget High Budget
LT1 Leader
Number of equilibria
0.00
40.00
18.00
0.00
40.00
18.00
Ave total parcels conserved
0.00
4.00
5.00
0.00
4.00
5.00
Ave LT1 parcels conserved
0.00
2.00
3.00
0.00
2.00
3.00
Ave total adjacencies
0.00
2.10
3.33
0.00
2.10
3.33
Ave LT1 net benefit
0.00
59.41
118.13
0.00
61.30
121.13
Ave LT2 net benefit
0.00
3.20
6.38
0.00
3.20
6.38
Ave social benefit
0.00
64.21
125.33
0.00
66.10
128.33
Max LT1 net ben (fract.)
0(0)
59.5 (0.25)
118.2(0.39)
0(0)
62.2 (0.25)
121.8 (0.39)
Min LT1 net ben (fract.)
0(0)
59.3 (0.15)
118.0(0.06)
0(0)
60.2(0.15)
119.8 (0.06)
Max LT2 net ben (fract.)
0(0)
3.2(1)
6.3803 (1)
0(0)
3.2(1)
6.3803 (1)
Min LT2 net ben (fract.)
0(0)
3.2(1)
6.3803 (1)
0(0)
3.2(1)
6.3803 (1)
LT2 Leader
Number of equilibria
12.00
57.00
48.00
12.00
57.00
48.00
Ave total parcels conserved
2.00
4.00
5.00
2.00
4.00
5.00
Ave LT1 parcels conserved
1.00
2.00
3.00
1.00
2.00
3.00
Ave total adjacencies
1.00
2.42
3.63
1.00
2.42
3.63
Ave LT1 net benefit
5.70
59.44
118.16
6.60
61.62
121.43
Ave LT2 net benefit
0.43
3.20
6.38
0.43
3.20
6.38
Ave social benefit
8.10
64.24
125.36
9.00
66.42
128.63
Max LT1 net ben (fract.)
5.7(1)
59.5 (0.42)
118.2(0.63)
6.6(1)
62.2 (0.42)
121.8 (0.63)
Min LT1 net ben (fract.)
5.7(1)
59.4 (0.58)
118.1 (0.38)
6.6(1)
61.2(0.58)
120.8 (0.38)
Max LT2 net ben (fract.)
0.43 (1)
3.2(1)
6.3803 (1)
0.43 (1)
3.2(1)
6.3803 (1)
Min LT2 net ben (fract.)
0.43 (1)
3.2(1)
6.3803 (1)
0.43 (1)
3.2(1)
6.3803 (1)
Simultaneous
Number of equilibria
0.00
57.00
48.00
0.00
57.00
48.00
Ave total parcels conserved
0.00
4.00
5.00
0.00
4.00
5.00
Ave LT1 parcels conserved
0.00
2.00
3.00
0.00
2.00
3.00
Ave total adjacencies
0.00
2.42
3.63
0.00
2.42
3.63
Ave LT1 net benefit
0.00
59.44
118.16
0.00
61.62
121.43
Ave LT2 net benefit
0.00
3.20
6.38
0.00
3.20
6.38
Ave social benefit
0.00
64.24
125.36
0.00
66.42
128.63
Max LT1 net ben (fract.)
0(0)
59.5 (0.42)
118.2(0.63)
0(0)
62.2 (0.42)
121.8 (0.63)
Min LT1 net ben (fract.)
0(0)
59.4 (0.58)
118.1 (0.38)
0(0)
61.2(0.58)
120.8 (0.38)
Max LT2 net ben (fract.)
0(0)
3.2(1)
6.3803 (1)
0(0)
3.2(1)
6.3803 (1)
Min LT2 net ben (fract.)
0(0)
3.2(1)
6.3803 (1)
0(0)
3.2(1)
6.3803 (1)
Social Planner
Total parcels conserved
3.00
4.00
5.00
3.00
4.00
5.00
Total adjacencies
2.00
3.00
4.00
2.00
3.00
4.00
Net benefit
20.00
54.70
113.40
21.80
57.40
117.00
Social benefit
27.20
64.30
125.40
29.00
67.00
129.00
Note: (+) LT1 adjacency values change agglomeration patterns but not overall crowding in. large adjacency values
don't affect patterns - they only increase benefits for LT1.
72
-------
Table 7: INCREASING Marginal Benefits: LT1 (+) ad.j, LT2 (+) ad.j
Small Adjacency Values
Large Adjacency Values
(Yi
= 0.1,y2 = 0.1)
(Yi = 1, Y2 = 1)
Low
Mid
High
Very Low
Low
Mid
High
Budget
Budget
Budget
Budget*
Budget
Budget
Budget
LT1 Leader
Number of equilibria
0.00
24.00
30.00
0.00
15.00
24.00
30.00
Ave total parcels conserved
0.00
4.00
5.00
0.00
3.00
4.00
5.00
Ave LT1 parcels conserved
0.00
2.00
3.00
0.00
1.00
2.00
3.00
Ave total adjacencies
0.00
3.00
4.00
0.00
2.00
3.00
4.00
Ave LT1 net benefit
0.00
59.50
118.20
0.00
26.60
62.20
121.80
Ave LT2 net benefit
0.00
3.50
6.78
0.00
2.40
6.20
10.38
Ave social benefit
0.00
64.30
125.40
0.00
29.00
67.00
129.00
Max LT1 net ben (fract.)
0(0)
59.5 (1)
118.2(1)
0(0)
26.6 (1)
62.2 (1)
121.8 (1)
Min LT1 net ben (fract.)
0(0)
59.5 (1)
118.2(1)
0(0)
26.6 (1)
62.2 (1)
121.8 (1)
Max LT2 net ben (fract.)
0(0)
3.5(1)
6.8(1)
0(0)
2.4(1)
6.2 (1)
10.4 (1)
Min LT2 net ben (fract.)
0(0)
3.5(1)
6.8(1)
0(0)
2.4(1)
6.2 (1)
10.4 (1)
LT2 Leader
Number of equilibria
15.00
24.00
30.00
0.00
15.00
24.00
30.00
Ave total parcels conserved
3.00
4.00
5.00
0.00
3.00
4.00
5.00
Ave LT1 parcels conserved
1.00
2.00
3.00
0.00
1.00
2.00
3.00
Ave total adjacencies
2.00
3.00
4.00
0.00
2.00
3.00
4.00
Ave LT1 net benefit
24.80
59.50
118.20
0.00
26.60
62.20
121.80
Ave LT2 net benefit
0.60
3.50
6.78
0.00
2.40
6.20
10.38
Ave social benefit
27.20
64.30
125.40
0.00
29.00
67.00
129.00
Max LT1 net ben (fract.)
24.8 (1)
59.5 (1)
118.2(1)
0(0)
26.6 (1)
62.2 (1)
121.8 (1)
Min LT1 net ben (fract.)
24.8 (1)
59.5 (1)
118.2(1)
0(0)
26.6 (1)
62.2 (1)
121.8 (1)
Max LT2 net ben (fract.)
0.6 (1)
3.5(1)
6.8(1)
0(0)
2.4(1)
6.2 (1)
10.4 (1)
Min LT2 net ben (fract.)
0.6 (1)
3.5(1)
6.8(1)
0(0)
2.4(1)
6.2 (1)
10.4 (1)
Simultaneous
Number of equilibria
0.00
32.00
34.00
0.00
15.00
32.00
34.00
Ave total parcels conserved
0.00
4.00
5.00
0.00
3.00
4.00
5.00
Ave LT1 parcels conserved
0.00
2.00
3.00
0.00
1.00
2.00
3.00
Ave total adjacencies
0.00
2.75
3.88
0.00
2.00
2.75
3.88
Ave LT1 net benefit
0.00
59.48
118.19
0.00
26.60
61.95
121.68
Ave LT2 net benefit
0.00
3.48
6.77
0.00
2.40
5.95
10.26
Ave social benefit
0.00
64.28
125.39
118.2
0.00
29.00
66.75
128.88
121.8
Max LT1 net ben (fract.)
0(0)
59.5 (0.75)
(0.88)
118.1
0(0)
26.6(1)
62.2 (0.75)
(0.88)
120.8
Min LT1 net ben (fract.)
0(0)
59.4 (0.25)
(0.12)
0(0)
26.6(1)
61.2(0.25)
(0.12)
Max LT2 net ben (fract.)
0(0)
3.5 (0.75)
6.8 (0.88)
0(0)
2.4 (1)
6.2 (0.75)
10.4 (0.88)
Min LT2 net ben (fract.)
0(0)
3.4 (0.25)
6.7(0.12)
0(0)
2.4 (1)
5.2 (0.25)
9.4 (0.12)
Social Planner
Total parcels conserved
3.00
4.00
5.00
2.00
3.00
4.00
5.00
Total adjacencies
2.00
3.00
4.00
1.00
2.00
3.00
4.00
Net benefit
20.00
54.70
113.40
4.20
21.80
57.40
117.00
Social benefit
27.20
64.30
125.40
9.00
29.00
67.00
129.00
* Budgeti = 0 parcels, Budget2 = 2 parcels. Note: Small adjacency values are similar to the results from the LT1 (+)
and LT2 (0) case; change agglomeration patterns even further, but no change to overall crowding in. Large adjacency
values (especially for LT2) do affect crowding-in (CI). CI occurs at lower budget/parcel conservation for LT1.
73
-------
Table 8: INCREASING marginal benefits: LT1 (+) ad.j, LT2 (-) ad.j
Small Adjacency Values
Large Adjacency Values
(Yi =
0.1, y2 = -0.1)
(Yi
= 1,Y2 = -1)
Low
Mid
High
Low
Mid
High
Budget
Budget
Budget
Budget
Budget
Budget
LT1 Leader
Number of equilibria
0.00
50.00
18.00
0.00
12.00
3.00
Ave total parcels conserved
0.00
4.00
5.00
0.00
4.00
5.00
Ave LT1 parcels conserved
0.00
2.00
3.00
0.00
2.00
3.00
Ave total adjacencies
0.00
1.00
3.00
0.00
1.00
2.00
Ave LT1 net benefit
0.00
59.30
118.10
0.00
60.20
119.80
Ave LT2 net benefit
0.00
3.10
6.08
0.00
2.20
4.38
Ave social benefit
0.00
64.10
125.30
0.00
65.00
127.00
Max LT1 net ben (fract.)
0(0)
59.3 (1)
118.1 (1)
0(0)
60.2(1)
119.8 (1)
Min LT1 net ben (fract.)
0(0)
59.3 (1)
118.1 (1)
0(0)
60.2(1)
119.8 (1)
Max LT2 net ben (fract.)
0(0)
3.1(1)
6.1(1)
0(0)
2.2(1)
4.4 (1)
Min LT2 net ben (fract.)
0(0)
3.1(1)
6.1(1)
0(0)
2.2(1)
4.4 (1)
LT2 Leader
Number of equilibria
12.00
33.00
18.00
0.00
6.00
18.00
Ave total parcels conserved
2.00
4.00
5.00
0.00
2.00
5.00
Ave LT1 parcels conserved
1.00
2.00
3.00
0.00
2.00
3.00
Ave total adjacencies
1.00
2.00
3.00
0.00
1.00
3.00
Ave LT1 net benefit
5.70
59.40
118.10
0.00
4.20
120.80
Ave LT2 net benefit
0.33
3.00
6.08
0.00
1.83
3.38
Ave social benefit
8.10
64.20
125.30
0.00
9.00
128.00
Max LT1 net ben (fract.)
5.7(1)
59.4(1)
118.1 (1)
0(0)
4.2(1)
120.8 (1)
Min LT1 net ben (fract.)
5.7(1)
59.4(1)
118.1 (1)
0(0)
4.2(1)
120.8 (1)
Max LT2 net ben (fract.)
0.3(1)
3(1)
6.1(1)
0(0)
1.8(1)
3.4(1)
Min LT2 net ben (fract.)
0.3(1)
3(1)
6.1(1)
0(0)
1.8(1)
3.4(1)
Simultaneous
Number of equilibria
0.00
0.00
6.00
0.00
0.00
0.00
Ave total parcels conserved
0.00
0.00
5.00
0.00
0.00
0.00
Ave LT1 parcels conserved
0.00
0.00
3.00
0.00
0.00
0.00
Ave total adjacencies
0.00
0.00
3.00
0.00
0.00
0.00
Ave LT1 net benefit
0.00
0.00
118.10
0.00
0.00
0.00
Ave LT2 net benefit
0.00
0.00
6.08
0.00
0.00
0.00
Ave social benefit
0.00
0.00
125.30
0.00
0.00
0.00
Max LT1 net ben (fract.)
0(0)
0(0)
118.1 (1)
0(0)
0(0)
0(0)
Min LT1 net ben (fract.)
0(0)
0(0)
118.1 (1)
0(0)
0(0)
0(0)
Max LT2 net ben (fract.)
0(0)
0(0)
6.1(1)
0(0)
0(0)
0(0)
Min LT2 net ben (fract.)
0(0)
0(0)
6.1(1)
0(0)
0(0)
0(0)
Social Planner
Total parcels conserved
3.00
4.00
5.00
3.00
4.00
5.00
Total adjacencies
2.00
3.00
4.00
2.00
3.00
4.00
Net benefit
20.00
54.70
113.40
21.80
57.40
117.00
Social benefit
27.20
64.30
125.40
29.00
67.00
129.00
Note: small (-) adjacency values prevent crowding in of LT2 until budgetl reaches high (rather than mid). Large (-)
adjacency values prevent LT2 from crowding in at all.
74
-------
Table 9: Hot Spots
Low Budget, (+)
Adj Values (yl =
0.1, y2 = 0.1)
Mid Budget, (+) Adj Mid Budget, Mixed
Values (yl = 0.1, y2 Adj Values (yl =
= 0.1) 0.1, y2 = -1.1)
LT1 Leader
Number of equilibria
2.00
3.00
2.00
Ave total parcels conserved
2.00
3.00
3.00
Ave LT1 parcels conserved
1.00
2.00
2.00
Ave total adjacencies
1.00
2.00
2.00
Ave LT1 net benefit
24.70
59.40
59.40
Ave LT2 net benefit
1.87
3.86
1.46
Ave social benefit
27.10
64.20
64.20
Max LT1 net ben (fract)
24.7(1)
59.4(1)
59.4(1)
Min LT1 net ben (fract)
24.7(1)
59.4(1)
59.4(1)
Max LT2 net ben (fract)
1.87(1)
3.86(1)
1.46(1)
Min LT2 net ben (fract)
1.87(1)
3.86(1)
1.46(1)
LT2 Leader
Number of equilibria
1.00
2.00
2.00
Ave total parcels conserved
1.00
2.00
2.00
Ave LT1 parcels conserved
1.00
2.00
2.00
Ave total adjacencies
0.00
1.00
1.00
Ave LT1 net benefit
5.60
22.30
22.30
Ave LT2 net benefit
2.46
4.27
3.07
Ave social benefit
8.00
27.10
27.10
Max LT1 net ben (fract)
5.6(1)
22.3 (1)
22.3 (1)
Min LT1 net ben (fract)
5.6(1)
22.3 (1)
22.3 (1)
Max LT2 net ben (fract)
2.46(1)
4.27(1)
3.07(1)
Min LT2 net ben (fract)
2.46(1)
4.27(1)
3.07(1)
Simultaneous
Number of equilibria
3.00
5.00
4.00
Ave total parcels conserved
1.67
2.60
2.50
Ave LT1 parcels conserved
1.11
2.08
2.08
Ave total adjacencies
0.67
1.60
1.50
Ave LT1 net benefit
18.33
44.56
40.85
Ave LT2 net benefit
2.07
4.03
2.27
Ave social benefit
20.73
49.36
45.65
Max LT1 net ben (fract)
24.7 (0.67)
59.4 (0.6)
59.4 (0.5)
Min LT1 net ben (fract)
5.6 (0.33)
22.3 (0.4)
22.3 (0.5)
Max LT2 net ben (fract)
2.46 (0.33)
4.27 (0.4)
3.07 (0.5)
Min LT2 net ben (fract)
1.87 (0.67)
3.86 (0.6)
1.46 (0.5)
Social Planner
Total parcels conserved
3.00
4.00
4.00
Total adjacencies
2.00
3.00
3.00
Net benefit
57.00
115.70
115.70
Social benefit
64.20
125.30
125.30
Notes: In low budget case, LT1 conserves 1 and LT2 conserves 2 - in mid budget, LT1 conserves 2 and LT2
conserves 1. No major differences except fraction of equilibria in which LT2 does not conserve is 1/3 in column 1 &
2/5 in column 2. Paragraph in text refers to column 1.
75
-------
Table 10: Summary Statistics of Variables in Illinois
Variable Name
Mean
S.D.
Min.
Max.
Privately protected areas
45.38
287.92
0
9397.73
(acres)
Publicly protected areas
489.1
2368.37
0
23366
(acres)
Population density
0.4466
1.545
0
23.09
(population/acre, year 2000)
Median household income
44,592
12,697
7.21
146,551
($, year 2000)
High school graduates as % of people over age 25
38.2
7.6
5.8
64.2
(%, year 2000)
College graduates as % of people over age 25
10.4
5.5
0.4
40.6
(%, year 2000)
Elevation heterogeneity
0.058
0.042
0.001
0.277
(standard deviation of elevation / mean elevation)
Cost of land
1651
1090
624
6141
($/acre)
Mean distance from municipal boundaries
3.62
1.018
0
28.8
(miles)
Area of surface water
867
17,349
0
570,600
(acres)
Area of agricultural land
16,272
8,067
0
54,156
(acres)
Area of forest
2,454
2,968
0
46,211
(acres)
Area of urban land
1,376
4,433
1.12
138,200
(acres)
Area of wetland
836
1,095
0
11,339
(acres)
-------
Table 11: Summary Statistics of Variables in Massachusetts
Variable Name
Mean
S.D.
Min.
Max.
Privately protected areas
1,664
2,045
0
11,815
(acres)
Publicly protected areas
2,810
2,891
4.49
21,856
(acres)
Population density
1.92
3.58
0
29.25
(population/acre, year 2000)
Median household income
63,014
19,656
29,861
160,084
($, year 2000)
High school graduates as % of people over age 25
26
9.9
0
46
(%, year 2000)
College graduates as % of people over age 25
20
8.6
0
44
(%, year 2000)
Elevation heterogeneity
0.374
0.234
0.071
1.182
(standard deviation of elevation / mean elevation)
Cost of land
6,803
5,119
0
42,565
($/acre)
Mean distance from municipal boundaries
2.97
3.6
0
19.7
(miles)
Area of surface water
543
850
6.67
9,210
(acres)
Area of agricultural land
1,013
1,142
0
7,372
(acres)
Area of forest
8,422
6,607
0
40,837
(acres)
Area of open urban land
433
478
1.78
5,649
(acres)
Area of wetland
451
557
0
4,970
(acres)
Area of priority habitat
2,390
3,377
0
28,974
(acres)
77
-------
Table 12: Results of Spatial-Lag Econometric Estimation for Illinois
Variable
Coef.
S.E.
Significance
Neighbors' private protected area
.33
.035
***
Publicly protected land
-.0069
.0032
**
Population density
-16.5
6.82
**
Median income
-.00030
.00088
High school
-111.02
119.6
College
-127.7
196.9
Elevation heterogeneity
439.5
212.5
**
Cost of land
.052
.013
**
Distance to city
-4.16
3.58
Surface water area
-.00028
.00044
Agriculture
-1.6 e"5
.001
Forest
.0048
.0028
*
Developed land
.00099
.0018
Wetland
.040
.0067
*
Constant
-34.1
62.6
N
R2
Log-likelihood
1691
.13
-11905.8
Notes:
1) The dependent variable is the number of acres of privately protected land in a township.
2) *** significant at 1% level; ** significant at 5% level; * significant at 10% level.
-------
Table 13: Results of Spatial-Lag Model for Massachusetts
Variable
Coef.
S.E.
Significance
Neighbors' private protected area
.24
.060
***
Publicly protected land
-.27
.048
***
Population density
21.2
30.8
Median income
.00022
.0071
High school
-2051.3
1032.6
**
College
2172.7
14706
Elevation heterogeneity
432.4
485.1
Cost of land
.0013
.0211
Distance to city
21.7
30.4
Surface water area
-.14
.12
Agriculture
.32
.090
***
Forest
.20
.025
***
Open urban land
.36
.19
*
Wetland
.18
.17
Priority habitat
.13
.033
***
Constant
-662.0
709.1
N
R2
Log-likelihood
351
.49
-3062.68
Notes:
1) The dependent variable is the number of acres of privately protected land in a township.
2) *** significant at 1% level; ** significant at 5% level; * significant at 10% level.
-------
Figure 1: Private Protected Areas in Illinois and Massachusetts
Note: Townships are shaded in increasing order of quintiles. The bottom quintile in Illinois is
equal to zero acreage.
-------
Pixels in place of parcels: Modeling urban growth using
information derived from satellite imagery
Rich Iovanna & Colin Vance
USEPA, National Center for Environmental Economics &
German Aerospace Center, Institute of Transportation Research
-------
Pixels in place of parcels: Modeling urban growth using information derived from
satellite imagery
Introduction
Over the past two decades, the conversion of farm and forestlands on city fringes
throughout the United States has continued unabated, with the urbanized area expanding from
approximately 51 to 76 million acres between 1982 and 1997 (Fulton et al, 2001). While partly
reflecting growing prosperity and preferences for increased living space, this trend has raised
concerns on several fronts. Through its strong association to the increase in impervious surfaces,
expansion of the urban frontier degrades and fragments natural habitats, contributes to poor air
quality through increased reliance on vehicle travel, and disrupts a multitude of ecosystem
services such as aquifer recharge and nutrient cycling. Such disruptions can impose significant
costs on municipalities, including damage from flooding, higher medical costs for air quality-
related illnesses, and increased expenditures for the provision of public services and
infrastructure. Social and aesthetic costs may further compound these ecological and health
impacts. The movement of populations away from central city areas has been argued to not only
contribute to urban blight (Jargowsky, 2001), but also to a loss of cultural heritage as farmland
and forest is replaced by what is often a pattern of helter-skelter development characterized by
strip malls, office parks and disconnected residential communities (Kunstler, 1994).
To the extent that development decisions create landscape mosaics that alter ecological
function and constrain the choice set of future land-use alternatives, efforts to understand urban
expansion have the potential to contribute greatly to land-use planning and environmental policy
processes. Models that are fine scale and spatially explicit are particularly meaningful because
ecologists and allied disciplines perceive an intimate connection between the provision of habitat
and other services by ecosystems and the pattern of the landscape mosaic in which the
ecosystems function. Complex landscape patterns attributable to development disproportionately
82
-------
impact the environment by fragmenting ecosystems and increasing the ratio of edge to interior
extent. Similarly, where development takes place (location in the watershed and proximity to
water bodies), rather than how much, is of paramount importance when considering how
development stresses aquatic ecosystems.
Attention toward the fine scale and spatial explicitness is noticeable in the recent
economics literature, with a recurring theme being how the spatial configuration of land use, by
virtue of its association to both accessibility and spatially determined externalities, is itself an
important determinant of conversion of open space to developed uses. In recent years, an
increasing number of studies have combined principles from landscape ecology with spatial-
econometric methods to account for how human decision-making, ecosystem function, and their
interaction effect landscape changes across different spatial scales (e.g. Turner, Wear and Flamm,
1996; Geoghegan, Wainger and Bockstael, 1997; Kline, Moses and Alig, 2001; Irwin and
Bockstael, 2002). Pioneering work in this area was undertaken by Geoghegan, Wainger and
Bockstael (1997), who capture externality effects on land values by including explanatory
variables measuring landscape diversity and fragmentation in a cross-sectional hedonic
regression. Subsequent work by Irwin and Bockstael (2002) adds a temporal dimension by
specifying the Cox proportional hazards model to examine the influence of spatially determined
spillover effects on the likelihood of land-use conversion. They capture spillover effects by
including a spatial explanatory variable that measures the percent of land already developed in a
roughly one-mile radius surrounding the parcel, but, unlike Geoghegan, Waigner, and Bockstael,
they include no control for the extent to which this development is fragmented.
Motivated by the hypothesis that "urban spatial structure is determined by
interdependences among spatially distributed agents," our efforts take as their point of departure
the work of Irwin and Bockstael (2002: 32). While the present paper further explores the role of
landscape pattern in land-use change, we decouple our exploration from reliance on parcel-level
data. By focusing on a consistent and finer unit of observation (a 60 X 60-meter satellite pixel),
83
-------
we develop a model that ventures beyond Irwin and Bockstael's ability to predict where
development may occur, to both where and by how much.1 Significantly, the structural equations
we develop indicate that neither price nor lot size data are required for our modeling approach. A
further implication is that the effect of variables measuring pixel-level amenities on the likelihood
of conversion is an empirical question, the sign of which cannot be hypothesized a priori.
