February 1985
Valuing Changes in Hazardous Waste Risks:
A Contingent Valuation Analysis
Volume I
Draft Interim Report
EPA Cooperative Agreement No. CR-811075
The Benefits of Hazardous Waste Management
Regulations Using Contingent Valuation
Prepared for
Benefits Branch, Economic Analysis Division
G.S. Environmental Protection Agency
Washington, D.C. 20460
George Provenzano, Project Officer
Prepared by
V. Kerry Smith
Senior Principal Investigator
Vanderbilt University
Nashville, Tennessee
William H. Desvousges
Principal Investigator
Research Triangle Institute
Research Triangle Park, North Carolina
A. My rick Freeman, HI
Principal Investigator
Bowdoin College
Brunswick, Maine
RTI Project No. 410-2699
-------
PREFACE
The objective of this report is to provide a summary of the research com-
pleted during the first phase of U.S. Environmental Protection Agency (EPA)
Cooperative Agreement No. CR-311075-01, "The Benefits of Hazardous Waste
Management Regulations Using Contingent Valuation."
When this Cooperative Agreement was initiated August 8, 1983, several
activities related to the research were already underway with other EPA fund-
ing. Chief among these was a project to use and evaluate focus groups in
developing contingent valuation survey questionnaires for valuing reductions
in the risk of exposure to hazardous wastes. With the initiation of complemen-
tary research under the Cooperative Agreement, the scope of the focus group
analysis was expanded to meet the specific needs of the research under the
Cooperative Agreement. Thus, the report submitted in December 1984, The
Role of Focus Groups in Designing a Contingent Valuation Survey to Measure
the Benefits of Hazardous Waste Management Regulations, was a joint product
reflecting activities undertaken both under EPA Contract No. 68-01-6596 (Sub-
contract 700-C, Work Assignment No. C-011) and under the Cooperative Agree-
ment. A detailed summary of the focus group activities was also prepared for
more limited distribution under these two agreements.
This volume is the draft interim report for the Cooperative Agreement.
It summarizes the research activities during and the findings from Phase I of
the Agreement. Volume II is the appendix material to the report. In addition,
we have also provided a third volume to supplement this report. Volume III
contains the 11 working papers prepared by various authors over the course
of the research with the support of the Cooperative Agreement. Some of these
articles will soon appear in print, but we have collected them here to ensure
easier access. While the findings of most of these working papers have been
integrated into this report, the papers sometimes provide more detailed treat-
ments or more extensive reviews of particular issues. However, due to budget
limitations, we have been able to prepare only a few copies of Volume If I for
our EPA Project Officer. The reader who desires access to Volume III is asked
to contact him.
In preparing this draft report which involved the complex interaction of
several authors and participants, it was often difficult to give all contributors
the opportunity to review the entire report. Consequently, to limit the liabil-
ity of specific individuals, we have prepared Table I, which describes the writ-
ing responsibilities for each chapter in this report. As the ones responsible
for the overall research, we are of course the most culpable. Table I lists
three categories of contribution--primary responsibility, contributor, and
assistance. Primary responsibiiity implies the individual responsible for com-
pleting the first draft of the chapter, for assembling comments or proposed
iii
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
TABLE I. RESPONSIBILITIES FOR CHAPTERS OF DRAFT INTERIM REPORT
Primary responsibility
Contributor
Assistance
William H. Desvousges
A. Myrick Freeman III
V. Kerry Smith
V. Kerry Smith
V. Kerry Smith
A . Myrick Freeman 111
William H. Desvousges
William H. Desvousges
William H. Desvousges
Hall B. Ashmore
William H. Desvousges
William H. Desvousges
V. Kerry Smith
V. Kerry Smith
V. Kerry Smith
V. Kerry Smith
V. Kerry Smith,
William H. Desvousges
V. Kerry Smith
V. Kerry Smith
Hall B. Ashmore,
Diane H. Brown,
V, Kerry Smith
V. Kerry Smith,
Bruce Jones
William H. Desvousges, 1
V. Kerry Smith
V. Kerry Smith
William H. Desvousges,
Matthew McGivney
A, Myrick Freeman II!
Lu Lohr
V. Kerry Smith
William H. Desvousges
William H. Desvousges,
Lu Lohr
V. Kerry Smith
-------
changes from other contributors, and for developing the draft provided in this
report, Contributor designates coauthor status achieved either through design
of the research or through involvement in key specific research activities.
Assistance implies that an individual provided key information, editorial sug-
gestions, and research assistance in the activities associated with a chapter.
With those chores behind us, we can now turn to our most important
task--that is, extolling the many who contributed to our research effort.
First, we would like to acknowledge the important role of our coauthor Rick
Freeman, Although Rick was primarily responsible for the conceptual analysis
of intrinsic benefits, he provided valuable comments throughout the research.
He also attended our interviewer debriefing sessions to help us interpret the
information gained in these sessions.
We would also like to point out the valuable roles of a number of EPA
personnel in our research activities, including structuring the research objec-
tives, commenting on the questionnaire design, reacting to the proposed design
of the empirical analysis, and attending the interviewer debriefing sessions.
George Provenzano, the EPA Project Officer, performed all of these tasks and
put up with Kerry's grumblings over administrative details of the project.
Ann Fisher also contributed substantively to the effort through her initiation
and supervision of the focus group project and in commenting on all aspects
of the research activities undertaken under the Cooperative Agreement. Along
with our field interviewers, we especially appreciated Ann's participation in
our training session. Also helpful were several members of the Office of Solid
Waste. Dale Ruther provided important guidance at a crucial stage; Peggy
Podolak, formerly of the Office, improved the questionnaire in several key
areas; and Jim Craig has continued the liaison with the Office.
We were also fortunate to receive comments on the development of the
questionnaire from a large number of individuals. Table II lists these brave
individuals, to whom we are grateful. We would like to note the contributions
of two individuals in particular: Robert Mitchell and Thomas Wallsten. Robert
provided two thorough reviews on short notice and helped us avoid several
potential problems. Tom helped in a number of ways. His thoughtful sugges-
tions and insights on questionnaire development and on the psychological litera-
ture on individual behavior under uncertainty were especially important to the
research design. In addition, he served as a member of the Research Advisory
Committee for the project and provided many helpful ideas as part of that
group as well.
The project was especially fortunate to have the guidance of an excellent.
Advisory Committee who assisted us at several key stages of the research de-
sign (and who, ultimately, will be our toughest and