From a dynamic profit- and consumer surplus-maximizing framework we derive an
empirical model to identify the determinants of land conversion from commodity-based to urban
uses across a 25,900 square kilometer swath in central North Carolina, an area that has undergone
extensive change over the last two decades. The data we use to estimate the model come
primarily from five satellite images spanning the years 1976-2001. Using 60 X 60-meter satellite
pixels as the unit of observation, we subsequently test for the significance of these factors with a
complementary log-log model derived from the proportional-hazards specification.
The model estimated has several distinguishing features: Unlike the Cox model, which
conditions out the parameters corresponding to the dynamics of the process being modeled, the
complementary log-log specification affords great flexibility for parameterizing the effect of time.
Because the data are observed at a very fine level of spatial resolution, we can additionally relax
the assumption commonly invoked in land-use shares models that all change occurs at the rural
urban interface (Hardie et al., 2000). Finally, the model includes a broad array of time-varying
covariates that measure the land allocation response to site, locational, and pattern attributes
associated with each pixel.
Our empirical results confirm the hypothesis that pixel-level characteristics - particularly
what surrounds a pixel - have a major influence on the likelihood of its conversion. We also find
that the omission of landscape pattern variables can lead to biased inferences regarding the
influence of other covariates, such as proximity to road and urban centers, which are commonly
identified as important determinants of land-use change. Finally, we uncover some
counterintuitive results with respect to the effects of amenities on the hazard of conversion,
84
-------
results that are suggested by our theoretical model to be attributed to trade-offs between plot
quality and plot size in the market for undeveloped land.
The Study Region
The study region straddles portions of the Piedmont and the Inner Coastal Plane of North
Carolina, two distinct physiogeographic zones that cut diagonally north-south across the state
(Figure 1). Across the state as a whole, hardwoods cover more than half of the timberland
acreage, while pine stands and oak-pine stands account for the remaining 33 and 14 percent,
respectively (Brown, 1993; Brownlow, Lineback and DeHart, 2000). Centuries of human
occupation have fragmented these forests into a patchwork that now includes croplands, fields in
varying stages of abandonment, and, increasingly, built-up areas.
INSERT FIGURE 1 HERE.
North Carolina is widely regarded as a state in which inefficiencies in land allocation are
leading to excessively costly expansion of the built environment. A highly publicized report
recently released from Smart Growth America (Ewing, Pendall and Chen, 2003) ranked
Greensboro and Raleigh-Durham as second and third among a listing of 83 U.S. cities in which
the spread of development far outpaces population growth. In Raleigh, for example, the
population increased by 32 percent between 1990 and 1996, while its urbanized land area
increased nearly twofold (Sierra Club, 1998).
Historical accounts suggest that the foundations for the sprawling patterns observed in
these and other North Carolina cities can be traced back to the 1880s, when a low-density urban
landscape emerged as a result of the proliferation of tobacco factories and textile mills (Orr and
Stuart, 2000; Ingals, 2000). These employment centers spawned a dispersed network of small
towns across the state that today serve as bedroom communities for regional metropolitan centers.
By 1900 there were 177 mills in the state, with over 90% of them in the Piedmont (Ingals, 2000).
To connect these emergent centers of economic activity, major investments in road infrastructure
85
-------
were undertaken with the result that by the early 1920s there were over 5,500 miles of roads
paved linking county seats (Ingals, 2000). These developments ushered in a transition from an
economy based largely on agriculture to one based on the service sector and on manufacturing,
with heavy reliance on the forest-products sector.
Although the state remains a major producer of tobacco, sweet potatoes, and hog
products, the area under agriculture has declined drastically since its peak in the early 1900s
(Lilly, 1998). The area under commercial timberland, by contrast, has remained relatively stable,
peaking in the early 1970s at 20.13 million acres and then dropping back down to approach the
1938 level of 18.1 million acres by 1990 (Brown, 1993). Nevertheless, a recent U.S. Forest
Service report projected that North Carolina will lose 30% of its privately owned, natural forest
by 2040, with the Interstate 85 corridor extending southward from Raleigh-Durham designated as
a "hotspot" of forest loss due to continuing urbanization (Prestemon and Abt, 2002; Wear and
Greis, 2002).
Formalization
Responding to concern about the rate and extent of land-use change requires
understanding the causes, timing and location of land-use change. If we know why and when
pressures to develop increase for a given tract, we will be in a better position to evaluate where
significant ecological consequences are likely to occur as well as the merit of conservation
responses. The decision to convert depends on a complex multiplicity of factors, including the
market value of output from the land in alternative uses, expectations about the future use of
neighboring lands, and the surrounding composition of land ownership. Following the work of
Capozza an Helsley (1989) and Boscolo, Kerr, Pfaff, and Sanchez (1998), the theoretical
approach taken here attempts to structure this complexity by assuming that a unit of land (referred
to hereafter as a 'pixel' to keep this discussion consistent with the data we ultimately use) will be
converted if the net present discounted benefits of doing so are greater than the net present
86
-------
discounted benefits of leaving the land under its present use. In other words, the land manager
converts pixel /' in period Tto maximize the following objective function:
T oo
(1) Maxr^A„S: +^D„S' -CrST ,
t=0 t=T
where
Ait is the return derived from a commodity-based use of the pixel in period t, i.e., the
agricultural or forestry rent;
Df is the return to development in period t, i.e., the development rent;
CT is the cost associated with conversion; and
S is the discount rate, l/(l+r).
Assuming irreversibility of the conversion process, there are two necessary conditions for
conversion to take place: The first is that the discounted stream of returns derived from
conversion are greater than that of leaving the plot in its present use, net of the one-time
conversion costs:
CO
(2) £(D„-4,)5'-Cr>0.
t= 0
The operative condition, however, is one that would be met well after that specified by
equation (2). Conversion will occur when the development rent just equals the opportunity cost,
()(', of developing that period as opposed to the next. Assuming development rents are rising
over time and conversion costs are declining, it is more profitable to the land owner to defer
development for at least another period before time T.2 After T, the landowner loses money every
period that development is deferred. More formally, a developed pixel is one in which
(3) Du>oq= 4,+(q-«;w).
If the development rent in period t exceeds the sum of agricultural rent and the cost
savings from deferring development, which relates to downward trend in costs as well as the fact
that costs are discounted an additional period, the pixel has already been developed. With
87
-------
equality, time Tis when conversion actually takes place. Equation 3 indicates and Figure 2
depicts how higher development rents hasten conversion, while higher agricultural rents,
conversion costs, and the rate of decline in costs defer conversion for one pixel relative to
another.3
INSERT FIGURE 2 HERE
To account for unobserved idiosyncratic factors associated with pixel i at time t. we add
an error term to equation (3) such that the greater it is, ceteris paribus, the less likely is
conversion. If we further specify s* as the amount that makes (3) an equality, then we find the
likelihood of conversion at time t to simply be the cumulative density of e evaluated at e*. In
other words, if the error for pixel i at time t is less than or equal e*, conversion occurs.
(4) A/ - OCit = Ait + (Cit - SCiM) + sit.
We can also depict the relationship between a pixel's development rent and opportunity
cost in a manner that makes explicit the contribution of factors affecting opportunity cost, both
known and unknown. The former comprises those exogenous factors of which the researcher is
aware, including agricultural prices and agronomic characteristics. Assuming for simplicity's
sake that the vector of known supply-side factors, X, is adequately represented by a single
indicator, we can describe a single-pixel analogue of a supply curve for pixels based on a pixel's
opportunity cost CDF in a given period. Depicted in Figure 3, the position of this curve is
determined by X, while the distribution of s determines its general shape.
INSERT FIGURE 3 HERE
At equilibrium, the total amount of conversion (in terms of pixels) that occurs in a market
or region must equal the summation across all pixels of conversion likelihoods:
(5) YJPoc{Dv,X)=TotDevt
i
where
Poc is the cumulative density of operating cost, and
88
-------
TotDevt is the overall areal extent of conversion.
We have shown thus far how the likelihood of conversion is simply determined the
comparison of development returns with opportunity costs. Before moving onto an empirical
model that estimates the likelihood of conversion, however, we must somehow deal not only with
the fact that the likelihood of conversion and development returns are jointly determined, but also
with the absence of price data for precisely the tracts in which we are interested, which precludes
recourse to modeling via simultaneous equations approach. To overcome this problem, we
explicitly consider an individual's residential choice.4
A pertinent abstraction of the individual's site selection process has them essentially
viewing from above the region they plan on living and considering where best to situate their lot.
They behold in their region undeveloped patches of varying levels of appeal (i.e., the patches'
quality varies), each of which a potential location for their new lot. A patch's quality results from
a vector of demand-side factors, denoted V, that individuals deem important, e.g., proximity to
water, proximity to the urban core, the landscape pattern of neighboring land, etc. We assume
that this vector, too, contains one element indicative of overall quality.5 In Figure 4, quality is
portrayed in a hypothetical region by color: the deeper the green, the higher the quality of a patch.
In addition to having discretion over where their new lot will be, these individuals also determine
how big a lot to carve out of the open space. Conceivable plots are portrayed by the rectangular
polygons.
The utility provided by the lot depends on both its size and the quality of the land on
which it resides. Development rents - or per pixel rental prices from the demand-side perspective
- will vary according to quality and the individual's desired lot size for a particular quality level
is determined by the first order condition equating marginal WTP with this price:6
INSERT FIGURE 4 HERE
dS
89
-------
where
WTP is the willingness to pay for a lot, and
S is the size of a lot.
Critically, pixel rental prices across quality levels adjust to ensure that the consumer
surplus the individual garners from a lot is the same regardless of the quality level. Expressed
mathematically, we have
(7) wtp(v,s)-das = CSVV.
It is relatively easy to see how a lot conceived on a relatively unappealing patch could be
larger than that on an amenity-laden patch: the pixel rental price for the former will be low
enough for one to carve out a larger lot size, compensating for the relatively low quality. At
market equilibrium, individuals will be indifferent to all quality-quantity combinations in their
choice set of potential lots. We can see the quality-quantity tradeoff in Figure 4, where the lots in
patches of higher quality are smaller (holding incomes constant).
The relationship between quality and quantity can also be depicted graphically, as in
Figure 5. The solid lines represent equilibrium pixel rental prices and lot sizes at low and high
quality levels; they are demand curves. The dotted line represents pixel rental price and lot size
combinations holding CS constant so the points of intersection illustrate the tradeoff that may
exist in the choice set between quality and quantity at market clearing prices. The areas bounded
by the two solid lines and their respective pixel rental prices (indicated by the points of
intersection) - reflecting surplus - are equal.
INSERT FIGURE 5 HERE
By incorporating the foregoing into Equation 5, we now express the total number of pixels
converted in terms of integration over the joint density of V and X:
(8) £ (£>„, Jf) = / • J J [g(V, X)-P^ (Dr (V, CS\ xtyxdV
i V X
where
90
-------
I is the total number of undeveloped pixels in a region, i.e., areal extent of the region.
g(V.X) is the joint probability density function for VandX
By dividing the likelihood of conversion for a pixel with V and X characteristics by the
equilibrium lot size (itself a function of V and CS) for such a pixel type, we have an expression
equal to the total number of lots sought over the entire market, considered as given:
With the distribution for demand and supply-side factors, along with the number of lots sought
known, this equation condition could be solved for CS.
Stepping back to focus on the landscape change process has involved combining into a
single framework decisions about where and how much to develop. The resulting equilibrium
condition implies that lot size and price information are not required for estimation of pixel
conversion probabilities. Apparent in the numerator of the expression, V and X are the relevant
covariates. The CS solving Equation 9 is actually irrelevant, as it is a market-level value that is
constant across all pixels in a given market and for a given time interval. As such, its effect on
the likelihood of conversion will manifest for all pixels in a constant term or set of fixed effects.
For land-use change and other phenomena, timing is a critical aspect of interest. Given
that conversion is the consequence of continuous processes and may occur at any point in time
during the period under observation, the appropriate means by which to estimate parameters that
affect all observations in a consistent manner is by recourse to duration - or survival - modeling.
Rather than modeling the direct influence of a covariate on conversion probabilities, duration
models are concerned with the hazard rate underlying the probabilities, i.e., the instantaneous risk
that pixel /' is cleared in period t conditional on not having been converted before I.7 While
conventional methods such as linear or logistic regression have been applied in these contexts,
they are ill-equipped to handle the features that often characterize duration data, including time-
varying explanatory variables and censoring or truncation of the dependent variable.8
91
-------
As our study data are interval censored, meaning that each observation's survival time is
known only to fall somewhere between two dates, the dependent variable assumes a value of one
if a conversion occurs over an interval between the dates and zero otherwise. To reconcile the
temporal continuity of the conversion process being modeled with this coarseness in the
measurement of timing, we specify a complementary log-log duration model. By doing so, the
relationship between our V and X covariates and the probability that opportunity costs are low
enough for conversion to occur is
(10) Poc= l-e-\
where
(11) h = ea+py+p!C-.
The complementary log-log model is a discrete analogue to Cox's proportional hazards
model, a highly flexible specification that is estimated using partial likelihood methods. Two
major advantages the models share are that they readily accommodate time-varying covariates
and require no assumptions on the functional form of the baseline hazard rate or on the factors
that may change this rate over time. This enables attention to be focused specifically on the effect
of the covariates on the relative risk of a transition. Additionally, and as the name implies, the
coefficients estimated by these proportional hazards models have a relative risk interpretation.
Unlike the Cox model, the complementary log-log model is estimated using maximum likelihood,
allowing one to readily generate estimates for the effect of time on the odds of a transition (See
Allison, 1995 for further discussion).
Data and Methods
The Dependent Variable
The econometric model presented in this paper is estimated using a time series of five
classified Thematic Mapper (TM) and Landsat Multispectral Scanner (MSS) satellite images over
central North Carolina for the years 1976, 1980, 1986, 1993 and 2001.9 The process of imagery
-------
classification was preceded by the standard pre-processing activities, including geometric
correction, spectral-spatial clustering, and radiometric normalization. Classification then
proceeded according to a hybrid change detection methodology combining radiometric and
categorical change techniques on a pixel-by-pixel basis. This procedure produced four land cover
classes: forest, non-forest vegetation, impervious surface, and water. From these classes, we
generated a binary dependent variable equaling 1 if a conversion from forest or non-forest
vegetation to impervious surface occurred between two dates and 0 otherwise.10 Conversions to
water were treated as censored, while pixels whose classification in the first year (1976) was
either water or impervious surface were eliminated from the data. Transitions between forest and
non-forest vegetation were also treated as censored as these may be attributable more to forest
rotations than permanent conversion from one land cover to another. After overlaying two GIS
layers of tenure data from ESRI (2000) and the North Carolina Department of Parks and
Recreation (2003), those pixels falling under public ownership (e.g. national, state, and municipal
parks) were also eliminated.
Upon classifying the imagery, a systematic sample of pixels was drawn that provided
65,991 pixels for model estimation. The grid pattern across the satellite scene was such that
roughly 1.2 kilometers separated each pixel from their nearest neighbors. Systematic sampling is
a commonly applied technique to handle spatial correlation of unobserved variables that affect the
probability of conversion (Turner, Wear, and Flamm, 1996; Cropper, Puri and Griffiths, 2001;
Kline, Moses and Alig, 2001). The consequences of spatial autocorrelation include inefficient but
asymptotically unbiased estimates. However, in cases in which the unobservable variables are
spatially correlated with the included explanatory variables, the coefficient estimates on the
included variables will additionally be biased (Irwin and Bockstael, 2001). A major source of
spatial autocorrelation arises from multiple observations falling under common landowners
(Kline, Moses and Alig, 2001). Given that the average size of private forest ownership in North
Carolina is 9.7 hectares (Powell et al., 1992), while the average farm size is approximately 75
93
-------
hectares (U.S. Census of Agriculture, 1997), 1.2 kilometer pixel separation in our sample was
deemed an adequate distance to sufficiently reduce the likelihood of this occurring.11'12
The Explanatory Variables
Several static and time-varying covariates are included in the model, the values for which
correspond to the start year of the interval given by the dates of the satellite imagery. The suite of
variables specified captures both site and locational attributes that are hypothesized to affect the
likelihood of land-use conversion. Table 1 presents descriptive statistics and the units of
measurement for each variable.
INSERT TABLE 1 HERE
To capture the influence of what Healy (1985) has termed juxtaposition effects - or
"spatially bounded externalities that affect adjoining or nearby land" (Alig and Healy, 1987: 225)
- we derived four time-varying window-based metrics from the imagery that measure the
landscape configuration surrounding a pixel. The first is the percent of the area within a window
of approximately two square kilometers that is classified as impervious (inner_imperv). The size
of the window is admittedly arbitrary, yet also based both on best professional judgment of a
typical developer's spatial frame of reference and on previous studies that have found window-
sizes of similar magnitude to capture spatial externalities (Geoghegan, Waigner, and Bockstael,
1997; Fleming, 1999; Irwin and Bockstael, 2002). The second metric complements the first, and
is the percent impervious in a region between the aforementioned window and another with sides
double the size of the first (outer_imperv). Thus, the metrics are non-overlapping, with
outer_imperv relating to a region that rings inner_imperv's.
Interestingly, the fact that we explore the potential for spatial externalities using the
amount of development within windows around a pixel, as opposed to a measure of how things
look around the perimeter of the lot within which the developed pixel would reside, is not a cause
for concern. Given that the parcel - like the pixel - must be in an undeveloped state in the data
upon which the metrics are based means that the two calculations lead to the same result. As can
94
-------
be seen in Figure 6, the amount of impervious surface within the windows is the same whether or
not one considers the lot, since it cannot contribute to that amount.
The two additional window metrics are based solely on the smaller window and are the
percent of area classified as water (p_water) and a fragmentation metric (frag). We use a
mu
formulation developed by Frohn (1998), which is defined as Fragit = , where i denotes the
n* X
pixel, t denotes the date of the image, m is the total number of patches in the window, n is the
total number of pixels in the window, and A, is a scaling constant equal to the area of the pixel.13
Because n and A, are constants in our data, the metric essentially reduces to a count of patches.14
Hence, as the landscape becomes more fragmented, frag increases.
In addition to the window-based metrics, time-varying proximity-based metrics are also
included in the specification. The first is the Euclidean distance to the nearest primary road
(road dist).15 The second is the Euclidean distance to the nearest woodchip mill (chipmilldist),
which is a potentially important cost attribute of forestry operations.16 The third proximity metric
(city_dist), a measure of the influence of market proximity, gives the Euclidean distance to the
nearest city with a population of over 50,000 (i.e., Charlotte, Durham, Fayetteville, Greensboro,
Raleigh, and Winston-Salem).
Another five variables are included in the model that do not change with time: elevation
(elev), slope, and dummy variables indicating forested pixels (forest), wetlands (wetland), and
whether either public lands (nearpub) are within a mile of the pixel, or whether hazardous waste
sites are (nearhaz).17
Varying by county and time interval, a returns to agriculture metric is also in the model
and assumed exogenous (ag_returns). This metric is calculated as county total farm receipts less
costs, divided by farm acreage in the county. This metric was associated with even forested
pixels: as we found agricultural returns to exceed forestry returns in all cases, we assumed that
agricultural production to be the relevant alternative use for land vis-a-vis development.
95
-------
Finally, we include a set of county dummies representing the 31 counties in the region, as
well as a dummy for each market—time interval combination (markets in our dataset are assumed
to be the Metropolitan Statistical Areas (MSAs) for Charlotte, Durham, Fayetteville, Greensboro,
Raleigh, and Winston-Salem). The former serve to limit omitted variable effects arising from
county-level differences in governance, zoning, and other factors that may be fixed over time.
The latter are a consequence of the formalization as per the discussion relating to Equation 9.
Each pixel in our sample was assigned to a market based upon the population weighted distance
from each to that pixel.
Results
Table 2 presents results of two complementary log-log models of the determinants of
increases in impervious surface. The second model is distinguished from the first by its inclusion
of the window-based metrics. In both models, the distance measures and the measures of
surrounding impervious surface are transformed as logarithms to allow for attenuated effects of
these variables with increases in their magnitude. Although interpretation of the coefficient
estimates from the complementary log-log model is complicated by the log-odds transformation
of the dependent variable, we can readily calculate their "risk ratio," which also is their marginal
effect. In the case of the linear (logged) continuous covariates, the risk ratio is interpreted as the
percent change in the hazard rate from a unit (percentage) increase in the covariate. These values
are obtained by subtracting one from e1' and multiplying the resulting value by 100 in the case of
the linear covariates, and by one in the case of the logged covariates. For the dichotomous
variables, the percent change in the hazard rate when the variable equals one is again 100 times
ep-l (Allison, 1995).
INSERT TABLE 2 HERE
While Models 1 and 2 are both highly significant, with chi square values of 1846 and
2707, respectively, a likelihood ratio test of the null-restrictions imposed by Model 1 on the
96
-------
effects of the window based metrics suggests that it be rejected in favor of Model 2. The chi
square value of the test is 860 with four degrees of freedom, providing clear-cut evidence that the
metrics improve the fit of the model. As an additional gauge of the predictive performance of the
two models, we calculated Goodman and Kruskal's gamma (Goodman and Kruskal, 1954, 1959
and 1963), a non-parametric, symmetric metric that is based on the difference between
concordant (C) and discordant (D) pairs of predicted and actual values of the dependent variable
as a percentage of all pairs ignoring ties. Gamma is computed as (C - D)/(C + D), and can be
interpreted as the contribution of the independent variables in reducing the errors of predicting the
rank of the dependent variable. The value of gamma calculated from the constrained model is
0.838, while that of the unconstrained model is 0.923. The improvement in the predictive ability
of the model with the inclusion of the window metrics is thus considerable, reducing the fraction
of uncertainty remaining in the constrained model by 52 percent.
With respect to the statistical significance and magnitude of the coefficient estimates on
the window metrics, the strongest result is seen for the inner ring metric, a 1% increase in which
induces a 1.18 percent increase in the hazard of conversion. The coefficient of the outer ring
metric is also positive and significant but of considerably lower magnitude, increasing the hazard
by 0.16 percent. It is notable that Irwin and Bockstael (2002) obtain contrary findings on
similarly constructed variables measuring the percent of developed area in two non-overlapping
rings surrounding a plot. Their study focuses on explaining leap-frog development of land
parcels limited to areas on the urban fringe, and they interpret the negative coefficients as
representing 'repelling effects'. Our attempt to replicate their result by limiting the sample to
pixels located beyond 10, 15, and 20 kilometer gradients of the nearest city of greater the 25,000
found the positive and significant parameter estimate on the outer ring variable to be robust.18
Increases in fragmentation, as measured by frag, decrease the hazard of conversion,
though the estimate is just within the range of significance at the 10% level. Increases in the
percent of water surface area, by contrast, have a positive effect that is just out of the range of
97
-------
significance. The former result may reflect disamenities associated with development on highly
fragmented land immediately surrounding the pixel, while the latter result is a likely consequence
of the positive spillovers generated by hydrological resources for both residential and industrial
uses.
Beyond improving the fit of the model, the inclusion of the window metrics produces
some noteworthy discrepancies with respect to the sign, significance and magnitude of the
remaining covariates. Elevation and the dummy indicating proximity to public lands, both
significant and positive in Model 1, are insignificant in Model 2. The negative and significant
coefficient on distance to road decreases by over threefold in Model 2, while the coefficient on
the variable measuring the distance to the nearest large city reverses its sign from negative to
positive. The former result is consistent with the intuition that decreasing primary road proximity
discourages peripheral location through increases in accessibility costs per kilometer. However,
the latter finding of a positive effect of distance to the nearest city in Model 2 contradicts the
conventional expectation that the value of land in developed use is a positive function of spatial
proximity to city centers. One plausible explanation for this finding is omitted variables bias: It
may be that what is most relevant to development potential is the existence of suitable
infrastructure, something better captured by the percent impervious metrics, than by the proximity
to some city center.
This result also serves to highlight the trade-off between lot size and pixel quality that
underpins the theoretical model outlined above. While decreased pixel quality, as measured here
by decreased proximity to the urban center, is expected to reduce the hazard of conversion, this
effect may be countered by a market equilibrium in which lower quality pixels are compensated
by larger lot sizes. These countervailing effects preclude hypothesizing the sign of the variable a
priori. Thus, Model 2's positive signing of the distance measure could also reflect the dominance
of size effects, which results in larger lots and hence a higher hazard of conversion.
98
-------
A final notable discrepancy between the two models is the sign reversal on the dummy
variable indicating proximity to a hazardous waste site, which is positive in Model 1 and negative
in Model 2. Specifically, the estimate from Model 2 suggests that the hazard of conversion for
pixels located within a mile of a hazardous waste site 82 percent of the hazard for pixels located
beyond a mile of such sites. The counterintuitive sign on this variable in Model 1 likely reflects
an upward bias imparted from the combined influence of the uniformly positive influences of
inner_imperv and outer_imper on the hazard of conversion together with the positive correlations
between these variables and the hazardous waste site dummy (which Spearman's rank correlation
tests support at the 99% level).
The remaining variables across the two models are largely in agreement. While the
coefficients of the 27 county dummies in the model are not shown in the table, using a chi-square
test of their joint significance we cannot reject the hypothesis at the 1% level that all of these
coefficients are zero in both models. Likewise, joint tests of the MSA-year interactions are found
to be statistically significant at the 1% level. The return to agricultural land uses, while having
the expected negative sign, is insignificant in both models. Turning to the pixel attributes, Model
2 indicates that the hazard of conversion for pixels identified as wetlands is 55 percent of the
hazard for those pixels not having this attribute, which is slightly higher than the magnitude
estimated in Model 1. These findings are consistent with the hypothesis of higher conversion and
opportunity costs associated with pixels under mature or ecologically important vegetation. The
two remaining pixel attributes - the forest dummy and slope - are insignificant in both models.
Finally, the negative coefficient on the chipmill_dist variable is noteworthy given a
continuing controversy over the socioeconomic and ecological impacts of satellite chip mills in
the state. Between 1980 and 1998 the number of such mills in this region increased from two to
18, a trend that many perceive as hastening environmental degradation and biodiversity loss
through the promotion of clear-cutting on non-industrial woodlots and monoculture tree farms.19
While not illuminating the question of clear-cutting, the result obtained in Models I and II
99
-------
indicate that closer proximity to chip mills does serve to increase the hazard that land is converted
from forest to urbanized use. The small magnitude of the coefficient estimate, however, suggests
that the economic significance of the mills for conversion may be minimal.
Discussion and Conclusion
This paper began with a theoretical model of land use change that takes into
consideration both the supply and demand sides of the market for undeveloped tracts. One of the
most salient results to emerge from the model is that data on parcel boundaries, lot size, and
prices are not required for the estimation of conversion probabilities, as these factors are absent
from the derived equilibrium condition. While such factors may play a role in land use
conversions, their effects play out at the market level and can hence be captured in the model
through the inclusion of fixed effects and time-market interaction terms. A second important
conclusion is that it is impossible to sign the effects of landscape amenities on the hazard of
conversion; to the extent that disamenities are compensated in equilibrium by larger lot sizes,
they may have the effect of actually increasing the probability that land is converted.
Based on the theoretical framework, the paper then presented an application of a hazard
model as a means of analyzing the effects of static and time-varying socioeconomic and
ecological covariates on the conditional risk that land is converted for developed use. By
specifying the complementary log-log derivation of the proportional hazards model, we employed
a methodology for modeling a continuous time process - the conversion of land to impervious
surface - using discrete time satellite data. Our analysis confirmed several findings uncovered
elsewhere in the literature, including significant impacts of ecological attributes and road
proximity on the likelihood of conversion. As in the works of Geoghegan, Wainger, and
Bockstael (1997) and Irwin and Bockstael (2002), we additionally find support for the hypothesis
that spatial interactions, as measured by the window metrics describing the landscape pattern, are
important determinants of the land conversion process. Unlike Irwin and Bockstael, however, we
100
-------
find no support for repelling effects; contrasting with their study, the two variables employed here
measuring the percent of impervious surface surrounding the pixel both have positive impacts.20
Our result may be attributed to the dispersed pattern of urban development, organized
around mill towns that emerged in North Carolina at the turn of the century. To the extent that a
leap frog pattern of development was already established at this time, subsequent development
occurring at the end of the century may have been driven largely by urbanization economies
arising from city size itself. Our result may also indicate simply that the repelling effect does not
operate at a macro enough scale to be evinced by a covariate like inner_imperv, but rather at finer
resolution that has implications for development patterns in a locale irrespective of its overall
density of development.
There are several possible extensions for using the empirical model estimated in this
paper to explore the issue of urbanization. Among the most promising would involve exploiting
the model's flexibility in incorporating the effects of time on the hazard of conversion (Allison,
1995). Rather than specifying time dummies, as done here, this could involve including a trend
variable measuring the time elapsed since some starting date of interest, such as a change in
tenure or the transfer of land ownership (see e.g. Vance and Geoghegan, 2002). Such an
approach would enable experimentation with different functional forms of the baseline hazard,
including the inclusion of squared and higher order trend terms to allow for non-linearities in the
hazard rate, and would provide a basis for simulating future landscape patterns under alternative
policy scenarios.
101
-------
Figure 1: The study region boundaries and physio-geographic zones of North Carolina
Source: Adapted from T.E. Stear, "Population Distribution,"pp.30-51, in North Carolina's
Changing Population (University of North Carolina, Carolina Population Center, 1973).
102
-------
Figure 2: Time paths for development rent and opportunity costs
103
-------
Figure 3: Given the equilibrium price of pixels of quality, V, the likelihood of conversion
depends on supply side characteristics
104
-------
Figure 4: Landscape mosaic illustrating patches of varying quality and potential lot choices
105
-------
Figure 5:
All lots at equilibrium, varying by size and quality, provide the same consumer surplus
106
-------
Figure 6: Equivalence between % impervious for a given pixel and an associated lot
107
-------
Table 1: Descriptive statistics
Variable name
Units Mean
Standard deviation
dep. var. (l=conversion)
0,1
0.01
0.10
forest
0,1
0.63
0.48
wetland
0,1
0.12
0.32
slope
degrees
0.60
1.19
elev
meters
137.18
65.81
ag_returns
$1000/acre
0.10
0.12
chipmilldist
kilometers
63.74
44.56
pubdum
0,1
0.09
0.28
city_dist
kilometers
42.35
20.07
roaddist
kilometers
1.41
1.32
nearhaz
0,1
0.05
0.21
inner imperv
percent
1.69
6.26
outer_imperv
percent
2.02
6.47
percent_water
percent
0.51
3.05
index
9.60
6.40
108
-------
Table 2: Complementary log-log model of the hazard of conversion to impervious surface
Model I
Model II
Coef. est.
% Chg
Coef. est.
% Chg
forest
0.035
(0.856)
3.562
-0.282
(0.140)
-24.573
wetland
-0.716
(0.000)
-51.130
-0.595
(0.003)
-44.844
slope
0.031
(0.453)
3.149
0.374
(0.386)
45.354
elev
0.006
(0.001)
0.602
-0.001
(0.738)
-0.100
ag_returns
-1.371
(0.200)
-74.615
-1.635
(0.122)
-81.269
chipmilldist
-0.007
(0.000)
-0.007
-0.005
(0.005)
-0.005
nearpub
0.612
(0.000)
84.412
0.006
(0.956)
0.602
city_dist
-0.560
(0.000)
-0.429
0.188
(0.020)
0.207
roaddist
-0.650
(0.000)
-0.478
-0.158
(0.002)
-0.146
nearhaz
0.610
(0.000)
84.043
-0.200
(0.053)
-18.127
inner imperv
0.770
(0.000)
1.159
outer_imperv
0.105
(0.051)
0.111
pcntwater
0.027
(0.116)
2.737
frag
-0.010
(0.095)
-0.981
intercept
-4.854
(0.000)
-5.520
(0.000)
chi2 county dummies (27)
138
(0.000)
65
(0.000)
chi2 MSA-year interactions (23)
196
(0.000)
185
(0.000)
LR chi2 (60, 64)
1846
(0.000)
2707
(0.000)
n obs
65991
65991
p-values in parentheses
-------
References
Alig, R. and R. Healy. 1987. "Urban and Built-Up Area Changes in the United States: An
Empirical Investigation of Determinants," Land Economics 63(3): 215-226.
Allison, P. D., 1995. Survival Analysis Using the SAS System: A Practical Guide. Carey: SAS
Institute Inc.
Bobyarchick, A.R. and J.A. Diener. 2000. "Land Regions and Geology," in: Orr, D.M. Jr. And
A.W. Stuart (eds.), The North Carolina Atlas: Portrait for a New Century. Chapel Hill: The
University of North Carolina Press.
Bollen, K.A. 1996. "An Alternative Two Stage Least Squares (2SLS) Estimator for Latent
Variable Equations," Psychometrika 61: 109-121.
Boscolo, M., S. Kerr, A. Pfaff, and A. Sanchez. 1998. "What Role for Tropical Forests in
Climate Change Mitigation? The Case of Costa Rica," Paper prepared for the HIID/INCAE/BCIE
Central America Project.
Brown, M.J. 1993. North Carolina's Forests, 1990. USDA Forest Service Technical Report.
Asheville, N.C.: USFS.
Brownlow, B., N. Lineback, J. DeHart. 2000. "Forest Sector Strategies," in Lineback, N. (ed.),
North Carolina's Sensible Greenhouse Gas Reduction Strategies. Report of the Department of
Geography and Planning, Appalachian State University.
Capozza, D. R., and R. W. Helsley. 1990. "The Fundamental of Land Price and Urban Growth,"
Journal of Urban Economics 26: 295-306.
Cooper, A.W. 2000. "Vegetation," in: Orr, D.M. Jr. And A.W. Stuart (eds.), The North Carolina
Atlas: Portrait for a New Century. Chapel Hill: The University of North Carolina Press.
Cropper, M., J. Puri and C. Griffiths. 2001. "Predicting the Location of Deforestation," Land
Economics 77: 172-186.
Environmental Systems Research Institute, Inc. (ESRI). 2000. U.S. National Atlas Federal and
Indian Land Areas, ESRI Data & Maps 2000 (CD 2).
Environmental Systems Research Institute, Inc. (ESRI). 2000. U.S. GDT Park Landmarks, ESRI
Data & Maps 2000 (CD 3).
Environmental Systems Research Institute, Inc. (ESRI). 2000. Census 2000 TIGER/ Line Data,
http://www.esri.com/data/download/census200Q tigerline/. retrieved March 2003.
Ewing, R., R. Pendall and D. Chen. 2003. "Measuring Sprawl and Its Impact,"
htto://www.smartgrowthamerica.com/sprawlindex/MeasuringSprawl.PDF. retrieved February
2003.
Fleming, M. 1999. "Growth Controls and Fragment Suburban Development: The Effect on Land
Values," Geographic Information Sciences 5(2): 154-162.
110
-------
Freddie Mac. CMHPI Data, http://www.freddiemac.com/finance/cmhpi/#ne w. retrieved January
2003.
Frohn, R.C. 1998. Remote Sensing for landscape Ecology: New Metric Indicators for Monitoring,
Modeling, and Assessment of Ecosystems. Boca Raton: CRC Press LLC.
Fulton, W., R. Pendall, M. Nguyen and A. Harrison. 2001. "Who Sprawls Most? How Growth
Patterns Differ Across the U.S." Center on Urban & Metropolitan Policy, The Brookings
Institution.
Gallagher, R. 1968. Information Theory and Reliable Communication. New York: John Wiley &
Sons.
Geoghegan, J., S. Cortina Villar, P. Klepeis, P. Macario Mendoza, Y. Ogneva-Himmelberger, R.
R. Chowdhury, B. L. Turner, II, and C. Vance. 2001. "Modeling Tropical Deforestation in the
Southern Yucatan Peninsular Region: Comparing Survey and Satellite Data." Agriculture,
Ecosystems, and Environment 84(1-3): 25-46, 2001.
Geoghegan, J., L.A. Wainger, andN.E. Bockstael. 1997. "Spatial landscape indices in ahedonic
framework: an ecological economics analysis using GIS," Ecological Economics 23: 251-264.
Goodman, L. and W.H. Kruskal. 1954. "Measures for association for cross-classification I,"
Journal of the American Statistical Association 49: 732-764.
Goodman, L. and W.H. Kruskal. 1959. "Measures for association for cross-classification II,"
Journal of the American Statistical Association 54: 123-163.
Goodman, L. and W.H. Kruskal. 1963. "Measures for association for cross-classification III,"
Journal of the American Statistical Association 58: 310-364.
Hardie, I.W., P. Parks, P. Gottlieb, D. Wear. 2000. "Responsiveness of Rural and Urban Land
Uses to Land Rent Determinants in the U.S. South," Land Economics 76(4): 659-673.
Hauser, J.R. 1978. "Testing the Accuracy, Usefulness, and Significance of Probabilistic Choice
Models: An Information-Theoretic Approach," Operations Research 26(3): 406-421.
Haynes, K.E. and A.S. Fotheringham. 1984. Gravity and Spatial Interaction Models. Beverly
Hills: Sage Publications.
Healy, R.G. 1985. "Population Growth in the U.S. South: Implications for Agricultural and
Forestry Land Supply," Rural Development Perspectives 2(1): 27-30.
Hosmer, D. and S. Lemeshow. 1999. Applied Survival Analysis: Regression Modeling of Time to
Event Data. New York: John Wiley and Sons, Inc.
Ingalls, G.L. 2000. "Urbanization," in: Orr, D.M. Jr. And A.W. Stuart (eds.), The North Carolina
Atlas: Portrait for a New Century. Chapel Hill: The University of North Carolina Press.
Irwin, E.G. and N.E. Bockstael. 2001. "The Problem of Identifying Land Use Spillovers:
Measuring the Effects of Open Space on Residential Property Values," American Journal of
Agricultural Economics 83: 699-705.
Ill
-------
Irwin, E.G. and N.E. Bockstael. 2002. "Interacting Agents, Spatial Externalities, and the
Endogenous Evolution of Residential Land Use Pattern." Journal of Economic Geography 2: 31-
54.
Jargowsky, P.A. 2001. "Sprawl, Concentration of Poverty, and Urban Inequality," in: Squires, G.
(ed.), Urban Sprawl: Causes, Consequences & Policy Responses. Washington D.C.: Urban
Institute Press.
Kline, J.D., A. Moses, and R.J. Alig. 2001. "Integrating Urbanization into Landscape-level
Ecological Assessments," Ecosystems 4: 3-18.
Kunstler, J.H. 1994. The Geography of Nowhere : The Rise and Decline of America's Man-Made
Landscape. New York: Touchstone Books.
Lee, M. 1998. "Off-farm Labor Supply and Various Related Aspects of Resource Allocation by
Agricultural Households," Ph.D. diss.Falkultat fur Agrarwissenschaften der Georg-August-
Universitat Gottingen,.
Lichtenberg, E. 1989. "Land quality, irrigation development, and cropping patterns in the
northern High Plains," American Journal of Agricultural Economics 71: 187-201.
Lilly, J.P. 1998. "North Carolina Agricultural History," North Carolina Department of
Agricultural and Consumer Services, http://www.ncagr.com/stats/historv/historv.htm. retrieved
February 2003.
North Carolina Division of Parks & Recreation, Resource Management Program,
http://www.ncsparks.net.
North Carolina Corporate Geographic Database Data Layers.
http://cgia.cgia.state.nc.us/cgdb/catalog.html
Orr, D.M. Jr. And A.W. Stuart (eds.). 2000. The North Carolina Atlas: Portrait for a New
Century. Chapel Hill: The University of North Carolina Press.
Plantinga, A.J., T. Mauldin, and D.J. Miller. 1999. "An Econometric Analysis of the Costs of
Sequestering Carbon in Forests," American Journal of Agricultural Economics 81: 812-824.
Powell, D.S., J.L. Taulkner, D.R. Darr, Z. Zhu, D. MacCleery. 1992. "Forest Resources of the
United States," USDA Forest Service, Gen. Technical Report RM-234.
Prestemon, J.P. and R.C. Abt. 2002. "Chapter 13: Timber Products Supply and Demand," The
Southern Forest Resource Assessment. USDA Forest Service.
Prestemon, J., J. Pye, D. Butry, and D. Stratton. Economic Research Unit, USDA Forest Service.
"Locations of Southern Wood Chip Mills for 2000,"
http://www.srs.fs.usda.gov/econ/data/mills/chip2000.htm. retrieved April 2003.
Saloner, G. and A. Shepard. 1995. "Adoption of Technologies with Network Effects: An
Empirical Examination of the Adoption of Automated Teller Machines," Rand Journal of
Economics 26(3): 479-501.
112
-------
Schaberg, R., F.W. Cubbage, and D.D. Richter. 2000. "Trends in North Carolina Timber
Product Outputs, and the Prevalence of Wood Chip Mills," paper prepared for the study on:
Economic and Ecological Impacts Associated with Wood Chip Production in North Carolina.
Shi, Yue Jin, T. T. Phipps and D. Colyer. 1997. "Agricultural Land Values under Urbanizing
Influences," Land Economics 73(1): 90-100.
Sierra Club. 1998. "1998 Sierra Club Sprawl Report: 30 Most Sprawl-Threatened Cities,"
http://www.sierraclub.org/sprawl/report98/ralcigh.asp. retrieved January 2003.
Stavins, R.N. and A.B. Jaffe. 1990. "Unintended Impacts of Public Investments on Private
Decisions: The Depletion of Forested Wetlands," American Economic Review 80: 337-352.
Stear, T.E. 1973. "Population Distribution," in: North Carolina's Changing Population
University of North Carolina, Carolina Population Center.
Thiel, H. 1967. Economics and Information Theory. North-Holland: Amsterdam.
Turner, M.G., D.N. Wear and R.O. Flamm. 1996. "Land Ownership and Land-Cover Change in
the Southern Appalachian Highlands and the Olympic Peninsula," Ecological Applications 6(4):
1150-1172.
U.S. Census of Agriculture. 1997.
USDA National Resources Conservation Service. Land Capability Classes,
http: //soils .usda. gov/soil survey/main .htm, retrieved January 2003.
U.S. Department of Commerce, Bureau of the Census, County and City Data Books 1982, 1987,
1992,1994,2000.
USGS. EROS Data Center. National Land Cover Data 1992,
http://edc.usgs.gov/products/landcover/nlcd.html. retrieved January 2003.
Vance, C. and J. Geoghegan. 2002. "Temporal and spatial modelling of tropical deforestation: A
survival analysis linking satellite and household survey data," Agricultural Economics 27: 317-
332.
Wear, D.N. and P. Bolstad. 1998. "Land-use Changes in Southern Appalachian Landscapes:
Spatial Analysis and Forecast Evaluation," Ecosystems 1: 575-594.
Wear, D.N. and J.G. Greis. 2002. The Southern Forest Resource Assessment Summary Report.
USDA Forest Service Southern Research Station.
113
-------
1 Irwin and Bockstael focus on which open-space tracts will be subdivided. In the sense that these tracts
can be of any size over five acres, their model does not predict the magnitude of conversion.
2 If non-decreasing rents over time seems a doubtful assumption, a weaker one can be employed to an
equivalent effect, namely that the change in the present value of a one-time return (selling, rather than
A
renting, the pixel) may be employed to similar effect, i.e., 2 A,8" -<5 z DUS
t-T+1
>0 Vr>r.
3
Note, too, that this formulation ensures that the change in returns due to conversion exceeds total costs.
This is evident if we simplify Equation (3) and examine the extreme case in which both C and net returns
are constant from time T. Equation (3) becomes ^ ^ '/\ — 5 ~ ^' '"uslral'n8 how discounted net
returns must exceed conversion costs.
4 Our focus exclusively on residential development is based on the assumption the majority of transitions to
industrial or commercial uses occur on already developed lands rather than on the undeveloped tracts that
comprise our data.
5 In actuality, the vectors X and V may overlap in terms of the metrics they include.
6 While we have assumed homogenous preferences for this exposition, an analogous result is attained when
preferences vary within a region due to, say, the income distribution.
7 It bears pointing out that the hazard rate itself is not a probability, but rather a measurement of the number
of events per unit interval of time, where an event is defined as some discrete transition across states.
8 Truncation and censoring are pervasive features of duration data, resulting respectively from the data
selection process inherent in the study design or from observation-specific random features that make
observations on survival time incomplete (Hosmer and Lemeshow, 1999).
9 The images are taken from the northern half of path 16, row 36 and the southern half of path 16, row 35 of
the Landsat satellite orbit. Data for the years 1976 and 1980 were derived from the MSS imaging system,
while the TM imaging system was the data source for the years 1986, 1993, and 2001. Because TM and
MSS data have different spatial resolutions - 58 X 79 meters for MSS and 30X30 meters for TM - the
data was spatially degraded to a 60 X 60 meter resolution for consistency.
10 Impervious surface includes paved surfaces, structures, and medium to high-density residential areas.
11 We also experimented with samples having 2.4 km separation between the pixels (n= 15,623) and
obtained similar model results with respect to the statistical significance and magnitude of the coefficient
estimates.
12 Another problem that models such as ours face is that of endogeneity bias: since it works in both
directions, the influence on a pixel's land cover of adjacent pixels' land covers leads to association among
the error terms. We circumvent this problem in two ways: through our modeling of the hazard of
conversion to an impervious surface, rather than simply the likelihood a pixel covered by impervious
surfaces, and through judicious use of systematic sampling.
13 Frohn (1998) suggests that unlike conventional measures of fragmentation, his metric allows
comparisons of landscape fragmentation across images having different spatial resolutions, raster
orientations, and numbers of land cover classes.
14 The metric does not assume only integer values because of the GIS algorithm used to calculate it.
15 Dis road is based primarily on the road network available from ESRI, but was modified using image
interpretation of Landsat data to reflect the conditions existing at the beginning of each interval.
16 The distance to the nearest chipmill was obtained by overlaying a GIS layer of woodchip mill locations
and their establishment dates that is available from Prestemon, Pye, Butry, and Stratton (2003) of the
Economic Research Unit of the USDA's Forest Service. To limit the effect of this variable on forested
pixels, we interact it with the forest dummy.
17 The measures of elevation, slope and the forest dummy were derived directly from the satellite imagery.
Soil quality data was taken from the Land Capability Classes of the USD A Soil Conservation Service,
which indicates the soil's suitability for agriculture. The wetland category was derived from the 1992 land
use and land cover data from the EROS Data Center of the USGS. Data on the location of public lands
were derived from the above referenced shapefiles produced by ESRI and the North Carolina Department
114
-------
of Parks and Recreation. The hazardous waste site data was obtained from the North Carolina Corporate
Geographic Database Data Layers.
18 As Irwin and Bockstael point out (and Geoghegan, Waigner and Bockstael confirm empirically), the
direction of landscape pattern effects may vary over different window sizes, a possibility that data
constraints precluded us from pursuing.
19 These concerns prompted Governor Hunt to commission a study by Schaberg, Cubbage and Richter.
(2000) on the ecological and economic impacts of the mills. Although the report found that the mills
increase the incentive to clear cut and raised the possibility of increased forest fragmentation and truck
traffic in areas around the mills, it stated that the mills are not expected to significantly shorten timber
rotations barring changes in the historical structure of timber product prices (p. v).
20 A possible explanation for this discrepancy is that our results suffer from positive biases imparted by
omitted variables. As Irwin and Bockstael note, such biases may emerge from positive spatial
autocorrelation among unobserved factors such as topography, school quality, and tax policy. Indeed, they
assert that because the net effect of this bias is positive, the estimated effect of their impervious surface
measures, which they refer to as the interaction effects, will bind the true effect from above. They use this
reasoning as an identification strategy, arguing that if "the estimated effect is negative, then it must hold
that the 'true' interaction effect is negative for at least some range of the sample and over some interval of
time (p. 43)."
One weakness with this reasoning is that it rests entirely upon positive spatial autocorrelation
among unobserved factors, which is argued to necessarily impart an upward bias on variables that are
included in the model. There is no justification for this expectation, even on net. The direction of the bias
from an omitted variable, x, will be largely determined by two factors, the sign of its correlation with the
included variable, and the sign of the coefficient estimate of x upon including it in the model. The direction
of the overall bias will depend on the combined influence of all relevant variables omitted from the model.
In fact, several of the omitted variables that Irwin and Bockstael themselves cite as important (p. 42-43)
may very well impart a bias opposite to the direction required for their identification strategy to be valid.
Hazardous waste sites are an example: It is plausible, and confirmed in the present study, that the
correlation between the incidence of hazardous waste sites and the percent of impervious surface is
positive, while the effect of hazardous waste sites on the likelihood of conversion for residential
development is negative. It can be readily demonstrated empirically that the bias imparted on the
interaction term from omitting the hazardous waste site is negative, thereby undermining the identification
strategy that motivates their approach. We thus do not share Irwin and Bockstael's confidence that a
negative effect on the interaction terms ensures that the true effect is negative over some range. Omitted
variables could impart either a negative or positive bias, and it is not possible to identify a priori the overall
direction of this bias with any certitude.
115
-------
Discussant Comments on Presentations in "Conservation and Urban Growth."
Sabrina Lovell, PhD. US EPA, National Center for Environmental Economics.
October 27, 2004.
Good afternoon. First, I'd like to say thank you to our three presenters for attending
today's workshop and sharing the results of their research with us. My goal today is to discuss
the policy relevance of the three presentations in this session and the general importance of
research on land use and habitat conservation.
The research by Iovanna and Vance focuses on the factors influencing transition from
rural to urban uses, by developing a model that identifies the likelihood of conversion from forest
and agricultural lands to developed uses. According to the Census Bureau, urban areas in the US
more than doubled from 1960 to 1990, and grew at about 1 million acres per year during that
time. Although developed land only accounts for about 3% of all US land area, and therefore, is
still a small proportion of our land base, the patterns and rates of urban growth can have negative
consequences (Heimlich and Anderson, 2001). These include disruption of ecological processes,
loss of habitat, higher costs of community services, and a loss of rural amenities such as open
space and scenic beauty. As a result of these potential negative impacts, it is important that
society understands the factors leading to urbanization and particularly their interactions. The
research by Iovanna and Vance specifically accounts for landscape pattern and changes in that
pattern over time and space, and gives us a better understanding of how interactions in these
patterns can influence the amount and location of development. Their model uses data over 2.5
decades and thereby accounts for long-term changes in land use. This is important because land
use transitions are often slow and cumulative. As a result of these subtle changes, the real
underlying patterns of change are often not apparent when looking at only one or a few years of
data as are used in other land use research papers. I found their use of satellite pixel, rather than
parcel level data, to provide an alternative way to analyze land use when parcel level data is
unavailable. However, I wonder whether or not satellite data at the pixel level is really a viable
and cost-effective alternative for most researchers. If not, I would like to pose the question
whether or not the model could be adjusted to use datasets of land cover and land use that are at a
116
-------
larger scale than either the pixel or parcel level. I would also suggest that they include more
discussion about the pros and cons of parcel versus pixel level data.
In terms of the policy relevance of this type of conversion model, the results
could be used to simulate future development patterns based on different policy scenarios, such
as building of new roads, and then the results could be linked to other environmental modeling
efforts. The first example that comes to mind is the impacts of different land use patterns on air
emissions. EPA has put out guidance on how States can use land use strategies and development
decisions to help meet their air quality planning requirements, and how to account for air quality
impacts of such decisions and strategies. A model such as the one developed by Iovanna and
Vance could be used to predict different development patterns and then based on such patterns,
air emissions models could generate potential air quality impacts.
Another way in which I see this type of model providing a contribution is to an
Alternative Futures Analysis (AFA). AFA is an environmental assessment approach that
provides a suite of alternative scenarios for the future land use in an area, as developed by
multiple concerned stakeholders. These scenarios are expressed as maps of future land use
patterns in an area, and then the potential effects of alternative scenarios on such things as
wildlife populations, water supply and quality, open space, and agriculture are assessed. It
provides communities with a clear picture of the consequences of many different potential
development decisions. EPA has sponsored a few such exercises, including one in the
Willamette River watershed in Oregon. Typically, however, these analyses use very simplistic
and deterministic economic factors to predict future growth if any are used at all. The relevance
of the hazard model developed by Iovanna and Vance is that it accounts for multiple factors
driving conversion to urban uses, in a more rigorous economic framework. Such a model would
be better able to capture the complexity associated with actual urbanization and could improve
AFA modeling.
Whereas the Iovanna and Vance research looks at the factors influencing conversion to
developed uses, the research by Albers and Ando looks instead at the issue of land preservation.
Land preservation is an important public issue as evidenced by the fact that 76 % of 801 non-
117
-------
federal ballot measures to protect land have passed in the last 6 years and generated $24 billion
in funds (Land Trust Alliance, 2004). The research presented today seeks to understand the
nature of interactions between private land trusts and government agencies and the impacts on
the environment based on different levels of cooperation. In their paper, they model the
government's conservation actions as exogenous to the land trust. However, it is often the case
that the government works together with a land trust to preserve an area, rather than each acting
independently as in the current model. For example, the former Mt Tom Ski area in western
Massachusetts was preserved in 2002 by a partnership between a land trust, two government
agencies, and another charity (Land Trust Alliance, 2004). One suggestion I have for expansion
of the model is to explicitly account for this potential scenario and to compare the results with
those where the government acts independently. Another suggestion is to explicitly incorporate
the actions and responses of developers into their model. The actions of developers influence
what land is in the most danger of being converted, and is often one of the criteria land trusts use
for choosing which parcels to protect. Finally, I think the authors should consider how both the
pattern and amount of land available for protection is affected when land owners donate lands.
In this case, the land trust is not choosing which lands to purchase, as in their model, but only
whether or not to accept the donation.
Like the other two papers in this session, there is a spatial dimension to the model that
provides for a more realistic analysis. In particular, I liked the way the authors incorporated the
idea that benefits depend on the spatial pattern of all lands protected and that benefits may not be
a linear function of amount conserved. This acknowledgement of the ecological implications
and interconnections of different preserved areas is increasing in policy work related to habitat
preservation, and I'm glad to see it accounted for in this model. One of the potential policy
applications I can see arising out of this research is that the model could be used for analyzing
the benefits of preserving a network of avian reserves. Cooperation among states, and between
the US, Canada, and Central and South American countries for setting aside migratory bird
habitat would be an excellent application of this type of model.
118
-------
The paper by Bauer et al. on different potential patterns of vernal pool preservation
relative to residential development provides an excellent example of an integrated ecological-
economic model. As we heard yesterday, EPA's Ecological Benefits Assessment Strategic Plan
(EBASP) calls for more research that integrates the two disciplines. I also found the particular
application used here to have significant policy relevance in that it focused on amphibians.
Across the globe, amphibians are experiencing a precipitous decline. In the United States, there
are 292 species, 21 of which are endangered or threatened (USFWS, 2004). Responding to this
concern, the federal government has a new national program to research and monitor the state of
amphibians, and devotes $4 million a year to identify threats to amphibians nationwide. Habitat
destruction poses one of the biggest challenges for their conservation, and research such as that
presented here can help us understand how best to address that challenge. In addition, while in
discussions to develop EPA's EBASP, amphibians where often overlooked when it came to both
ecological risk assessment and economic valuation and this study provides a first step to
correcting that gap.
The importance of accounting for species dispersal and connectivity of habitat for species
preservation as discussed in this paper is a good example of the point made in the Albers and
Ando paper that benefits of preservation depend on the pattern and total amount of land
preserved. The use of different types of habitat by amphibians, here wetlands and upland areas,
is just one example of the importance of preserving a variety habitats for different life-stages or
activities of wildlife. Another application of this model could be for preservation of habitat for
songbirds that often use different habitats for feeding, nesting, and migrating. Habitat
fragmentation, particularly for birds dependent on forest for their primary habitat, is thought to
be a big reason for the declines we are now seeing in many songbird species. One question that
arises in transferring this type of model to another species, is the data requirements - how
detailed does the information have to be to widely apply a model such as this to different areas or
species?
119
-------
The three presentations we just heard offer different but complementary approaches to
understanding the complex issues involved with land use and land conservation. Land use
patterns and development from rural to urban uses are indeed complex and the impacts of land
use decisions have wide spread and important consequences for both people and the
environment. The research just presented describing the factors influencing land use change, the
impact of interactions between different conservation organizations on conservation outcomes,
and the potential impacts on ecological resources of different patterns of development, is
therefore critical to helping understand the complex tradeoffs between the many alternative land
use choices.
References:
Heimlich, R.E. and W.D. Anderson. Development at the Urban Fringe and Beyond. US
Department of Agriculture. Agricultural Economic Report Number 803. June 2001.
Land Trust Alliance, http://www.lta.org
US Fish and Wildlife Service. http://endangered.fws.gOv/wildlife.html#Species
120
-------
Discussant Comments: "Conservation and Urban Growth: Finding the Balance"
Andrew J. Plantinga
Associate Professor
Department of Agricultural and Resource Economics
Oregon State University
Corvallis, OR 97331
plantinga@oregonstate.edu
The first two papers, by Bauer et al. and Albers and Ando, explicitly consider the
designation of land for conservation uses. The Bauer et al. paper focuses on designating habitat
for amphibians and places particular emphasis on modeling ecological relationships. In this
paper, the decision-maker is a social planner. In contrast, Albers and Ando consider the
provision of conservation land by the private sector. The paper focuses on modeling the strategic
behavior of land trusts and exploring their interaction with a government conservation agency.
The last paper, by Iovanna and Vance, presents an econometric analysis of private land
development decisions. The focus of the paper is on estimation with high resolution spatial data
and measuring the effects of spatial measures of development patterns on land conversion
decisions. To some degree, all of the papers present preliminary work. Therefore, my comments
will focus on the methodologies employed, rather than on specific results.
The paper by Bauer et al. has two parts. The first presents a land allocation model
constrained to protect amphibian populations from adverse effects of development. The second
is concerned with identifying a reserve network to minimize the probability of extinction subject
to a budget constraint on the total foregone value of developed land. This research has several
strengths. It integrates ecological and economic modeling in logical ways. The ecology
component is a finite-patch metapopulation model, which accounts for the spatial arrangement of
patches. This seems to be an appropriate framework to study amphibians. Finally, the authors
draw on data from an actual landscape to develop the simulation and optimization models.
121
-------
In the first part of the paper, the ecological model includes land development as a barrier
to dispersal between any two patches. The more development that occurs between two patches,
the smaller is the contribution of those patches to the long-term persistence of the species. A
refinement of the model could allow for different types of development. More intensive
developed uses, such as roads, are likely to be a greater barrier to dispersal than less intensive
development, such as low-density residential housing. This may allow the authors to consider a
larger, and more realistic, set of policies. Rather than only evaluating preservation policies, they
may be able to consider policies that allow only certain types of development. Zoning policies
that restrict intensive development, but allow residential housing, for example, may be relevant
alternatives.
Another refinement should address an inconsistency between the initial landscape used in
the simulations and the method by which land values are measured. Data on land values and lot
sizes are assembled from local tax assessments. A simple regression model is estimated and
used to derive the per-acre value of land for use in the simulations. These data on land values are
conditioned on the current landscape, reflecting such factors as the existing road network.
However, in the simulations, the initial landscape is assumed to be completely undeveloped. The
first step in addressing this inconsistency might be to re-estimate the land value model with
distance to roads as a regressor and to include the current road network in the initial landscape.
This should help to make the cost estimates presented in the results section more meaningful.
However, it raises other challenges, such as needing to account for the barriers to dispersal posed
by the existing road network.
The authors compare the costs of current policies to a reference case corresponding to the
optimal solution. The current policies are representative of existing policy approaches. The
122
-------
reference case assumes that the planner operates free of transactions or information costs. For a
given metapopulation size, Figure 3 shows the divergence between the costs of the current
policies and the reference case. Future work might investigate the costs of implementing each of
these policies, accounting for transactions, measurement, and monitoring costs. This would
permit an evaluation of whether the optimal solution is truly the least-cost approach. Because
implementation costs are likely to be highest with the reference case, current policies may be
more efficient. Plantinga and Ahn (2002) and Antle et al. (2003) analyze implementation costs
for land-use and carbon sequestration policies.
In the second part of the paper, the authors evaluate three methods for conserving habitat
for amphibians: conserving ponds, conserving corridors between ponds, and conserving both
ponds and corridors. The authors should clarify what the conservation of corridors between
ponds entails. First, if the land between ponds is already developed, then in many cases it will be
infeasible to acquire and restore the land for use as habitat. Consider the cases of land currently
used for roads or residential housing. Second, if the land is not developed, then why is
conservation necessary? Might the land remain in the undeveloped state? If so, should the
opportunity costs of conservation be zero?
The paper by Albers and Ando presents theoretical and empirical analyses of private land
trust decisions. This paper addresses the interesting and policy-relevant issue of coordination
between private conservation groups and public sector conservation agencies. The theoretical
model involves two land trusts (one of which could be the government) deciding which parcels
to conserve. The set-up is seven parcels arranged in a line, land trusts with inter-related benefit
functions, spatial externalities (negative or positive) between adjacent parcels, and competition
for parcels in the manner of Cournot duopolists or a Stackleberg leader and follower. The
123
-------
decision by the authors to specify a discrete state space (land parcels) means that only numerical
solutions can be found. In my view, it would be preferable to specify a simpler model with a
continuous state space that permits analytical solutions.1 The structure of the land trust model
has analogues in the industrial organization literature. For example, product differentiation
models involve Cournot competition over where firms locate in product space. This competition
gives rise to location externalities similar to the spatial externality in the present model. Other
relevant models include those concerned with network externalities and product space location.
One technical point is that early models in the industrial organization literature represented
locations along a line, but it was found that the existence of equilibria is influenced by the
endpoints. Circle models (e.g., Salop, 1979) were developed to avoid this problem.
The structure of the theoretical model requires that land trusts conserve parcels in the
same geographic area. An empirically-relevant alternative is for land trusts to operate
independently. According to the Land Trust Alliance, local and regional land trusts conserve 6.2
million acres of land in the U.S., which represents less than 0.5% of the U.S. private rural land
base. The authors might define their benchmark case as one in which the land trusts do not
compete. This raises the issue of how to specify the benefit functions when land trusts operate
independently. If the benefits from conservation are pure public goods, then the functions might
still be inter-related, though the spatial externalities would not be present.
In the theoretical model, land prices are assumed to be exogenous, suggesting that land is
competitively supplied. However, if a set of parcels provides unique conservation values, then
landowners may be in a position to extract the rents that, in the current formulation of the model,
1 If simulations are used, my view is that the results would be more interesting and relevant if the authors use an
actual landscape, as in the Bauer et al. paper, or at least a more realistic landscape.
124
-------
accrue to the land trusts. That is, one can envision agents having market power on the supply
side of the market, as well as on the demand side.
The second part of the Albers and Ando paper presents empirical models of land trust
acquisition in Massachusetts and Illinois. The dependent variable is the number of privately
protected acres in a township and regressors include the area of protected land in neighboring
townships, the area of publicly protected land, and sociodemographic and physiographic
characteristics of townships. Because townships differ in size, the variables measured in acre
units should be expressed as shares (i.e., normalized on total township area). A more
fundamental issue is whether the dependent variable should be defined in terms of townships. If
the purpose of the empirical exercise is to test the implications of the theoretical model, in which
the decision-making unit is the land trust, then ideally the dependent variable would be defined
in the same way. This point is obviously moot if there is a one-to-one correspondence between
townships and land trusts. If this is not the case,2 there may be problems with interpreting the
results. The authors find that that the acres of privately protected land in neighboring townships
have a positive effect on the dependent variable. This is not necessarily evidence that land trusts
are coordinating their activities. Township boundaries may simply cut across the holdings of a
single land trust.
This result that land trust acres are affected by neighboring land trust acres has another
plausible interpretation. Presumably, land trusts will want to preserve lands that have not been
developed for urban uses or, in some cases, agricultural uses. We might then ask: why have
these lands remained undeveloped? One answer is that these lands were never profitable to
develop. Characteristics of such lands include inaccessibility or topography that limit the
2 The Nature Conservancy, which is an international land trust, protects 76,000 and 22,000 acres of land in Illinois
and Massachusetts, respectively.
125
-------
productivity of the land for agriculture. Lands with these characteristics are often clustered
spatially. For example, lands unproductive for agriculture are often found in upland or
unglaciated areas. The spatial pattern of federal lands in the western U.S. corresponds strongly
to the distribution of unproductive lands.
A final point relates to the exogeneity of certain regressors. It is not difficult to imagine
that population, the number of endangered species, and the cost of land could be endogenous to
the amount of privately protected land. The authors should consider testing for the exogeneity of
these variables (e.g., Hausman 1978).
The paper by Iovanna and Vance begins with a theoretical model of private land-use
decisions that informs the development of an empirical model of land development. My general
comment on the theoretical model is that it should be worked out in full so that all of the
derivations and assumptions are transparent. Currently, the model is a mix of equations, graphs,
and verbal arguments. A main purpose of the theoretical model is to justify replacing
development rents, which the authors do not observe at the pixel level, with site-specific
characteristics. The authors invoke an equilibrium argument: in a spatial market equilibrium,
prices for developed land will have adjusted so that individuals are indifferent to the available set
of residential lots. Because in equilibrium the demand for lots equals the supply of lots, the
equilibrium price for developed land can be expressed as a reduced-form function of exogenous
demand- and supply-side factors, which include site-specific attributes.
This spatial equilibrium framework underlies hedonic property value studies and urban
spatial models. Consider, for example, the closed-city model of Capozza and Helsley (1989).
The equilibrium price of developed land is given by equation 17. As shown, it depends on
exogenous model parameters, the distance of a particular parcel to the CBD, z, and the location
126
-------
of the city boundary relative to the central business district (CBD), z(t) ,3 Thus, the equilibrium
price depends on a single site-specific attribute, z, and on time. The dependence on time comes
through the exogenous population level, N(t), which affects the position of the city boundary
according to equation 11. If the population level is constant, then the boundary is fixed and the
city does not grow. If the population level increases, then the city grows as development rents at
the boundary of the city rise above the exogenous agricultural rent (equation 14 and section IV).
Two points deserve emphasis. First, if population is constant, the equilibrium price of developed
land varies spatially according to a single site-specific attribute (location), and is constant over
time. Second, in order for land development to occur in equilibrium, there must be a change in
an exogenous factor such as population.
Iovanna and Vance also model equilibrium land development. The question raised by the
foregoing discussion is: what exogenous factors are changing that result in land development?
Many of the variables in the empirical model are constant over time. Variables that are not
constant may not change enough to explain a large portion of the development (e.g., agricultural
returns, chipmill distance) or are not exogenous from a regional perspective (i.e., spatial
variables that are a function of development within the region). The authors should work to
reconcile the empirical model with an equilibrium model of land development. One approach
would be to specify the empirical model in terms of changes in site-specific attributes, rather
than start-of-period levels.
The authors estimate their model with high-resolution (60 meter) pixel data. Modeling
land-use decisions with these data poses a number of challenges, not least of which is managing
such as a large amount of information. The authors are to be commended for going beyond
3 Note that future increases in development rents (the last term in equation 17) are a function of the city boundary
location, J(t), by equation 13.
Ill
-------
descriptive analysis and taking on the difficult task of estimation. In this regard, the chief
advantage of these data—relative to plot data or aggregate data—is that it allows spatial
processes to be modelled explicitly. The authors include three spatial metrics to explain land
development: two measure the density of development around each pixel (innerimperv,
outerimperv) and one measures the fragmentation of development (contagion).
The results indicate that the spatial metrics are important determinants of development.
However, it is difficult to know how to interpret their effects. First, all impervious surfaces are
classified as developed land. As such, the data does not distinguish between development for
residential housing and development for commercial or industrial uses. One can imagine that
residential development decisions are affected by development patterns, but why would it affect
other types of development? Even if most development is for residential uses, the same value of
a spatial metric could indicate very different types of development. For example, fragmented
development could correspond to an appealing wealthy neighborhood with low housing
densities, or to unappealing low-density commercial development. Second, the development
density measures may just be picking up effects of omitted variables. Many factors that explain
past development decisions are likely to explain current development decisions. For example,
the extent of the transportation network (e.g., road density, not just the distance to the nearest
road) is likely to matter. Because no road density variable is included, its influence may be
coming through the development density variable. Third, and related to the second point, the
density measures cause identification problems for variables that do not change over time. For
example, suppose the elevation of a pixel is the same (or similar to) the elevation of surrounding
128
-------
pixels included in the density metrics. Then, the effect of elevation is not identified because it
affects past and current development.4
For modeling land-use decisions, plot and aggregate data have several advantages relative
to pixel data. At the current time, these data provide much broader coverage. For example, the
National Resources Inventory (NRI) contains 1 million plots spread across the continental U.S.
Because these data are collected through on-the-ground inventories, the NRI provides a wealth of
plot information (e.g., detailed land use, land quality). The broader geographic coverage is
advantageous because it allows the modeler to take advantage of spatial variation in economic
variables. For example, economic returns to forestry vary spatially because climatic differences
affect which tree species are dominant. One of the important uses of land-use models is for
policy analysis. Particularly for economists, it is valuable to include economic policy levers like
net returns in the model so that market-based incentives can be analyzed (see, for example,
Plantinga and Ahn 2002).
References
Antle, J., S. Capalbo, S. Mooney, E. Elliott, and K. Paustian. 2003. Spatial Heterogeneity,
Contract Design, and the Efficiency of Carbon Sequestration Policies for Agriculture.
Journal of Environmental Economics and Management 46(2):231-50.
Capozza, D.R., and R.W. Helsley. 1989. The Fundamentals of Land Prices and Urban Growth.
Journal of Urban Economics 26:295-306.
Hausman, J.A. 1978. Specificiation Tests in Econometrics. Econometrica 46:1251-72.
Plantinga, A. J., and S. Ahn. 2002. Efficient Policies for Environmental Protection: An
Econometric Analysis of Incentives for Land Conversion and Retention. Journal of
Agricultural and Resource Economics 27(1): 128-45.
4 Consider what happens to the elevation variable as we move from Model I to Model II. Its coefficient changes
from being highly significant to being highly insignificant.
129
-------
Salop, S.C. 1979. Monopolistic Competition with Outside Goods. The Bell Journal of
Economics 10:141-56.
130
-------
Summary of the Q&A Discussion Following Session V
Nancy Bockstael (University of Maryland)
Directing her comment to Rich Iovanna, Dr. Bockstael asked for clarification as to how
he measured his dependent variable.
Richard Iovanna (U.S. EPA, NCEE)
Mr. Iovanna responded, "It's a binary variable that's "1" if over an interval of time, say
1976 tol980, a pixel went from being . . . primarily agricultural uses to an impervious
surface."
Nancy Bockstael
Dr. Bockstael questioned Mr. Iovanna further, asking if the variable was related to "some
percentage of impervious surface" or whether there was just "one measure per pixel—and
that's either impervious or not."
Richard Iovanna
Mr. Iovanna answered, "It's the latter. At this level of resolution, it's quite a research
challenge to go beyond whether or not that's clearly classified as one or the other."
Nancy Bockstael
Dr. Bockstael continued with "just a few comments, and you'll probably be able to
resolve these, but my experience dealing with parcel-level data and struggling with it
leads me to ask. First of all, I think that the size of development is considerably larger
than one of your pixels, if I'm doing my multiplication and division right, at least in
Maryland. Since there are economies of scale in development, it seems to me that you
always have difficulties if you don't have observations at the decision level. What looks
like a surrounding land use effect may really be just the effect of the same decision
because the decision is overlapping your unit of observation—so that the pixel next door
is getting developed either concurrently or in the next wave of development. So, I'm
always reluctant to interpret the neighboring land use having an effect on the current
pixel if it's based on pixel-level rather than parcel-level data."
She continued by questioning the resolution capability of the LANDSAT data that remote
sensing experts use, stating, "We actually had a project with NASA to try to figure out
whether the remote sensing people could pick up low-density residential use—lots that
are 2 acres or more. Something like 80 or 90 percent of the land that's been developed in
the last 10 years has been 2-acre or larger lots. They can't do it. . . . and we had very
accurate data on certain places in Maryland about exactly what's going on. We even
matched it out against the actual houses from tax maps, and they basically gave up on
that part of the project. So, despite what remote sensing people will say about some of
the LANDSAT data, I would ask them whether or not they're sure that they're picking up
low-density residential, which is such an important part of development these days. Also,
I would question whether you could go back in time and get anything like that very
accurately, because the accuracy of LANDSAT data has been increasing over the last 25
131
-------
years. I wonder, also, whether there are situations in which you might actually be
"picking up" the resolution of stuff that happened a long time ago, but you're just seeing
it now because of the changing technology in LANDSAT data."
Dr. Bockstael concluded by adding "One other thing: You mentioned that you sort of
assume this relationship between size of the parcel and value. However, if you introduce
regulations like minimum lot size, which is a common zoning regulation in Maryland . . .
Trying to keep developers from developing, we make the minimum lot size larger and
larger, and if you do that, then it introduces distortion into that relationship. Since I
haven't seen the data, I'm not absolutely certain how you would introduce the typical
land-use regulations into your model. Over the 25 years, I don't see anything [in your
model] that would reflect any changes in land-use regulations over that time, and I would
expect that there would have been. So, if there's no place in which to pull strings on
regulation, you may not be able to answer the policy questions that you'd like."
Richard Iovanna
Mr. Iovanna responded by saying, "With regard to your last question, I think you're
absolutely right—at this stage of the game, we intended to pick up a lot of those
influences simply with the fixed effects, and we are planning to revisit that issue as soon
as we can find the data. What I find particularly interesting is the impact of minimum lot
size and how that could be, if possible, incorporated into our model, which is right now at
a pretty high level of abstraction."
He continued, "With regard to the low-density residential, you're absolutely right—that's
been an issue sitting in the back of our minds since we received the data from the
contractors, who initially assured us that things could be sliced ten different ways. When
we finally received the product, they sheepishly admitted that with low-density
residential, given the fact that there are laws and the fact that much of it will be under tree
cover, it often looks like forest."
David Martin (Davidson College)
Addressing Heidi Albers and Amy Ando, Dr. Martin said, "With respect to Andrew's
(Plantinga) comment about land trusts competing, they also compete with foundation
grants. So, it's not just land prices ... I was on the board of directors of a land trust, and
it wasn't always clear to me whether we were better off cooperating with another land
trust or competing for a particular grant—there's that budget constraint issue as well."
Speaking to "both Stephen (Swallow) and Amy (Ando)," Dr. Martin stated, "My other
career is as a local elected official, so I'm wondering about your welfare functions for a
decision maker. For example, I care about frogs and affordable housing. From my
perspective, how do we achieve that balance of getting frogs and affordable housing? I
guess I'm seeing in your model that if the decision maker likes frogs, this is the best thing
to do, but what happens if the decision maker likes more than one thing that affect the
conservation? So, if I like affordable housing, I may like to have more houses, higher
132
-------
density, and what not. In your case, Amy, when I buy public land, what kind of uses do I
use that land for—do I put in soccer fields or hiking trails? Or do I put soccer fields in
the middle and have the edges for the hiking trail and then let the land trust conserve the
land right next to the hiking trails? What I'm encouraging is a richer consideration in
terms of the decision maker."
Dr. Martin closed with a personal anecdote, saying, "I found it easier to get re-elected
because people will have moved to my community because they liked high taxes and
open space, and they moved elsewhere if they liked asphalt and low taxes. In the
dynamics of this, it was easier for me to support environmentally friendly things."
Stephen Swallow (University of Rhode Island)
Dr. Swallow responded that the issue of affordable housing is "kind of on my radar
screen. My understanding is that certainly at state levels, and, I believe at the federal
level also, the requirement, so to speak, for affordable housing trumps everything. In
fact, in Massachusetts and Rhode Island I've been hearing environmental horror stories . .
. where a developer can come in, and if he's having trouble getting a proposal through"
for a development involving $250,000 houses, all he has to do is say that he'll do
affordable housing and "everything's out the window. There are situations developing, at
least in Massachusetts, where those buildings are going next to the wetlands, and they're
nicely mosquito-infested" with all the attendant disease possibilities. Dr. Swallow closed
by saying, "It's certainly beyond what I'm getting to, but I wanted to take the time to say
that I think there's some bad policy there."
Amy Ando (University of Illinois at Urbana-Champaign)
Dr. Ando replied, "Our model doesn't have endogenous budgets. There are some in the
literature where fundraising is endogenous and land trusts might differentiate each other
in order to facilitate fundraising. So, if we take Andrew's (Plantinga) suggestion of
developing an integral model, we might go that route. . . . We certainly do not have a
government objective function that is as complicated as the one you just described—
government has an arbitrarily chosen objective function. If you model the social planner,
the social planner maximizes total net benefits (total benefits minus total costs)—it's
pretty simple. I have a feeling that making it more complicated will have to wait until the
next round of research."
Stephen Swallow
Dr. Swallow added, "Let me be a little less dismissive: Clearly, in principle, you could
add another constraint and deal with that, among some other things like that."
John Tschirhart (University of Wyoming)
Dr. Tschirhart addressed Amy Ando, saying, "You mentioned that there are conditions
under which hot spots may not be the best places to preserve—I was wondering what
those conditions might be. Also, you've worked in the past on explaining government
decisions about endangered species and so forth." With all the elections around the
133
-------
country, Dr. Tschirhart said he wondered what the characteristics are of the local
governments at the county and city level who are deciding on passing taxes in support of
environmental/ecological programs.
Addressing Stephen Swallow, Dr. Tschirhart asked, "Who owns these vernal pools, and
if they're preserved would there be public access or are they so small that no one really
wants to visit them anyway?" He closed by saying, "Also, I was wondering if you could
explain to me why the toads on my property are declining precipitously over the last ten
years."
Amy Ando
Dr. Ando said, "You know, it's funny—I do a lot of work on political economy and the
behavior of government agents and here I am cranking out papers in which our
government doesn't behave in anything like an interesting or realistic manner. . . . Thank
you, John, for reminding me of that stuff—it's certainly something that it would be good
to be thinking about more actively as we proceed. Right now we don't have plans to
model government behavior in a more realistic way, but maybe we should."
Heidi Albers (Oregon State University)
Responding with what she classified as a "quick answer on that hot spot thing," Dr.
Albers said, "We're just starting to develop that, but one example where the government
might leave the hot spot unprotected is when the land trust perhaps has a lower marginal
value for protection. So, if the government comes in and protects the hot spot, then they
[the land trust] might not protect anything, but if the government comes in and protects
something else instead, then the marginal value of protecting the hot spot is still high
enough for them to come in—so you get a larger amount of area, overall, conserved,
including the hot spot."
Stephen Swallow
Dr. Swallow commented, "I think what may not have come through in the presentation is
that we're using the amphibians as an indicator of trying to keep a functioning ecosystem
across the landscape. If we can keep the amphibians functioning, then there's hopefully
going to be room for many other species, including birds. What I had in mind with
preservation, if you link this with the land trust unit. . . Rhode Island has more municipal
land trusts, which are town agencies, than probably any other state in the country despite
the fact that we are only slightly larger than Yosemite. If the land trusts are going to own
the land, they're probably going to be involved with providing some public access when
it can be managed in conjunction with ecological attributes."
He continued, "I want to point out one thing related to Amy's talk: The grant application
process that many of these land trusts face I think actually causes the land trusts to alter
their objectives. So, a lot of times, you might be picking up the objectives of the funding
agency."
Susan Durden (U.S. Army Corps of Engineers, Institute of Water Resources)
134
-------
Addressing a brief comment to Amy Ando, Ms. Durden stated, "I'm pretty familiar with
townships in Illinois, and I think that Massachusetts townships are probably more
analogous to counties in Illinois because at the [Illinois] township government level,
there's not a lot of activity other than clearing the snow off of the roads."
Continuing with a second comment directed at Dr. Ando, Ms. Durden said, "This was
bothering me a little bit in some of the papers, and you said something that reminded me
of it: When you were talking about the econometric models (and I'm a great fan of using
real logic and looking at the outputs and seeing if they make sense), you mentioned that
the government investment went up and the private investment in conserving land went
up, and just as an aside you said, "well, there were some problems with the econometrics
as if the observation didn't make any sense. I may have misunderstood that, because I
certainly didn't take in the whole paper, but it concerns me that in some cases we may
have an idea that this or that would make sense or would not make sense. I would think
particularly in Illinois there would be a lot of reasons that might explain what you
observed—the younger farmers who are taking over from older farmers might be more
inclined to convert lands, or people contributing money could indicate simply that the
economy is good. Again, I may not have gotten it right, but I think it's important to
realize that those two could move together rather than necessarily moving in opposite
directions."
Amy Ando
Responding first to the township comment, Dr. Ando said, "Yes, they are totally
different. We used them because they are a nice size. We wanted a spatial unit of
observation that wasn't too big and that wasn't too small. In California, we ended up
using quadrangles because California doesn't have townships and their counties are too
big. But, you're absolutely right—they are not functionally similar" and we will try hard
to emphasize that in Massachusetts the townships merely provided convenient boundaries
for our study. Ms. Ando also said that they also benefited from the "wonderful
coincidence" that in Massachusetts, voting data are also essentially gathered at the
township level.
Dr. Ando presented the following clarification to address the second point that was made:
"The comment that I made about a paper that had analyzed potential crowding out having
some econometric problems—the particular issue is that they didn't have spatial data on
the locations of private protected areas. All they knew was how much land was protected
by different land trusts but not where. Since many land trusts operate in multiple
counties, these authors were struggling with this econometric difficulty of how to cope
with the lack of spatial data in their data set. They have a panel and we don't, so in that
sense their work has an advantage over ours. I actually think that in a year or so, there
will be two papers in the literature, ours and theirs, . . . approaching a similar question
from very different points of view with very different data, and it will be very
interesting."
135
-------
Dr. Ando closed by stating, "We're not making any assumptions about whether public
and private protected areas are likely to be positively related or correlated—we'll let the
data tell us what they say."
Nancy Bockstael
Saying that her comment was actually a follow-up on this last question, Dr. Bockstael
stated that". . . preservation decisions are simple sorts of things, and we've been dealing
with this as a supply and demand issue, basically. The landowner has to decide his
reservation price, so there's this wide range in this market. Now, if one's willingness to
sell development rights, which is what we're talking about, is affected at all by how many
people around them are also selling development rights, then it seems to me that that side
of the model has to be dealt with if you're going to deduce anything from the results and
looking at outcomes. One of the problems with the literature in this area is that we
analyze the outcomes, and half the papers analyze the outcomes as though it's the result
of people's willingness to preserve their land, and the other half analyze the output as
though it's the agency's decisions as to what to purchase, when in fact it's the interaction
of the two. We have found, at least in the agricultural preservation area, a landowner's
reservation price is definitely affected by how many people around him are willing to
preserve—the property preserved is a lot more valuable if other people are preserving,
whether it's agriculture or if it's for a state kind of effect. If the development happens all
around you, you wish you really hadn't sold at the development price because the value
of the property isn't as high—and I'm saying that in analyzing the output, you'd have to
separate that out in order to make deductions about the interactions with the government
policy."
Amy Ando
Dr. Ando replied, "Thank you—that's a very good observation, and I'm sitting here
thinking about scale. The story I was just telling was a story at the landowner level or
parcel level, . . . and you would end up with clusters because of that just because
reservation prices depend on what's going on around you, so either you end up with
everybody selling or nobody. I don't know whether a similar story translates when your
unit of observation is a town—is much larger . . ."
Nancy Bockstael
Dr. Bockstael interrupted, saying, "Well, I'm not sure how big the townships are—if
nothing else, it will induce spatial autocorrelation, because anybody around you
obviously is affected by people around them—just something to think about."
Kerry Smith (North Carolina State University)
Dr. Smith commented, "I just want to put in a plug for Andy's (Plantinga) comment
about structural modeling. There's a fellow by the name of Randy Walsh at Colorado
who has been using this in assorted models to look at an interaction between public
136
-------
choices and private choices where there may be substitutes and complements. So, you
locate government space or you protect areas, and the private undeveloped land may
actually become developed in communities as a consequence of those decisions, so it's
not just the interaction between land trusts—it's the interaction between the land trusts
and other private decisions that influences the outcomes. Now, he's [Walsh] been able to
solve that in a framework that allows you to look at Nash equilibrium and a variety of
other things. Now, you could take that and extend it a little bit, using some of the median
voter models from public economics and think about the decision process of local
governments or another kind of framework that would describe land trusts' decisions. It
would require a structural model for each of the agents participating in the model, in the
supposed market, and what it would do is give you the ability to look more specifically at
how those decisions influence prices. There are certainly easier models than the one he's
working with—there's stuff in the public economics literature using some of the random
utility models that we heard about for recreation yesterday. So, it would be worth
looking at."
END OF SESSION V Q&A
137
-------
Valuation of Ecological Benefits: Improving the Science
Behind Policy Decisions
PROCEEDINGS OF
SESSION VI: METHODOLOGICAL ADVANCES IN STATED PREFERENCE
VALUATION
A WORKSHOP SPONSORED BY THE U.S. ENVIRONMENTAL PROTECTION
AGENCY'S NATIONAL CENTER FOR ENVIRONMENTAL ECONOMICS (NCEE)
AND NATIONAL CENTER FOR ENVIRONMENTAL RESEARCH (NCER)
October 26-27, 2004
Wyndham Washington Hotel
Washington, DC
Prepared by Alpha-Gamma Technologies, Inc.
4700 Falls of Neuse Road, Suite 350, Raleigh, NC 27609
-------
ACKNOWLEDGEMENTS
This report has been prepared by Alpha-Gamma Technologies, Inc. with funding from
the National Center for Environmental Economics (NCEE). Alpha-Gamma wishes to
thank NCEE's Cynthia Morgan and the Project Officer, Ronald Wiley, for their guidance
and assistance throughout this project.
DISCLAIMER
These proceedings are being distributed in the interest of increasing public understanding
and knowledge of the issues discussed at the workshop and have been prepared
independently of the workshop. Although the proceedings have been funded in part by
the United States Environmental Protection Agency under Contract No. 68-W-01-055 to
Alpha-Gamma Technologies, Inc., the contents of this document may not necessarily
reflect the views of the Agency and no official endorsement should be inferred.
11
-------
TABLE OF CONTENTS
Session VI: Methodological Advances in Stated Preference Valuation
Embedding in the Stated-Preference Methods
Michael Hanemann, University of California - Berkeley; Jeff Lazo,
National Center for Atmospheric Research 1
Experimental Tests of Provisions Rules in Conjoint Analysis for
Environmental Valuation
Laura Taylor, Georgia State University; Kevin Boyle, University of
Maine; Mark Morrison, Charles Sturt University 6
Internet-Based Stated Choice Experiments in Ecosystem Mitigation:
Methods to Control Decision Heuristics and Biases
John P. Hoehn, Frank Lupi, Michael D. Kaplowitz, Michigan State
University 8
Discussant
Joseph Cooper, USD A, Economic Research Service 44
Discussant
Thomas H. Stevens, University of Massachusetts - Amherst 51
Summary of Q&A Discussion Following Session VI 57
in
-------
Embedding in the Stated-Preference Methods
By
Michael Hanemann, University of California, Berkeley
Jeff Lazo, National Center for Atmospheric Research
(Summary of Dr. Hanemann's presentation)
Dr. Hanemann began by stating that the reason he and his colleague focused their
attention on embedding is that he believes "embedding is one of the most serious issues
that remains not satisfactorily resolved in contingent evaluation, and it's a major focus of
the critics." He continued, "This research wants to try and come to grips with what is
causing embedding, whether these causes operate in market valuation as opposed to non-
market valuation, whether they occur in stated preference based on conjoint analysis, and
lastly, what can be done to deal with embedding in stated preference generally."
Stating that different terminologies are often used, Dr. Hanemann went on to clarify the
meaning of the term "embedding" as he uses it. He stated that embedding involves three
elements: (1) insensitivity to scope, so that a larger item is not valued more than a smaller
item; (2) sub-additivity, meaning that the value of a set of items is less than the sum of
the values of the items individually, and (3) order effects—the order in which an item is
valued affects its value. He said that he believes "one can write down utility functions
which explain all three effects in terms of diminishing marginal rate of substitution,
income effects, and substitution effects."
Referring to a formula with variables representing public goods and income, Dr.
Hanemann said he thinks "something like this can represent mental accounting; that is,
mental accounting can be expressed as a form of utility function. He continued, "What I
want to stress here is that I don't think the economic structure of preferences is all that's
going on with embedding, and I want to focus on other features, such as features of the
questionnaire and features of the elicitation format, but also, more basically, features of
how people think and make judgments about items." Dr. Hanemann explained that the
methodology of the research is to replicate some of the existing studies in the literature,
using the same sort of setting and the same survey mode while at the same time adding
features to the survey which are designed to explore some of the hypotheses that focus on
the three items he identified.
Starting off by looking at scope effects, Dr. Hanemann asked, "Why might somebody
give you the same willingness to pay for a larger item as for a smaller item?" He
suggested five possible explanations: (1) The survey is flawed and doesn't really capture
what the person feels. (2) The person doesn't see the larger item as any better than the
smaller item. (3) The person feels that if he pays for the smaller item, he actually gets the
effect of the larger item anyways, so there's no point in offering more money. (4) The
person feels that the larger item isn't feasible, and therefore pointless. (5) The person
thinks that the larger item actually only costs the same. He stated, "To explore these
explanations, one needs to incorporate what are called manipulation checks, that is,
1
-------
questions in the survey or "think-alouds" or verbal protocols to get at what the respondent
was thinking of. We focus on monetary valuation and get anomalous results, and we feel
that's because of a flawed elicitation of monetary valuation. However, it may not be
flawed—it may simply be that these other things are going on, but we don't normally
look for them. So, the focus of the research and the replication that I'm conducting is to
investigate these other explanations. Of 60 or 80 scope studies, there are only 4 or 5 that
I've seen that do this. A very nice recent one is coming out shortly in the Journal of
Environmental Economics and Management by Heberlein, Bishop, and Schaefer. They
express that conventional economists look at economic scope, what they call an affective
scope and cognitive scope, and they get at these things by asking a series of questions—
this is what I was referring to as manipulation checks. For instance, in a question to
evaluate wolves, they ask, "How important are wolves to you, personally? -Not at all
important—Somewhat important—Etc." Or: "In valuing a population of 800 wolves
versus a population of 300 wolves in northern Wisconsin, how would you rate a
population of 800 wolves? -Extremely bad—Somewhat bad—Bad—Neither bad nor
good—Etc."
Dr. Hanemann revealed that "what they find when they use these manipulation checks is
that they line up with the monetary valuation. When respondents like the larger item
more, they give it a higher value. Sometimes the respondents like the smaller item more,
and then they value that item accordingly compared to the larger item." Providing
another example, which he said "is not widely reported in the study that Bill Desvouges,
Kevin Boyle, et al. did on birds" Dr. Hanemann said that there was actually a
manipulation check in the survey. He stated, "Remember, the focus was on covering
waste oil holding ponds on the flyway to protect birds from being killed—2,000 birds—
20,000 birds—200,000 birds. The researchers posed the question: Covering waste oil
ponds will not significantly affect populations—Strongly agree?—Agree?—Neither
agree nor disagree?—Etc. There was the same sentiment. That is, most people felt that
this didn't make a big difference. So, it's not surprising to me that they then found no
difference in monetary value between those items."
Moving on to the issue of feasibility, Dr. Hanemann cited Baruch Fischoff s paper in
which he looked at willingness to pay for pollution cleanup along variable segments of
the Susquehanna River. Dr. Hanemann focused on some debriefing questions that were
not used in the study's data analysis but that were reported. Fischoff found that in a post-
survey phone interview people remembered poorly how many miles of the river were to
be cleaned up, but the people who thought there were more miles had a higher
willingness to pay. Dr. Hanemann summarized, "In other words, some of the noise in the
willingness to pay responses seems to correlate with noise in what the size of the
commodity was. Also, a significant fraction of people didn't think that a thousand miles
could be cleaned up, and again that appears to have influenced their responses."
Dr. Hanemann went on to note that "most of these studies use the open-ended format—
how much are you willing to pay?—and that introduces additional complications because
in addition to valuing the item, people don't want to pay more than their fair share and
they don't want to pay more than the item costs. In this context, one issue is maybe a
2
-------
larger item will involve more people, and the cost per household for the larger item may
not be any greater than the cost per household for the smaller item." He cited some
evidence from a phone survey he had done in Ohio regarding river cleanup in which he
found that willingness to pay is correlated with cost, and cost doesn't vary with scope.
He said, "What we are finding is that things that ought to correlate, rationally, with the
willingness to pay also don't vary with scope. So, the focus of this research is to
replicate some of the scope studies but to add these questions to see if this holds in some
of those studies."
Saying he wanted to relate all of this to the literature on market research, Dr. Hanemann
cited "a very interesting series of papers by Chris Hsee at Chicago, who has worked on
what he calls joint versus separate evaluation." He said that in implementing the strategy,
you first describe a market goods item to someone and then ask, "How much would you
pay for this?" Another group of people is asked the same question regarding a different
item. A third group is then asked to evaluate the two items together, so you achieve both
separate and joint evaluation. Dr. Hanemann said that Hsee's premise is that "assessing
an item in isolation is more difficult than assessing two or more items together, and
because of this difficulty, people adopt different response or judgment strategies in
assessing a single item in separate evaluation than in joint evaluation." He stated that
Hsee frames his comparison "in terms of evaluability: When assessing an item in
isolation, the judgment is influenced more by attributes that are easy to evaluate, even if
they are less important than other attributes which are hard to evaluate." However, when
people assess two or more items together, it is easier to compare the attributes—one
against the other—and more weight is placed on the more important attributes. Through
your choice of things, you can therefore switch the ranking of the choice of items.
To illustrate the point, Dr. Hanemann cited one of Hsee's examples concerning the
purchase of a music dictionary as a gift for a friend. Given the choice between a
dictionary with 20,000 entries that has a torn cover and a dictionary in perfect condition
but with only 10,000 entries, he revealed that in isolation people chose the smaller
dictionary more often, but when they used joint evaluation people chose the larger
dictionary more often and tended to overlook the blemish. Dr. Hanemann said that
studies from the environmental literature show the same thing. He went on to reiterate
the widely recognized "greater difficulty of doing separate evaluation as opposed to joint
evaluation."
Dr. Hanemann said Hsee also points out "the link to another concept in psychological
theory called norm theory: when evaluating an item in a separate evaluation, people
think about the larger category to which they think the item belongs and then they
compare it to the norm for that category. In joint evaluation they compare the two items
as opposed to comparing each item with the norm from an imagined category." He said
he stresses this point because he thinks that "a// cognition is relative, not absolute, and
the norm theory suggests that if I don't give you a standard of comparison but instead just
ask you to evaluate a single item, you invent a standard of comparison." This ends up
being "more noisy" because it's not controlled by the researcher.
3
-------
"Here are the implications. I see this as essentially the same phenomenon as what some
have termed coherent arbitrariness. As I said, cognition is relative to something—to a
norm—and that something affects the evaluation. It makes evaluation in isolation noisy,
or arbitrary, but I think it's arbitrary within some range." Dr. Hanemann said the
important thing, he believes, is that "this applies not just to the monetary evaluation
(expressed willingness to pay) but to all dimensions of liking and judgment for an item.
So, again, the aim is to test this by replicating some studies. I also think mathematically
you would say that separate evaluation involves an element of noise, and then there's less
noise or less uncertainty when a second item is considered, and so one could write down
some mathematical formalisms."
Dr. Hanemann continued, "This also relates to some recent work by Ian Bateman that
was published this year looking at order effects—looking at the evaluation of multiple
items and comparing what he calls step-wise versus advanced-disclosure designs." He
explained, "With step-wise you get to see one item and you are asked to evaluate that.
Then you get to see another item, a subset or something—then you get to see another
item—but each time you value an item before knowing what else is coming. With full
disclosure, on the other hand, you're shown everything—it's laid out before any of the
valuation questions are asked." Dr. Hanemann stated, "What Ian (Bateman) finds is that
the order effects appear very regularly with the step-wise design and go away with the
advanced disclosure. How does this relate? I think this is like the distinction between
separate evaluation and joint evaluation—it's the same type of phenomenon. It's not
limited in any way to non-market goods, and it's not a feature of monetary evaluation as
opposed to other dimensions of evaluation, and I think it moves things around much more
than the economic formalisms of income effects and substitution effects."
"So, what I'm doing is replicating Ian Bateman's work, comparing step-wise with
advanced disclosure, but also measuring not just monetary evaluation but other
dimensions of liking and valuation for the goods in non-monetary terms and seeing if
they have order effects in one treatment and not in the other treatment."
"If there is a difference between separate evaluation, thinking of items in isolation, versus
thinking of them together, the question arises: Which is better?" Dr. Hanemann stated
that "the NOAA panel argued strongly for separate evaluation for external tests of scope,
not internal tests of scope." He added, "One reading of Hsee's work is that the external
tests of scope, the separate, are much more noisy. They're rooted less securely. In a
sense, they go against how human cognition works, and in fact, what happens is a person
invents something with which to contrast or compare the item being evaluated."
Dr. Hanemann noted that "Hsee says in a recent paper that this also explains the
difference that's been observed between predicted utility and experienced utility, because
when I ask you to predict your behavior or your choice, that's like a joint evaluation
because you imagine several outcomes and you compare them. But what actually
happens in life is you choose one of them (or one of them gets chosen) and you
experience that—you decide to move to the West Coast, for example—and then three
years later you're asked how did you like it. That's more like separate evaluation; you
4
-------
experienced one thing and you didn't experience anything else, and he argues that
separate evaluation may be more realistic for assessing actual experienced utility."
Dr. Hanemann continued, "My inclination is to prefer joint evaluation, but—this is the
point I want to emphasize—it seems to me that joint evaluation is susceptible to the same
underlying forces. Now, I just said that in joint evaluation you have contrast, but the
contrast is influenced by the particular items that are involved in that contrast. If you had
other items the assessment might have come out differently, and there is in fact a large
literature in market research on, for example, the number of alternatives, the variety of
alternatives, the number of different attributes or dimensions on each alternative, a range
of values." He went on to offer this example: "Joel (Huber) and a colleague did a
beautiful work on decoy effects in asymmetric dominance. You choose between A and
B. Then I add an item C, which is actually dominated by A or B but shifts your choice
between A or B because it makes one of the items look better. Why? Because you're
evaluating attributes of a particular item relative to the range of attributes in the choice
set. It's the same sort of cognitive imperatives potentially affecting joint evaluation, and
Ian (Bateman) has a CV study showing the presence of decoy effects in multi-item
evaluation."
In closing, Dr. Hanemann said, "I want to end with two points. First, one of the
criticisms of non-market valuation relative to market valuation is that you tell me you
would pay so much for this particular item, but there are other items out there—there are
other brands, etc. How much would you pay for the other items and does it add up? A
major difference with public goods is there's not a set of other items. If I ask you to
value a particular flavor and brand of yogurt, we both know you can walk down the road
to the supermarket where there's a whole shelf of other brands of yogurt. If I ask you to
value a program to protect frogs in a particular part of Rhode Island—well, it's true that
you could have programs to protect other creatures in other parts of New England, but the
point is there's no reason to believe that anybody's planning to do anything about other
creatures in Massachusetts or any place else—it's not obvious that these other public
goods are out there. ... So, one issue is what people think are the other items when you
ask them about a public good. With market goods, you know what's in the
supermarket—you don't need to ask them."
"I'm anticipating possible conclusions, but I haven't reached them. The surveys that
we're doing now are meant to add features which test the speculations I've told you
about. If they come out, this might be an assessment of embedding effects—that in some
sense, joint evaluation with advanced disclosure is preferred, and that means that order
effects can be controlled, and I think the scoping sensitivity gets controlled also and
becomes less of an issue. The remaining issue, which is incorrigible, is that in any sort of
joint evaluation the results are sensitive to the set of designs, and we need to think of
ways of standardizing this or controlling this—we can't escape it, but we could perhaps
agree on a protocol for doing this so we bring this effect under control."
"Thank you."
5
-------
EPA Agreement Number: R830820
Title: Experimental Tests of Provisions Rules in Conjoint Analysis for Environmental
Valuation
Investigators: Laura Taylor, Mark Morrison and Kevin Boyle
Institution: Georgia State University
In recent years there has been a movement away from using contingent valuation to
estimate non-use values towards using various forms of conjoint analysis. Conjoint analysis is
becoming the technique of choice in major government sponsored valuation exercises both in the
US and abroad. This movement in part reflects concerns about possible biases associated with
contingent valuation that are assumed to be less prevalent in conjoint analysis (Hanley, et al
1998).
An important part of the contingent valuation literature was the development of an
incentive compatible provision rule that is made explicit to survey respondents (Arrow, et al,
1993). With conjoint analysis applications, however, respondents are simply asked to reveal
their preferences through various evaluation tasks. Studies will describe a payment mechanism
(such as a tax-price or a user-fee), however the actual rule used to determine which of the
options presented in the survey will be the option that is implemented, if any at all (the provision
rule) is left unspecified.
This study investigates the impacts of provision rules within conjoint choice
questionnaires. Using both private and public goods, we collect conjoint choice data using three
different provision rules and ten different treatments. First, we use an incentive-compatible,
individual provision rule (IPR) involving real payments and purchase. We then conduct a
treatment using an individual provision rule, but with hypothetical payment. Next, we use a
group provision rule (GPR) in which the option that receives the greatest support in the survey is
the option that is actually provided to every subject, regardless if this was his or her preferred
option. This provision rule is not incentive compatible, but important to understand as it mimics
the likely inferred provision rule in past conjoint surveys valuing public goods. Lastly, we
conduct treatments where no provision rule (NPR) is described to subjects. This treatment is
consistent with all previous conjoint applications for environmental valuation. Because no
provision rule is specified, this treatment can not be conducted in an actual-payment scenario —
6
-------
only hypothetical surveys can be conducted.
The treatments we conduct allow us to (1) examine the differences in choices due to
hypothetical versus real-payments (i.e., explore hypothetical bias), an issue that has received
considerable attention in the contingent valuation literature, (2) examine the effects of moving
away from incentive compatibility toward mechanisms that are more realistic (but not incentive
compatible) for conjoint exercises valuing public goods, and (3) examine whether the results
from the treatment with no decision rule (NPR) converges on the results from the incentive
compatible decision rule (IPR) or the group decision rule (GPR).
Preliminary results, based on nearly 2,000 subjects indicate that provision rules have
important effects on the responses to choice surveys. Results indicate that in surveys using
private goods, subjects opt to purchase the private good more often when either the non-
incentive compatible group-provision rule is described (GPR), or when no provision rule is
described (NPR). Results from surveys using a public good as the object of choice indicate a
similar pattern. In particular, conditional logit and random parameter logit models indicate that
the provision rule treatment affected the marginal values subjects revealed in the surveys.
Subjects were significantly less responsive to price in both the GPR and NPR treatments as
compared to the IPR treatment.
Lastly, preliminary results comparing the results from the IPR treatment in which
payments are hypothetical with the IPR treatment in which subjects actually pay for the good
and receive it as a result of their decisions in the choice survey, indicate significant differences in
behavior between these two treatments. Interestingly, while it is clear that subjects "opt out" of
the market more frequently when actual payments could result from their decisions, there may
not be significant differences in the subjects' responsiveness to prices across the two treatments.
However, these results are very preliminary in nature at this time.
7
-------
Internet-Based Stated Choice Experiments in Ecosystem Mitigation:
Methods to Control Decision Heuristics and Biases
John P. Hoehn,
Department of Agricultural Economics
Frank Lupi,
Department of Agricultural Economics and
Department of Fisheries and Wildlife
and
Michael D. Kaplowitz,
Department of Community,
Agriculture, Recreation and Resource Studies
Michigan State University
East Lansing, MI 48824
Acknowledgments
The reported research was supported, in part, by the: U.S. Environmental Protection Agency,
Science to Achieve Results (STAR) Grant #R827922; the Michigan Sea Grant Program; and the
Michigan Great Lakes Protection Fund. All opinions expressed and errors herein are the authors'
alone.
November 12, 2004
8
-------
Abstract
Internet-Based Stated Choice Experiments in Ecosystem Mitigation:
Methods to Control Decision Heuristics and Biases
John P. Hoehn, Frank Lupi, and Michael D. Kaplowitz.
Michigan State University
The research developed internet-based stated choice questionnaires to evaluate wetland ecosystem
mitigation and restoration. The goals were to estimate the in-kind values of ecosystem attributes and
test hypotheses about the performance of the questionnaires. A key question was whether the
ecosystem information and program descriptions were sufficiently detailed to meet the informational
needs of respondents, without overwhelming them with too much information. Behavioral research
shows that respondents' decisions are inconsistent and biased when confronted with too much
information. The research used a multistage design process to reduce the informational and choice
complexity perceived by respondents. The final questionnaire presented ecosystem information in
a tabular format that enabled respondents to easily identify the choice attributes and to compare
attribute levels and qualities across wetland pairs. A text format was developed as a control to
determine the degree that the tabular format controlled decision heuristics and biases.
Results indicated that the tabular format was successful in simplifying a complex choice without
eliminating relevant information. In-kind values estimated with the tabular data were consistent
with intuition and statistically significant. In contrast, text format responses were insensitive to high
quality wetlands and highly sensitive to poor quality wetlands, as expected when loss aversion
biases are present. Text responses were also more variable than the tabular responses. The results
suggest that a systematic questionnaire design process reduces the subjective complexity of
ecosystem choices without reducing the objective quality of information.
9
-------
Internet-Based Stated Choice Experiments in Ecosystem Mitigation:
Methods to Control Decision Heuristics and Biases
Wetland ecosystems are regulated under an array of Federal and state regulations. The goal
of many state regulations is similar to the Federal objective of "no net loss" of wetlands (National
Research Council 2001). To avoid a net loss of wetland services, Federal and state regulations may
require mitigation for activities that impair or destroy wetlands. Mitigation raises the issue of
determining what and how much should be done to offset the loss or impairment of a wetland. One
way that losses are offset is by restoring wetlands in locations near a destroyed or impaired wetland.
The amounts and types of restoration are typically determined on ecological grounds. However, a
purely ecological assessment may not adequately address wetland attributes that are valued by
human beings. If the latter values are overlooked, a net economic loss may be incurred despite the
no-net-loss goal.
Previous research shows that the ecological qualities of wetlands are indeed valued by
ordinary citizens (Heimlich etal. 1998;Kosz 1996; Phillips, Haney, and Adamowicz 1993; Stevens,
Benin, and Larson 1995). Both use and nonuse values are recognized as important to the economic
value of wetlands (Woodward and Wui 2001). Previous research is less clear about the values of
specific wetland attributes and qualities, such as wildlife habitat or access for recreation by the
public. Reported research tends to focus "on the question of 'what is the value' and not enough on
what, in particular, people value" (Swallow, 1998, p. 17).
Identifying the relevant wetland ecosystem attributes is important for both policy and
valuation. Wetlands are complex ecosystems that may be evaluated in different ways. Different
technical approaches characterize wetlands using different metrics and different attributes, such as
hydrogeomorphic types, wetland functions, wetland processes, and ecological values (National
10
-------
Research Council 2001). No net loss policies may mistakenly result in real and costly losses when
wetland policies ignore economically important attributes, or use metrics that are only partially
correlated with attributes that are valued by the general public.
Identifying the subset of economically relevant attributes is also important to the reliability
of stated choice experiments. Recent experiments show that choice complexity reduces the
consistency of stated choices, and increases the variance of stated choice results (Breffle and Rowe
2002; DeShazo and Fermo 2002; Swait and Adamowicz 2001). Including irrelevant attributes and
ill-defined attributes makes stated choices unduly complex. Such complex information sets are
likely to increase the variance of responses and reduce the statistical significance of estimated
values.
Behavioral research also indicates that complexity increases the respondents' use of
simplifying decision heuristics (Payne, Bettman, and Schkade 1999). Loss aversion is one of the
common decision heuristics recorded in behavioral research (Kahneman 2003; McFadden 1999).
Faced with a complex decision involving both losses and gains, loss-avoiding respondents make
decisions that myopically avoid losses, and fail to account for gains. Loss-avoiding respondents
tend to focus only on the potential losses and ignore potential gains. Such complexity-induced
heuristics are likely to result in severely biased value estimates to the extent they are evoked by
unnecessarily complex stated choice experiments.
The research reported below developed and tested internet-based questionnaires as a means
of eliciting in-kind values for wetland mitigation from members of the general public. The
questionnaire design drew on behavioral research for strategies to reduce the choice complexity
perceived by respondents. These strategies were incorporated into a four-stage questionnaire design
11
-------
process. Focus groups were used to identify the wetland attributes that were most salient to
respondents. Group and individual interviews were used next to test alternative information and
choice formats. Verbal protocol analysis was employed to identify questionnaire content and
attributes that were confusing or misleading to respondents. The final questionnaire presented
ecosystem information in a tabular format that made it easy for respondents to identify choice
attributes and to compare attribute levels across wetland pairs. A conventional text format was
developed as a control to determine the degree that the tabular format controlled decision heuristics
and biases.
Data from the tabular and text formats was used to estimate a mitigation equation that gave
the acreage of a restored wetland necessary to compensate respondents for the loss of an existing
wetland. The amount of restored acreage was conditioned on the acreage of the destroyed wetland,
the quality differences between the two wetlands, and the demographic characteristics of
respondents. With the data from the text format, the estimated in-kind values were consistent with
uncontrolled loss aversion bias. Text format responses were highly sensitive to poor quality wetlands
and insensitive to both wetland size and high quality wetlands. In contrast, tabular responses were
sensitive to wetland size, low quality wetlands, and high quality wetlands. The tabular format
appeared to facilitate informed and balanced tradeoffs.
Economic Model of Mitigation Choices
Wetland mitigation compensates for the loss of wetland services with the restoration of
wetland services in a different location. As such, wetland mitigation offers a natural setting for
eliciting pair-wise stated choices between a restored and drained wetland. An individual may be
asked whether a restored wetland of a given acreage and quality is sufficient to compensate for the
12
-------
loss of a destroyed wetland of a given acreage and quality. This section presents a utility theoretic
framework for such choices and derives the in-kind values associated with wetland acreage and
wetland ecosystem qualities.
Stated preference techniques are widely applied in market research (Louviere 1991),
transportation economics (Bates 2000), development economics (Rubey and Lupi 1997), and
environmental economics (Adamowicz et al. 1998; Boxall et al. 1996; Mackenzie 1993; Opaluch
et al. 1993; Swallow et al. 1998). Stated choices are usually estimated within a random utility
formulation. In this analysis, we derive a choice model based on offered restoration versus a desired
amount of utility compensating restoration.
The analysis begins with the preferences of a respondent drawn from the general public. The
respondent has preferences over wetland size and wetland qualities that are conditioned on the
respondent's demographic characteristics. These preferences are summarized by a utility function,
(1) u = u(x,q,c)
defined on wetland acreage, x, a /^-element vector denoting the quality of wetland services, q, and
a TV-element vector of individual respondent characteristics, c.
Consider the loss of a wetland that is d acres in size with qualities qd. The amount of
restored acreage, /??, with qualities qm, that compensates for the loss of d with qualities qd is
(2) u(m,qm,c) = u(d,qd,c)
13
-------
Equation (2) states the amount of compensatory restoration, m, as an implicit function of lost
wetland acreage, lost and restored wetland service qualities, and individual respondent
characteristics.
A compensatory mitigation equation is derived by inverting the left-hand side of equation
(2) about the amount of restored acreage,
(3) m = u^iq^uiq^d))
Equation (3) may be rewritten as a mitigation function,
(4) ™ = m(d,q m,q d,c)
Equation (4) is similar to an income compensation function (Chipman and Moore 1980) except that
the mitigation compensation function is denominated in restored acreage rather than income. The
mitigation equation states the amount of quality adjusted restored acreage required to compensate
for the loss of an existing wetland of a given size and quality.
The mitigation function is approximated with a linear function of destroyed wetland acreage,
the difference between the qualities of the restored and destroyed wetland, respondent demographic
characteristics, and a stochastic term,
K N
(5) m = Po + hd + E P Mk + E Y nCn + e
k-1 n-1
where P0 is an intercept coefficient, is the coefficient of the acreage of the destroyed wetland,
d; P^ is the coefficient of the difference between the restored and destroyed wetland in the kt\i
14
-------
wetland quality, A qk; yn is the coefficient of the //th respondent characteristic, cn, such as income
level or having never visited a wetland; and e is a stochastic error term. The stochastic term, e,
represents random choice effects that are unobserved by the researcher.
The stated choice experiments present respondents with a pair of wetlands, the restored
wetland with its quality attributes, and the destroyed wetland with its quality attributes. The
respondent can either accept or reject the restored wetland as compensation for the loss. A
respondent accepts the restored wetland as compensation if the size of the restored wetland is greater
than the compensating mitigation described by equation (5), given the size of the destroyed wetland,
the quality differences between the restored and destroyed wetland, and individual characteristics.
A respondent rejects the restored wetland as compensation if the size of the restored wetland is less
than the compensatory mitigation described by equation (5).
Given the stochastic term in equation (5), a respondent's decision is not known with certainty
by a researcher. However, the probability that an individual accepts restored acreagew with
qualities qw is
[accept w \d,qd,qw] = Pr(w > m \d,qd,qw)
^ = Pr(W > P0 + $dd + £ PMk + £ VnCn + 6)
(O) k=\ n=1
= Pr(w - P0 - vdd ~ £ VAqk - £ YrPn > e)
k=\ n=1
When the stochastic term, e, is an independently distributed normal random variable, the probability
of accepting the offered restored wetland is
15
-------
K N
(7) Prob[accept w \d,qd,qw] = ®[(w - p0 ~ $dd ~ X) PA* ~ X) Y«c«)/oe]
£=1 n=1
where
-------
Equation (8) is a density function conditioned on the random variable v representing the
individual effect. The random effects probit model is derived by setting up the likelihood equation
for the ordinary probit and computing the expectation of the likelihood equation with respect to v.
The expected likelihood equation is then evaluated by Gaussian quadrature to obtain maximum
likelihood estimates of the coefficients and standard deviations (Butler and Moffitt 1982).
Stated Choice Questionnaire
The purpose of the questionnaire design process was to develop questionnaires that
accurately describe wetland qualities and choices to respondents, while controlling the perceived
complexity of wetland information and tradeoffs. Previous research shows that such complexity can
introduce inconsistencies across stated choices (Breffle and Rowe 2002; DeShazo and Fermo 2002;
Swait and Adamowicz 2001) and potentially lead to characteristic biases due to loss aversion and
other decision heuristics (Kahneman 2003; McFadden 1999).
Stated choice research tends to treat complexity as an objective phenomenon (Breffle and
Rowe 2002; DeShazo and Fermo 2002; Swait and Adamowicz 2001), but behavioral research
indicates that it is subj ective and conditioned on the structure of a particular informational treatment
(Carlson, Chandler, and Sweller 2003; Ganier, Gombert, and Fayol 2000; Simon 1974). An
informational treatment may focus on relevant information or it may force a reader to sort through
relevant and irrelevant data. Focusing on the important and salient features reduces the number of
features to be evaluated by respondents and thereby reduces one aspect of complexity. In addition,
the same information may be presented in complex or simple ways. For example, the structure of
human memory appears to make it easier for people to remember more information when it is
17
-------
presented with both text and graphical representations than when it is presented solely as text (Luck
and Vogel 1997).
The research used the four-stage process described in Figure 1 to build the lessons of
behavioral and cognitive research into a stated choice questionnaire. The objective of the first stage
of the process was to determine the types of information that respondents find most relevant to
mitigation choices. The first stage examined the general public's knowledge about wetlands, their
experience with wetlands, and the priorities they placed on protecting and managing specific
wetland qualities and services.
Data for the first stage analysis was obtained in focus group discussions. Participants were
recruited using telephone numbers drawn randomly from the Lansing-area phone book.
Respondents were asked whether they could attend a discussion group concerning 'critical policy
issues'. Respondents were selected so that the mix of focus group participants was representative
of the demographic characteristics of mid-Michigan. Focus group discussions were guided by
trained moderators using a written discussion guide. The discussion guide began by eliciting top
environmental concerns, and then probing these concerns to assess their connections with aquatic
and wetland ecosystems. The discussion guide gradually probed the topic of wetlands and elicited
respondents' baseline knowledge and experience with wetlands. The last segment of the discussion
guide presented a wetland mitigation case and asked respondents to evaluate the case in terms of the
adequacy of compensatory mitigation.
Results from the first stage group discussions were used to construct alternative information
and choice formats. The alternative formats varied in three dimensions. The first dimension was
the way the formats used hierarchical taxonomies to categorize wetland attributes. Behavioral
18
-------
research shows that information is readily assimilated and used when it is organized in subjectively
meaningful categories or chunks (Gobet et al. 2001; Simon 1974). For instance, it is more difficult
to recall the letters BMICSIACB than when they are arranged in as the chunks IBM-CIA-CBS
(Bettman, Payne, and Staelin 1987). Similarly, wetland species may be chunked into taxonomic
categories, such as wading birds, song birds, and amphibians. A format that divides species into
species categories and lists specific species within that category is likely to appear less complicated
to respondents than a format that simply lists individual species. Such a hierarchical listing may
also appeal jointly to non-expert respondents and lay experts. Non-expert respondents may focus
on the species categories while lay experts may find the sets of specific species meaningful to their
evaluations and choices.
The second dimension that varied across the alternative formats was the way that the restored
and destroyed wetlands were presented. For instance, the information about each of the two
wetlands may be presented sequentially on separate pages or on the same page. Tabular designs
may array wetland attribute information in corresponding columns to facilitate comparisons across
the destroyed and restored wetlands. Previous research indicates that tabular designs reduce the
perceived task complexity and reduce the amount of time needed for task completion (Carlson,
Chandler, and Sweller 2003; Ganier, Gombert, and Fayol 2000).
The third dimension was a method of describing quality changes across wetlands. Focus
group respondents tended to describe wetland experiences in terms of what they saw as they drove
by or walked through wetlands. These comments suggested that metrics based on transect sampling
may be a meaningful way to represent quality differences between wetlands. Transect sampling
plots a path through a given area and counts all the features of interest along that path within a given
19
-------
time (Buckland et al. 1993). A wetland visitor might be thought of as conducting an informal
transect where the quality of the wetland ecosystem influences the chance of seeing different
categories of species, so one of the quality indicators evaluated was based on the chance that a lay
visitor might see certain species during a visit to the wetland.
The second stage of the questionnaire design process evaluated the alternative information
treatments in group and individual interviews. Participants for the group and individual sessions
were drawn from the Lansing area through random selection using a local telephone directory. The
interviews were divided into interview and debriefing segments. During the interview segment,
respondents completed one of the alternative questionnaires. The debriefing segment began once
the questionnaires were completed. The debriefing segment was led by a trained moderator who
followed a written discussion guide. The guide asked respondents to discuss how they understood
the tasks required by different parts of a questionnaire, examined any difficulties that respondents
had in completing a questionnaire, and ended with a short quiz to assess respondents'
comprehension of the questionnaire.
The debriefing interviews identified one questionnaire design as superior to the others. The
questionnaire used hierarchical categories, a tabular layout, and placed the qualities of the drained
and restored wetlands in adj acent columns on a single page of the questionnaire. Respondents using
this format made specific and repeated references to wetland qualities during the debriefings. No
specific quality seemed to dominate their recollections. Rather, respondents seemed to have
balanced and nuanced perceptions of the wetland attributes being compared. The tabular format
appeared to facilitate choice-related comparisons and tradeoffs across the two wetlands. Several
20
-------
respondents commented that the choices were almost "too easy," despite the fact that the choices
involved nine attributes, three possible quality levels for each attribute, and two wetlands.
The third stage of the questionnaire design process used individual pretest interviews to test
and revise the prototype questionnaire. Pretest respondents were again drawn from the Lansing, MI
area using a random selection method based on area telephone books. Participants were paid an
honorarium to attend the pretest at a specified day and time on the Michigan State University
campus. Pretest interviews were divided into questionnaire self-administration and debriefing
segments. In the survey self-administration segment, each participant completed a prototype
questionnaire. Once the questionnaire was complete, the respondent was guided to a private office
for an in-depth debriefing interview.
Debriefing interviews followed a detailed written debriefing guide. The guide began by
asking a respondent to recall and describe their thoughts as he or she completed particular segments
of the questionnaire. Additional questions focused on how and whether the respondent understood
the information and choice segments of a questionnaire. The debriefing ended with several
knowledge-based questions to determine whether respondents understood important aspects of the
questionnaire and choice question. Debriefing data were used to revise the prototype questionnaire,
with the resulting revised questionnaires subjected to further testing.
The final stage of the design process programmed the questionnaire for use on the internet.
The programming was done to preserve the appearance of a paper questionnaire as much as possible.
The draft internet questionnaire was pretested over the internet with respondents from the Lansing
area, primarily to test the technical characteristics of the questionnaire. Respondents were recruited
through random identification from telephone records and paid an honorarium to complete the
21
-------
questionnaire and debriefing interview. Respondents used a variety of operating systems, e-mail
systems, web browsers, computer displays, and different internet service providers. Despite the
wide range of situations, relatively few problems arose with the web-based questionnaire. The minor
problems that did arise were readily remedied by some minor reprogramming.
The final questionnaire focused on a subset of wetland attributes and presented these
attributes in a tabular design. Attributes were selected for inclusion based on the data obtained in
the qualitative research and pretesting. Attributes included wetland size, type of vegetative cover,
accessibility by the general public, and suitability as habitats for plant and animal species.
Vegetative cover was categorized as marsh, wooded wetland, or a mix of both marsh and wooded
wetland.
Figure 2 shows a portion of the final tabular information format included in the final
questionnaire. The tabular form arrayed the relevant wetland choice information in two adjacent
columns, one for each wetland under consideration. Wetland habitats were described in five
dimensions; habitat quality for amphibians and reptiles, habitat quality for small mammals, habitat
quality for song birds, habitat quality for wading birds, and habitat quality for wild flowers.
Each type of habitat was described with a rating of poor, good, or excellent based on the
transect sampling concept discussed above. Habitat quality ratings were provided for both the
drained and restored wetlands. A narrative box at the bottom on the table explained each of the
quality ratings. The ratings were based on what a visitor was likely to see during a visit to the
wetland. A poor rating was indicated by " and was defined as a wetland habitat that supported
"these species in very small numbers...[so] a trained observer is unlikely to find any of these
species." A "good" rating meant that the wetland habitat supported "these species in average
22
-------
numbers... [so] a casual observer is likely to see a few of these species." An "excellent" rating meant
that the wetland habitat supported "these species in better than average numbers...[so] a casual
observer is very likely to see a variety of these species."
Internet-Based Stated Choice Experiments
The obj ective of the internet experiment was to test the performance of the developed questionnaire.
The hypothesis was that the tabular format reduced complexity and encouraged reasoned decisions
informed by a balanced view of all wetland attributes. To test the hypothesis, an experimental
control was developed based on a text version of the information format. The text version contained
information that was obj ectively identical to the tabular questionnaire. The only difference between
the tabular and text questionnaires was that the text format used sentences to convey the information
about wetland attributes and qualities. Figure 3 gives an example of the text format.
Two empirical consequences were expected if the tabular format reduced the complexity
perceived by respondents. First, prior research showed that reduced complexity increases the choice
consistency and reduces the variance of choice responses (Breffle and Rowe 2002; DeShazo and
Fermo 2002; Swait and Adamowicz 2001). In the present case, reducing perceived complexity
should result in greater consistency and smaller standard deviations for both the cross-section effect,
e, and the respondent effect, v. Thus,theestimatedstandarddeviationsforthetabularformatdata, og
and ov, should be smaller than the estimated standard deviations for the text format data.
Second, behavioral research indicates that complexity leads to increases in the use of
decision heuristics and biases (Payne, Bettman, and Schkade 1999). Viscusi and Magat (1987)
found that text formats had less impact on risk avoidance behavior and willingness to pay than
tabular formats. Psychological research stresses that cognitive constraints lead to characteristic
23
-------
biases when dealing with complicated decisions (Kahneman 2003; Payne, Bettman, and Schkade
1999). Loss aversion is one characteristic and common decision bias (Kahneman, Knetsch, and
Thaler 1991;McFadden 1999). With loss aversion, respondents overweight losses and underweight
gains.
With complex wetland choices, it was hypothesized that loss aversion would lead
respondents to overweight wetlands with poor quality attributes and underweight wetlands with
excellent quality attributes. As a result, a mitigation equation estimated with the text data was
expected to have larger coefficients for variables indicating poor quality than a mitigation equation
estimated with the tabular data. In contrast, a mitigation equation estimated with text data would
be expected to have smaller coefficients for variables indicating excellent quality. The text
coefficients for variables indicating excellent quality may be ignored by respondents, with the result
that their coefficients may not be statistically different from zero. In contrast, the statistical
significance of the coefficients estimated with tabular data is likely to be more evenly distributed
across poor and excellent quality indicators.
Data to estimate the mitigation equations and test these empirical hypotheses was collected
in a large-scale internet experiment with Michigan residents. Access to a panel of potential web-
based respondents was purchased from Survey Sampling International (SSI), a commercial provider
of sampling frames and databases. The SSI panel is a self-selected sample of potential respondents
with known demographic characteristics.
The web-based experiment was implemented in multiple stages beginning in October and
ending in December, 2003. E-mail invitations to 16,936 members of the SSI panel, resulted in 3,420
clicks on a welcome page to the web-based questionnaire. From the welcome page, 25 percent of
24
-------
respondents were randomly assigned to the text format. In all, 2,689 respondents began the first
page of the questionnaire. Usable questionnaires with at least one completed mitigation choice
and complete demographic information numbered 1,326. This was 8 percent of the number of e-
mail invitations and 40 percent of those visiting the welcome page. Eight percent is a midrange rate
for recent internet experiments (Berrens et al. 2002).
Results
The tabular and text formats yielded two sets of data suitable for an analysis of mitigation choices
and values. The data pertaining to the tabular format were the preferred, core data set, since the
tabular design was subject to the full iterative design process. The purpose of the text format data
was to provide a baseline for evaluating the performance of the tabular design. By hypothesis, the
text format leads to (1) more inconsistency in stated choices and (2) cognitive biases that overweight
losses in wetland qualities and underweight gains in wetland quality.
The text and tabular data contained three types of variables. First, there were the wetland
choice variables. Respondents were given five mitigation scenarios and were asked to determine
whether the restored wetland was sufficient to offset the loss of a drained wetland. Hence, each
individual recorded accept or rej ect choices for up to five restoration scenarios. Second, there were
the variables that described the acreage and qualities of both the drained and restored wetlands.
Third, there were demographic variables for each respondent.
Table 1 lists demographic characteristics for respondents to the tabular and text versions of
the questionnaire. There were 937 respondents to the tabular version and 363 respondents to the text
version who had responses complete enough to the used in the choice analysis. The choice analysis
required complete responses for the variables listed in Table 1.
25
-------
Mean levels of income, education, age, and gender were similar for respondents to both the
tabular and text versions. One exception was for the age of respondents where the text data set
contained about 8 percent more respondents who were over 65 years of age. The mean income level
for respondents to both versions was about the same as the 2002 Census mean for the State of
Michigan. Respondents to the questionnaires were somewhat more schooled with some college
study and were more likely to be female and over 65.1 Finally, 15 percent of the respondents in each
sample had never visited a wetland.
The tabular and text data was used to estimate a mitigation equation (5) using the random
effects probability model of equation (8). Table 2 lists the general characteristics of the two
estimated equations. The data included 4,685 choices from the 963 respondents who used the
tabular format and 1,811 choices from the 363 respondents who used the text format. The tabular
and text equations performed about equally well in predicting both yes and no responses.
The tabular and text equations are noticeably different in the standard deviations for both the
cross-sectional and respondent effects. The standard deviations for the text data are more than twice
the size of those for the tabular data. The third column shows that the differences between the two
sets of standard deviations are statistically different from zero at the 90 percent level of significant.
These results indicate that respondents make more consistent choices with the tabular questionnaire
format than the text questionnaire format. The tabular format appears to be successful in reducing
perceived complexity, at least as indicated by the variability of choices.
'The sample selection procedures were intended to be weighted by the Census proportions
for males and females in the 2000 Census. However, an error occurred in subcontractor's sample
selection process during the waves 1 and 2 of the experiment. The error was corrected for waves
3 to 6 and the sample size was increased to meet the demographic criteria for the initial sample
design.
26
-------
Table 3 lists the wetland attributes and demographic variables used to estimate the mitigation
equation coefficients. Wetland size was one of the variables and ranged from 5 to 19 acres for the
drained wetlands and from 4 to 48 acres for the restored wetlands. Other wetland characteristics
were described as categorical variables. The drained and restored wetlands (a) allowed access by
the public, denoted by a "yes," (b) allowed access to the pubic with developed trails, denoted by
"yes-trails," or (c) made no provision for public access, denoted by "no." The type of wetland was
either a marsh, a wooded wetland, or a mixture of marsh and woodlands.
The changes in wetland characteristics variables, Ax ., were transformations of the data in
the questionnaires. The change in access variable indicated whether there was a change in public
access in the restored wetland relative to the drained wetland. The change in access variable was
given a value of 1 if the restored wetland allowed public access while the drained wetland did not.
Change in access was -1 if the restored wetland did not provide for public access while the drained
wetland did provide for public access. In other cases, change in access was set to 0.
The change in wetland type variable was a simple, unsigned dummy variable. It was given
a value of 1 if there was a change in wetland type between the restored and drained wetlands and
set to 0 if there was no change in type.
The changes in wetland habitat variables were computed from dummy variables representing
the poor and excellent categories. The first step was to assign a dummy variable for each of the poor
and excellent quality levels of the drained and restored wetlands. Each of the "poor" dummy
variables was given a value of 1 if a particular habitat category was poor in quality, and was set to
zero otherwise. Each of the "excellent" dummy variables was given a value of 1 if a particular
habitat quality was excellent in quality, and was set to zero otherwise. Dummy variables were
27
-------
created for the "poor" and "good" variables for four habitats (reptiles/amphibians, song birds,
wading birds, and wild flowers) and both wetlands, so there were 8 initial dummy variables for
quality. The habitat dimension for small animals was kept constant across the choice experiments,
so no dummy was created to indicate the quality of habitat for small animals.2
The second step in computing the habitat change variables was to compute the difference in
the habitat dummy variables between the restored and drained wetlands. For instance, the change
in poor dummies for reptiles/amphibians was the difference between (a) the poor reptiles/amphibians
dummy for the restored wetland and (b) the poor reptiles/amphibians dummy for the drained
wetland. A value of 1 for the latter variable meant that the reptiles/amphibian habitat was poor for
the restored wetland and not poor for the drained wetland. A value of -1 meant that the
reptiles/amphibian habitat was not poor in the restored wetland and poor in the drained wetland.
A value of 0 meant no change in the habitat quality for the reptiles/amphibians habitat across the two
wetlands. Similar habitat change variables were computed for computed for the poor and excellent
dummies variables for each of the 4 habitat categories, resulting in 4 variables to reflect changes in
poor quality habitat and 4 variables to reflect changes in excellent quality habitat.
The demographic characteristics variables were simple levels or categorical dummy
variables. Income was measured in thousands of dollars. The remainder of the respondent variables
were categorical dummy variables, taking the value of 1 if the respondent had the characteristic, and
taking the value of 0 otherwise.
2The small animals habitat quality was kept constant across the two wetlands to reduce the
size of the experimental design. Because the small animals are generalists, this type of habitat was
not thought to vary across substantially across the common wetlands under consideration, and the
other habitat categories were sufficient to demonstrate the role of habitat quality with respect to
respondents' preferences.
28
-------
Table 3 display s the estimated mitigation coefficients for the tabular and text equations. The
second and third columns list the estimated normalized coefficients for the tabular and text data.
The final column lists the differences between the coefficients of the tabular and text coefficients.
The coefficients for the tabular equation have plausible signs and are mostly statistically different
from zero at the 95 percent level. The normalized coefficient for drained acreage is equal to 1.42.
A acreage coefficient equal 1 would mean that restored wetland acreage is a very close substitute
for drained acreage. However, the coefficient is 42 percent larger than one and statistically different
from 1 at the 95 percent level. The coefficient implies that the mean respondent requires
compensation of 1.42 restored acres for each acre of drained wetland, even when the two wetlands
are otherwise identical in access, wetland type, and habitat quality.
The premium of 42 percent on the drained wetland acreage is similar to Mullarkey' s finding
that natural wetlands are more valuable than restored wetlands (Mullarkey 1997). However,
Mullarkey found a much larger premium on dollar value of natural wetlands, perhaps due to
unaddressed differences in wetland qualities.
Public access and wetland type also have a significant impact on the amount of mitigation
acreage that compensates for loss of the drained wetland. The public access coefficient indicates
that providing public access reduces the compensating number of mitigated acres by 5.76 acres. A
change in wetland type increases the compensating amount of mitigation by 4.69 acres.
The change in habitat variables are all significantly different from zero for the tabular data
and have algebraic signs consistent with intuition. Reductions in habitat qualities from good to poor
require additional acreage to offset the loss in quality. A change in a reptile/amphibian habitat from
good to poor requires 8.19 additional restored acres to offset the loss of quality. A reduction in a
29
-------
wild flower habitat from good to poor requires 2.33 acres of additional restored acreage.
Improvements in habitat quality relative to the drained wetland reduce the amount of restored
acreage required for mitigation. A change from a good wading bird habitat in the drained wetland
to an excellent habitat in the restored wetland reduces the number of restored acres by 5.09 acres.
An improvement from a poor habitat in the drained wetland to an excellent habitat in the restored
wetland is assessed by summing the appropriate coefficients. For instance, for song birds, a change
from poor to excellent reduces the number of restored acres by 6.56 plus 3.80, an overall reduction
of 10.36 acres.
Several demographic characteristics affect the level of mitigation that compensates for
wetland loss. Increases in respondents' income and schooling tend to reduce the size of
compensatory mitigation projects. Having visited a wetland at some point in the past also leads to
reductions in the amount of compensating mitigation acres. The latter variable is interesting since
it indicates that individuals who have some experience with common wetlands are more inclined to
accept the replacement of existing wetlands with restored wetlands.
The notable feature of the text coefficients is the large size of the poor quality habitat
coefficients and the small size of the excellent quality habitat indicators. Respondents who were
randomly given the text-based choice question require more acreage compensation for loss in
quality than the respondents who were randomly selected to receive the tabular-based choice
question. Alternatively, for improvements in restored habitat quality relative to the drained wetland,
text respondents behave in just the opposite fashion; they underweight improvements.
The final column of Table 3 shows that these asymmetries are statistically significant for
each of the poor habitat coefficients and are significant as a group for the excellent habitat
30
-------
coefficients. The results suggest that relative to the tabular format the text respondents fell prey to
decisions biases that have been noted by psychologists: respondents tend to overweight losses and
underweight gains. The tabular questionnaire appears unaffected by such biases. Coefficient
estimates are relatively precise and the differences between coefficients seem reasonable and
consistent with intuition. The iterative design process appears successful in deriving a questionnaire
that supported balanced, reasoned decisions for rather complex mitigation choices.
The strong asymmetry in the resulting data from the text choice questionnaire also appears
in estimating mitigation acreage requirements. Suppose one is considering mitigation for the
drainage of a 20-acre wetland with good habitat quality in each of the four habitat categories.
Consider two restoration projects: the first involves restoration that results in all four habitat
qualities being in poor condition, and the second involves restoration that results in all four habitat
qualities being in excellent condition. In the first case, the mitigation equation estimated with the
tabular data requires 49 acres of restored wetland acres as compensation, but the equation estimated
with the text data requires 106 acres of restored wetland as compensation. Conversely, in the second
case involving restoration with excellent habitat quality, computing compensating restoration
acreage with the tabular equation requires 11 acres of compensation while the text equation requires
28 acres as compensation.
The mitigation examples highlight the differences between the text and tabular data, and the
hypothesized superiority of the tabular questionnaire. With the text questionnaire, respondents
appear to overweight losses in habitat quality and underweight gains. The underweighting of gains
is rather extreme, since the individual habitat coefficients for improvements are small in size and
statistically indistinguishable from zero. In contrast, the tabular data results in coefficients that are
31
-------
economically significant, statistically different from zero, much more balanced in their assessment
of wetland gains and losses, and accord with the respondent feedback from the focus groups and
pretest interviews.
Conclusion
The research demonstrates that stated choice experiments with complex ecosystems are
feasible for the general public. Careful research on baseline knowledge and systematic pretesting
appear essential for obtaining reasonable, unbiased stated choice results. The tabular questionnaire
format that resulted from a four-stage design procedure appeared to perform well. The research also
used a simple text-based information treatment as an example of the type of questionnaire that might
be developed without the iterative questionnaire design process. The simple text-based
questionnaire revealed the kinds of asymmetric biases anticipated on the basis of recent
psychological and economic research (McFadden 2001). The text-based descriptions resulted in
losses in ecosystem quality being overweighted and gains in quality being underweighted relative
to those estimated using the tabular format. Thus, while ecosystem choices may be complex enough
to strain respondents' decision capacities, systematic questionnaire development seems able to help
researchers arrive at formats that reduce or eliminate the impact of characteristic biases on the estimated
values.
The results demonstrate that wetland qualities and services are valued by members of the general
public. From qualitative research, wetland habitats for small animals, birds, and special plants were
found to be of special interest and value to respondents (Hoehn, Lupi, and Kaplowitz 2003; Kaplowitz,
Lupi, and Hoehn 2004). Respondents had direct experience with the latter types of wetland habitats and
saw them as directly impacted by mitigation activities. The importance of habitat quality emerged
32
-------
consistently at all stages of the research including the initial focus groups, the pretest phase, and the
web-experiments. This finding is similar to other recent research on wetland ecosystems (Azevedo,
Herriges, and King 2000; Johnston et al. 2002; Stevens, Benin, and Larson 1995; Swallow et al. 1998).
Two aspects of the research need to be kept in mind in interpreting the results. First,
respondents to both the qualitative and quantitative research were drawn from residents of Michigan.
Michigan's climate is characteristic of the humid north-central portion of the United States. Wetlands
are a common landscape feature, so Michigan residents may have more experience with wetlands than
those in other parts of the United States, especially those living in arid regions. Second, while the study
provides estimates of how to adjust mitigation ratios to account for differences in habitat quality, it
should be considered a first step. The obj ective of this research was not to estimate values representing
a particular population, but to develop and evaluate stated choice valuation methods and procedures.
Further research is needed to implement the developed procedures in a statistically representative
sample. Third, the wetlands considered here were common types which are regularly subj ect to permit
actions in Michigan. The study results do not apply to rare wetlands, rare habitats, or rare species.
Likewise, in the wetland choices studied here, respondents were explicitly asked to hold other functions
of wetlands constant.
33
-------
References
Adamowicz, W., P. Boxall, J. Louviere, J. Swait, and M. Williams. 1998. Stated Preference Methods
for Valuing Environmental Amenities. In Valuing Environmental Preferences: Theory and
Practice of the Contingent Valuation Method in the U.S., E.C., and Developing Countries,
edited by I. Batemen and K. Willis. Oxford: Oxford University Press.
Azevedo, Christopher, Joseph A. Herriges, and Catherine L. King. 2000. Iowa Wetlands: Perceptions
and Values. Ames: Center for Agricultural and Rural Development Iowa State University.
Bates, John. 2000. History of Demand Modelling. In Handbook of Transport Modelling, edited by D.
A. Hensher and K. J. Button. Handbooks in Transport, vol. 1. Amsterdam; New York and
Oxford: Elsevier Science Pergamon.
Berrens, R.P., A.K. Bohara, H.C. Jenkins-Smith, C.L. Silva, and D.L. Weimer. 2002. Information and
Effort in Contingent Valuation Surveys: Application to Global Climate Change Using National
Internet Samples. Journal of Environmental Economics and Management 47:331-363.
Bettman, J.R., J.W. Payne, and R. Staelin. 1987. Cognitive Considerations in Presenting Risk
Information. In Learning About Risk, edited by K. W. Viscusi and W. A. Magat. Cambridge,
MA: Harvard University Press.
Boxall, P., W. Adamowicz, J. Swait, M. Williams, and J. Louviere. 1996. A Comparison of Stated
Preference Methods for Environmental Valuation. Ecological Economics 18:243-253.
Breffle, W.S., and R.D. Rowe. 2002. Comparing Choice Question Formats for Evaluating Natural
Resource Tradeoffs. Land Economics 78 (2):298-314.
Buckland, S.T., D.R. Anderson, K.P. Burnham, and J.L Laake. 1993. Distance Sampling: Estimating
Abundance of Biological Populations. London: Chapman and Hall.
Butler, J.S., and Robert Moffitt. 1982. A Computationally Efficient Quadrature Procedure for the One-
Factor Multinomial Probit Model. Econometrica 50 (3):761-764.
Cameron, T.A., and M.D. James. 1987. Efficient Estimation Methods for 'Closed-Ended' Contingent
Valuation Surveys. Review of Economics and Statistics 69:269-276.
Carlson, R, P. Chandler, and J. Sweller. 2003. Learning and Understanding Science Instructional
Material. Journal of Educational Psychology 95 (3):629-640.
Chipman, J.S., and J.C. Moore. 1980. Compensating Variation, Consumer's Surplus, and Welfare.
American Economic Review 70 (5):933-949.
DeShazo, J.R., and G. Fermo. 2002. Designing Choice Sets for Stated Preference Methods: The Effects
of Complexity on Choice Consistency. Journal of Environmental Economics andManagement
44:123-143.
Ganier, F., J.E. Gombert, and M. Fayol. 2000. The Effects of Instructional Formats on Procedural
Learning when Using Technical Documents. Travail Humain 63 (2): 121-152.
Gobet, F., P.C.R. Lane, S. Croker, P.C-H. Cheng, G. Jones, I. Oliver, and J.M. Pine. 2001. Chunking
Mechanisms in Human Learning. Trends in Cognitive Sciences 5 (6):236-243.
Greene, W.H. 2000. Econometric Analysis. 4th ed. Upper Saddle River, NJ: Prentice-Hall.
Heimlich, Ralph E., Keith D. Wiebe, Roger Claassen, Dwight Gadsby, and Robert M. House. 1998.
Wetlands and Agriculture: Private Interests and Public Benefits. Washington, D.C.: Resource
Economics Division, Economic Research Service, U.S. Department of Agriculture.
34
-------
Hoehn, J.P., F. Lupi, and M.D. Kaplowitz. 2003. Untying a Lancastrian Bundle: Valuing Ecosystems
and Ecosystem Services for Wetland Mitigation. Journal of Environmental Management
68:263-272.
Johnston, Robert J., Gisele Magnusson, Marisa J. Mazzotta, and James J. Opaluch. 2002. The
Economics of Wetland Ecosystem Restoration and Mitigation: Combining Economic and
Ecological Indicators to Prioritize Salt Marsh Restoration Actions. Am J Agricultural
Economics 84 (5): 1362-1370.
Kahneman, D., J.L. Knetsch, andR.H. Thaler. 1991. Anomalies the Endowment Effect, Loss Aversion,
and Status Quo Bias. Journal of Economic Perspectives 5 (1): 193-206.
Kahneman, Daniel. 2003. Maps of Bounded Rationality: Psychology for Behavioral Economics.
American Economic Review 93 (5): 1449-1475.
Kaplowitz, Michael D., Frank Lupi, and John P. Hoehn. 2004. Multiple-Methods for Developing and
Evaluating a Stated Preference Survey to Value Wetland Ecosystems. In Methods for Testing
and Evaluating Survey Questionnaires, edited by S. Presser, J. M. Rothgeb, M. P. Couper, J.
T. Lessler, E. Martin, J. Martin and E. Singer. New York: John Wiley and Sons.
Kosz, Michael. 1996. Valuing Riverside Wetlands: The Case of the "Donau-Allen" National Park.
Ecological Economics 16:109-127.
Louviere, J. J. 1991. Experimental Choice Analysis: Introduction and Overview. Journal of Business
Research 23:291-297.
Luck, S.J., and E.K. Vogel. 1997. The Capacity of Visual Working Memory for Features and
Conjunctions. Nature 390 (6657):279-281.
Mackenzie, J. 1993. A Comparison of Contingent Preference Models. American Journal of Agricultural
Economics 75:593-603.
McFadden, Daniel. 1999. Rationality for Economists? Journal of Risk and Uncertainty 19 (l-3):73-105.
McFadden, Daniel. 2001. Economic Choices. American Economic Review 91 (3):351-378.
Mullarkey, Daniel J. 1997. Contingent Valuation of Wetlands: Testing Sensitivity to Scope.
Dissertation, Agricultural and Applied Economics, University of Wisconsin, Madison, WI.
National Research Council. 2001. Compensating for Wetland Losses under the Clean Water Act.
Washington, DC: National Academy Press.
Opaluch, J. J., S. Swallow, T. Weaver, C. Wessels, and D. Wichlens. 1993. Evaluating Impacts from
Noxious Waste Facilities: Including Public Preferences in Current Siting Mechanisms. Journal
of Environmental Economics and Management 24:41-59.
Payne, J.W., J.R. Bettman, and D.A. Schkade. 1999. Measuring Constructed Preferences: Towards a
Building Code. Journal of Risk and Uncertainty 19 (l-3):243-70.
Phillips, W. E., T. J. Haney, and W. L. Adamowicz. 1993. An Economic Analysis of Wildlife Habitat
Preservation in Alberta. Canadian Journal of Agricultural Economics 41 (4):411-418.
Rubey, L, and F. Lupi. 1997. Predicting the Effects of Market Reform in Zimbabwe: A Stated
Preference Approach. American Journal of Agricultural Economics 78:89-99.
Simon, H.A. 1974. How Big Is a Chunk? Science 183:482-488.
Stevens, Thomas H., Samuel Benin, and Joseph S. Larson. 1995. Public Attitudes and Economic Values
for Wetland Preservation in New England. Wetlands 15:226-231.
Swait, J., and W. Adamowicz. 2001. Choice Environment, Market Complexity, and Consumer
Behavior: A Theoretical and Empirical Approach for Incorporating Decision Complexity into
35
-------
Models of Consumer Choice. Organizational Behavior and Human Decision Processes 86
(2): 141-167.
Swallow, S., M. Spencer, C. Miller, P. Paton, R. Deegen, L. Whinstanley, and J. Shogren. 1998.
Methods and Applications for Ecosystem Valuation: A Collage. In Proceedings of the first
Workshop in the Environmental Policy and Economics Workshop Series, edited by B.
Kanninen. Washington, D.C.: Office of Research and Development and Office of Policy, U.S.
Environmental Protection Agency.
Viscusi, W. Kip, and Wesley A. Magat. 1987. Learning about Risk. Cambridge, MA: Harvard
University Press.
Woodward, R.T., and Y-S. Wui. 2001. The Economic Value of Wetland Services: A Meta-Analysis.
Ecological Economics 37:157-270.
36
-------
Figure 1. Four Stage Questionnaire Design Process
Stage
Methods
Result
Stage 1:
Respondents'
Baseline
Knowledge
Focus Groups
(3 groups,
n=27)
and
Expert
Review of
Mitigation
Scenario
Alternative
Information
Treatments
Stage 2:
Evaluation
and Revision
• •
c
\
Group
Interviews
(3 groups,
n=21)
and
Individual
Interviews
(n=3)
• •
Questionnaire
Prototype
J
Stage 3:
Pretests
r \
Individual
Individual
Interviews,
o
'to
Interviews,
Debriefing
>
Debriefing
(n=29)
CCh
(n=28)
• •
Final
Questionnaire
Stage 4:
Internet
Functionality
Internet
Pretest
and
Telephone
Debriefings
(n=l 5)
\
y
• •
Large-Scale
Internet Survey
37
-------
Figure 2. Tabular Choice Format
Wetlands Scorecard #1
Mow do (lie I)i ninod and Restored Wetlands Compare?
Wo) I nil (I Choice #1
Wetland Features
Drained
Wetland
Restored
Wetland
Is il marsh, wooded, or a mix of march and woods?
Wooded
Mixed
1 low large is it?
14 acres
23 acres
Is il open to public?
Yes
No
Arc lIktc (rails and nature signs0
No
Xo
How good is the habitat lor dilTerenl species?
Amphibians and reptiles like frogs and turtles
IaccIIciii
—
Small animals like raccoon, opossum, and fox
(iood
Good
Songbirds like warblers, waxing, and vireo
—
Good
Wading birds like sandpiper, heron, or crane
—
Good
Wild flowers0
(iood
—
38
-------
Figure 3. Text Choice Format
Wetlands Scorecard #1
How do the Drained and Restored Wetlands Compare?
Wetland Choice #1
Drained Wetland
The drained wetland is 14 acres in size. It is a wooded wetland. It is open to the
public. It has no trails or nature signs. This wetland is excellent habitat for
amphibians. Small animals such as raccoon, opossum, and fox have good habitat in
this wetland. The habitat is poor for warblers, waxwing, vireo, and other songbirds.
It is poor habitat for wading birds such as cranes, heron, and sandpipers. The growing
conditions for wild flowers are good.
Restored Wetland
The restored wetland is 23 acres in size. It is a mix of marsh and wooded wetland.
It is not open to the public. It has no trails or nature signs. This wetland is poor
habitat for amphibians. Small animals such as racoon, opossum, and fox have good
habitat in this wetland. The habitat is good for warblers, waxwing, vireo, and other
songbirds. It is good habitat for wading birds such as cranes, heron, and sandpipers.
The growing conditions for wild flowers are poor.
39
-------
Table 1. Respondent Characteristics
Variable
Tabular
Text
Michigan,
Census 2000
Households
937
363
3.8 million
Income ($1,000)
54.4
54.1
57.4
Some college
79%
79%
52%
18 to 25 years
8%
8%
9%
Over 65 years
38%
47%
12%
Female
56%
60%
49%
Never visited a wetland
15%
15%
-
40
-------
Table 2. General Properties of the Tabular and Text Mitigation Equation Estimates
Variable Tabular3 Text3 Difference:
Text-Tabular
No of observations
4685
1811
2874
Correct predictions of yes responses (%)
63
64
-1
Correct predictions of no responses (%)
64
65
-1
Log-likelihood
-2814
-1151
--
Cross-sectional effects, standard deviation (oe)
22.3
49.8
27.5
(1.64)
(15.33)
(15.42)
Respondent effects, standard deviation (au)
17.0
40.1
23.1
(1.25)
(12.37)
(12.43)
a.. Asymptotic standard errors are given in parentheses.
41
-------
Table 3. Coefficient Estimates for the Tabular and Text Mitigation Equations
Variable
Tabular3
Text3
Difference:
Text-Tabular3
Acreage of drained wetland
1.42
0.99
-0.42
(0.19)
(0.68)
(0.70)
Change in public access
-5.76
-9.76
-3.99
(1.01)
(4.00)
(.412)
Change in wetland type
4.69
1.81
-2.86
(1.14)
(4.18)
(4.33)
Change in poor habitat
Reptiles/amphibians
8.19
23.46
15.3*
(1.17)
(8.10)
(8.98)
Wading birds
5.76
21.11
15.3 *
(1.14)
(7.53)
(7.62)
Song birds
6.56
21.33
14.8*
(1.16)
(7.12)
(7.21)
Wild flowers
2.33
12.51
10.2*
(1.14)
(5.41)
(5.53)
Change in excellent habitat
Reptiles/amphibians
-4.76
1.00
5.8*
(0.76)
(3.27)
(3.35)
Wading birds
-5.09
-1.12
4.0*
(0.74)
(3.19)
(3.28)
Song birds
-3.80
-1.76
2.0*
(0.76)
(3.10)
(3.19)
Wild flowers
-1.94
-3.44
-1.50*
(0.73)
(3.12)
(3.21)
Income ($ 1,000s)
-0.06
-0.03
0.03
(0.02)
(0.07)
(0.06)
Some college
-4.25
3.91
8.16
(1.83)
(7.49)
(7.71)
18 to 25 years of age
2.53
3.35
0.82
(2.70)
(9.62)
(10.0)
65 years of age and over
0.41
-3.29
-3.70
(3.17)
(12.65)
(13.05)
42
-------
Table 3. Coefficient Estimates for the Tabular and Text Mitigation Equations
Variable
Tabular3
Text3
Difference:
Text-Tabular3
Female
-2.9
0.59
3.50
(1.54)
(5.76)
(5.96)
Never visited a wetland
8.26
-0.54
-8.80
(2.14)
(7.92)
(8.21)
Intercept
4.75
6.68
1.93
(2.94)
(11.27)
(11.65)
a.. Asymptotic standard errors are given in parentheses. A indicates that the habitat quality coefficients are
significantly different from zero when evaluated as a group of coefficients.
43
-------
Discussion
Session VI: Methodological Advances
n Stated Preference Valuation
Joseph Cooper
Economic Research Service - USDA
US EPA NCEE Workshop
Valuation of Ecological Benefits: Improving the Science
Behind Policy Decisions"
Washington, DC Oct 26-27, 2004
• • a*
• • •
• • •
I. Specific Comments on:
• Experimental Tests of Provision Rules in Conjoint
Analysis for Environmental Valuation (Taylor et al.)
II A USDA Employee's Perspective on Methodological
Advances in Stated Preference Techniques
The views presented herein as those of the authors, and do not necessarily represent the views of the Economic
Research Service or the United States Department of Agriculture.
44
-------
I. Comments on Taylor et al.
At least two sources of incentive incompatibility associated
with conjoint surveys:
1) Incentive incompatibility due to exclusion of an explicit
provision rule
°This source of incentive incompatibility is the focus of Taylor
et al.
°This is also an issue with dichotomous choice reference
surveys
2) Incentive incompatibility of referendums with three or more
choices (or treatments)
• • •
• • •
I. Comments on Taylor et al.
With regards point 2),
Gibbard-Satterwait Theorem
An election mechanism for 3 or more alternatives
which is:
¦ Unanimous
¦ Strategy proof
is a dictatorship.
Other election methods are not incentive compatible
• • •
45
-------
I. Comments on Taylor et al.
General Question for Conjoint Analysis:
What is the potential response bias associated with 3
or more alternatives inherent in the voting
mechanism itself?
• • of
• • •
• • •
I. Comment on Hoehn et al.
Comparison of the tabular choice (fig. 2) to the text
choice format (fig. 3):
• The text choice format has quality rankings of
"Poor", "good", and "excellent."
• The tabular choice format has quality rankings of
"good", and "excellent."
• Substituting for "poor" in tabular choice format
would seem to limit comparability of the two formats.
• • •
• • •
46
-------
Ill, A Governmental Perspective
How stated preference techniques can be useful to
ERS
• Measuring the success of conservation programs
> e.g. Conservation Reserve Program (CRP), Environmental
Quality Incentives Program (EQIP), Conservation Security
Program (CSP)
• Monetizing the environmental impacts of commodity
programs
II. A Governmental Perspective
Example of relationships we want to measure:
Farmers' mimagemonl practice* affect ambierrt environmental quality...
Source: Smith and Weinberg # # #
• • •
47
-------
II. A Governmental Perspective
Another Example:
Agricultural Trade Liberalization
¦u
Change in world prices and quantity
u
Changes in production practices,
input use, and outputs
Changes in physical measures of Changes in returns to
environmental impacts agricultural production
u
Changes in economic measures of
environmental impacts
• • s*
• • •
• • •
II. A Governmental Perspective
In stylized fashion,
to evaluate an agr-environmental program, we want to
be able to approximate d V7dG
where 6V/6G = tiV/tiE ¦ dE/dF ¦ dF/dG
and where V = Nonmarket benefits, G = government
payments or expenditures, E = environmental
impacts, F = farm management practices (e.g.,
nitrogen application rates, etc).
• • •
• • •
48
-------
II. A Governmental Perspective
• But in general, CVM survey scenarios are designed
to produce discrete points in dWdG or d V/6E , e.g.,
V I G = $ level 1 anC' V I G = $ level 2
or
' I E= pollution ievel 1 ° ^ I E = pollution level 2
• There is probably little one can do to design a CVM
scenario that approximates d VltiG
• • a*
• • •
• • •
II. A Governmental Perspective
• Hence, the best path is likely to choose CVM
scenarios to maximize the possibility of achieving a
statistically significant relationship
V = f(E,F,G).
• This means choosing a benefit measure V (and
CVM scenario) that maximizes the possibility of
obtaining significant relationships between the V
and policy relevant variables £, F, and G.
• This suggests considering the available
environmental process models and data when
choosing V. ;;
49
-------
1)1, A Governmental Perspective
Examples of notable agri-environmental process models:
• USDA's Erosion-Productivity Impact Calculator (EPIC)
• USDA's Soil & Water Assessment Tool (SWAT)
• USGS's SPAtially Referenced Regressions On
Watershed Attributes (SPARROW)
II. A Governmental Perspective
With physical scientists, we need to stress that
• To make full use of environmental indicators to inform
decisions, the development and collection of these indicators
need to be coordinated and integrated with the development
and collection of behavioral data.
With economists, we need to stress that
- To make full use of behavorial data to inform decisions, the
development and collection of these indicators need to be
coordinated and integrated with the development and
collection of environmental indicators data.
• • •
• • •
50
-------
Section VI. Methodological Advances in Stated Preference Valuations
Discussion by T. H. Stevens
University of MA, Amherst
I. Embedding in Stated-Preference Methods (Michael Hanemann and Jeff Lazo)
Although I have been unable to obtain a copy of this paper, I found Hanemann's
presentation both stimulating and useful. Of particular importance is the notion that some types
of embedding, such as geographical scope effects, can often result from rational, well informed
decision-making. For example, many respondents may express the same value for preserving a
nearby wetland as they do for preserving all wetlands in a larger region simply because the
nearby wetland is the only one that is really important to them. Many other logical reasons for
scope effects were outlined in this presentation which suggests that (a) it is important to examine
psychological factors that might influence respondent's decision making and (b) debriefing
should be an important component of the stated preference methodology. Many important issues
remain to be addressed, including definition of the relevant choice set for valuation of public
goods.
II. Experimental Tests of Provision Rules in Conjoint Analyses for Environmental Valuation
(Taylor, Boyle, Morrison).
This paper focuses on a very important issue. Conjoint (choice) analyses is being used
widely, but little is known about potential biases that might be associated with this technique. In
particular, since provision rules are generally not incorporated within the conjoint format, this
method might produce inaccurate results.
The experiments involving hypothetical payments conducted by the authors suggest that:
1. Respondents were more likely to "purchase" a private good when the conjoint
question did not include an incentive compatible provision rule.
51
-------
2. Subjects were less responsive to the price of a public good when the conjoint
question did not include an incentive compatible provision rule.
Taken together, these findings imply that results derived from the traditional conjoint
approach (without provision rule) are likely biased upward.
Experiments involving real payments were also conducted, but these results were not
available prior to Taylor's presentation. However, the presentation seemed to suggest that
explicitly stated provision rules reduced hypothetical bias associated with the conjoint analyses.
If so, then it is very important to incorporate appropriate provision rules in conjoint analyses.
It is important to note, however, that comparisons were not made between an incentive
compatible CV format and an incentive compatible conjoint format. Such a comparison is
important because CV and conjoint techniques differ in several respects other than the provision
rule. That is, even if conjoint methods are modified to incorporate appropriate provision rules,
conjoint and CV results may still diverge because of other differences between these formats.
For example:
(1) Substitutes are made explicit in the conjoint (CJ) format and this may encourage
respondents to explore their preferences and tradeoffs in more detail. Indeed, as noted by Gan
and Luzar (1993), conjoint analysis 'can be characterized as an extension of the referendum
closed-end CV method in which large numbers of attributes and levels can be included in the
analysis without overwhelming the respondents' (p. 37). As shown by Boxall, et al. (1996),
when compared to CJ, CV results may therefore be biased upward because respondents to the
'typical' CV survey are usually asked to consider fewer substitutes.
(2) From a psychological perspective, the process of making choices in the CJ format
may be quite different from that associated with making decisions about WTP (Irwin, et al.,
52
-------
1993; McKenzie, 1993). That is, respondents may react differently when choosing among
commodities that have an assigned price as compared to making dollar valuations of the same
commodities. Moreover, Irwin, et al. (1993) found that CV questions lead to relatively greater
preference for improved commodities, such as TVs and VCRs, while choice questions yielded
relatively greater preference for environmental amenities like air quality. Similar results were
reported by Brown (1984). Irwin, et al. (1993) concluded that if monetary prices are an attribute,
they carry more weight in determining a response measured in dollars (e.g. CV) than they do in
determining a rating or choice response. This arises from the fact that choices seem to be driven
from reason and arguments to a greater extent than are pricing responses.
(3) CJ respondents can express ambivalence or indifference directly. As a result, CJ
surveys may result in relatively less non-response and protest behavior. Moreover, allowing for
respondent uncertainty may have a significant effect on the WTP of those who do respond. For
example, Ready, et al. (1995) compared a dichotomous choice CV format to a polychotomous
choice format. Their CV question asked respondents to determine whether or not they preferred
a given program while the polychotomous choice format gave six options (i.e., definitely prefer,
probably prefer, maybe prefer, maybe not prefer, probably not prefer, definitely not prefer). This
format was motivated by the belief that respondents might be more comfortable answering
valuation questions when given the opportunity to express strength of conviction; since the
polychotomous method allows for a range of answers, it might produce a more accurate
description of respondents' preferences. In two empirical studies, preservation of wetlands and
horse farms, the polychotomous format yielded a higher rate of usable responses and much
higher WTP estimates.
53
-------
More recently, Champ, et al. (1997) found that although contingent values were greater
than actual donations for an environmental good, when the contingent values were restricted to
respondents who said they were very certain to contribute, mean CV and actual donations were
not statistically different. Ekstrand and Loomis (1997), Alberini, et al. (1997) and Wang (1997)
also found that contingent value estimates vary widely depending on how respondent uncertainty
is incorporated in the analysis.
In summary, conjoint (choice) and CV formats differ in several ways, and correction for
provision rule may not resolve many of the differences between traditional CV and conjoint
estimates.
III. Stated-Choice Experiments to Estimate In-Kind Values for Ecosystem Mitigation (John
Hoehn, Frank Lupi, Michael Kaplowitz).
This paper addresses another very important issue—do respondents suffer from
information overload in the stated choice format, and if so, what are the consequences and what
can be done about this potential problem?
The authors use a split sample approach to compare text and tabular information formats
with the result that the tabular presentation was successful in reducing information complexity
and information overload.
Specific comments are as follows:
1. An internet survey that produced an 8 percent response rate was used in this
study. Much more information is needed with respect to non-respondents.
2. Another interesting research question would be whether the differences observed
in this study are also found in mail surveys.
54
-------
3. In this study, text respondents tended to exhibit loss aversion while tabular
respondents did not. But if loss aversion is part of "human behavior", elimination of loss
aversion might produce biased results. So, in this sense is a tabular format really "better" than a
text version?
55
-------
References
Alberini, A., Boyle, K., Welsh, M., 1997. Using multiple-bounded questions to incorporate
preference uncertainty in non-market valuation. In: Englin, J. (Compiler), Tenth Interim
Report, W-133 Benefits and Costs Transfer in Natural Resource Planning, University of
Reno, Reno, NV.
Boxall, P., Adamowicz, W., Swait, J., Williams, M., Laviere, J., 1996. A comparison of stated
preference methods for environmental valuation. Ecol. Econom. 18, 243-253.
Brown, T., 1984. The concept of value in resource allocation. Land Econom. 60,231-246.
Champ, P., Bishop, R., Brown, T., McCollum, D., 1997. Using donation mechanisms to value
nonuse benefits from public goods. J. Environ. Econom. Manag. 32 (2), 151-162.
Ekstrand, E., Loomis, J., 1997. Estimated willingness to pay for protecting critical habitat for
threatened and endangered fish with respondent uncertainty. In: Englin, J. (Compiler), Tenth
Interim Report, W-133 Benefits and Costs Transfer in Natural Resource Planning, University
of Reno, Reno, NV.
Gan, C., Luzar, E.J., 1993. A conjoint analysis of waterfowl hunting in Louisiana. J. Agric.
Appl. Econom. 25, 36-45.
Irwin, J.R., Slovic, P., Licktenstein, L., McClelland, G., 1993. Preference reversals and the
measurement of environmental values. J. Rick Uncertrain. 6,5-18.
McKenzie, J., 1993. A comparison of contingent preference models. Am. J. Agric. Econom.
75, 593-603.
Ready, R., Whitehead, J., Blomquist, G., 1995. Contingent valuation when respondents are
ambivalent. J. Environ. Econom. Manag., 29 (2), 181-197.
Wang, H., 1997. Treatment of don't know? Responses in contingent valuation survey: a random
valuation model. J. Environ. Econom. Manag. 32, 219-232.
56
-------
Summary of the Q&A Discussion Following Session VI
Elizabeth David (Stratus Consulting, Inc.)
Dr. David introduced herself by saying that prior to working for Stratus she worked for a
number of years with the State of Wisconsin's Department of Natural Resources, and she
stated that she would "feel much better" if she knew that John Hoehn had "familiarity
with questionnaire design." She cited the University of Michigan's Survey Research
Center within the Institute of Social Research, where there is a "whole section that
worries about how to design questions." Acknowledging that the wetlands problem is
particularly difficult and extremely complicated with "so many services associated with
it," she wished that Dr. Hoehn had "grounded his questions about wetlands in the existing
literature about how to deal with a very complicated set of interactions."
John Hoehn (Michigan State University)
Dr. Hoehn responded by stating, "We actually did work with a number of people from
the University of Michigan as consultants on the project. We certainly did try to access
that knowledge base."
Joel Huber, (Duke University)
Dr. Huber commented, "I wanted to talk to the issue of whether we're trying to get from
people the best answer or their first answer. There's a certain notion that there's true
utility up there and we need to ask them without them thinking about it too much and get
it out. I actually would have gone the other way. If you think about it, what you really
want to do is not get what they would do quickly but what they would do if they thought
about it. Because here we're talking about policy, and most of these are rich issues and
they're deep issues. So, I really applaud, John, what you're doing in terms of trying to
simplify and trying to actually test. . . . This can be sort of discouraging, because much of
the economics becomes harder to apply—the appearance of the answer depends on how
you ask the question, and knowing that puts you in a type of limbo because there are all
kinds of skills needed that you don't normally have. Typically in your work, though, you
have options of doing different versions, so I would suggest taking advantage of this.
You can have version A and version B and you don't necessarily need to mention it—if
they work out the same, great. What you're aiming for is what a person would do if they
thought about it a lot—it's quite different from what I would call the sort of implicit
utility."
Michael Kaplowitz (Michigan State University)
Dr. Kaplowitz said, "I just want to make one comment, because a lot of this discussion
the last two days has been on economics and ecology, and the work that John (Hoehn),
Frank (Lupi), and I are doing—and the work that I think many people here have done—is
really work that spans economics and survey research. For example, I never would have
thought that we would be publishing in survey research outlets, but our four-step design
process is now something that survey researchers are using or are thinking about using in
their work in other fields. So, I think there's a lot of crossover and lessons we can learn."
57
-------
When asked by John Hoehn what book was forthcoming on this topic, Dr. Kaplowitz
replied, "Stanley Presser and a bunch of people have a questionnaire development and
evaluation text being published by Wiley Press."
Elena Besedin (Abt Associates,Inc.)
Addressing Michael Hanemann, Ms. Besedin commented that Dr. Hanemann, in
reporting on his study, had mentioned an internal scope test in which the survey
participants indicated that they held widely disparate views as to how many birds would
represent a "significant" effect on bird population. She raised the question of whether
this really represents a scope issue "compared to, for example, a background information
issue." She noted that "sometimes scientists have difficulties measuring bird and other
wildlife populations." Ms. Besedin added that you "can't really judge whether an effect
is large or small" without having some idea of the total population figure. She concluded
by asking Dr. Hanemann whether this information (the total population figure) was
available to the focus group who gave the noted response, and if so, how would he
explain their conclusions.
Michael Hanemann (University of California-Berkeley)
Dr. Hanemann responded by saying that "the survey gave famously ambivalent
information, because it said: 2,000 birds or under 1%; 20,000 birds or about 1%; and
200,000 birds or under 2%. From one perspective, these are all speaking of 1 or 2%—
little to no difference. ... I interpret this as saying that, in fact, people were looking at the
percentages, and there's abundant literature from Slovik and others that indicates that
percentages are what people think of. The difference between under 1% and under 2% is
unimportant, and I think the attitude question about what this means to the population
suggests that they were looking at the percentages, and so they had a real basis.
Obviously, if they were looking at the numbers, that's striking, but it seems they were
focusing on the percentages, and the differences are unimportant."
John Hoehn
Dr. Hoehn referred to Joel Huber's comment that the decision they were trying to get at is
one that a person would make if they had a little more time to assimilate information and
to think about it. He said, "I think that is certainly a target. This format problem is
trying to make the assimilation task easier for respondents, so they don't have to spend as
much time on assimilation and can put more time into the decision and focus on that
problem. . . . It is a difficult problem, and some of the work we're doing is contributing to
that literature on survey design . . . because these are different sorts of questions than
asking, "Who are you going to vote for for president?"—and even there you see a lot of
variability these days. You know, we are asking people to make difficult choices when
we address wetlands issues—they're a distinct kind of problem in terms of survey design
and the issues they raise with respect to human cognitions."
58
-------
Michael Hanemann
Dr. Hanemann added, "There's a different strategy, in sociology at least, in attitude
measurement. The strategy in attitude measurement is to ask a large battery of questions
and then to reduce them. I see this relating to the Lancaster Model. If, by an attitude,
you mean something broad, such as patriotism or law-and-order, then it makes sense that
there's a large number of questions that would touch on that, and you could average
them. ... So, when you're measuring something very broad, then it's possible to have a
large number of imprecise measurements."
Joel Huber
Dr. Huber followed with these comments: "If you're trying to get attitudes, what you
want is quickness—that is, you want to see how a person reacts to a certain "picture."
That is often mediated by thought. ... So, some things are attitude questions, but the
tradeoffs are what I'd call "rational." They're very different modalities—one is fast and
the other is slow, and you're actually overcoming your initial thoughts. So, depending on
what you want to do, you'd go one way or the other."
END OF SESSION VI Q&A
59
------- |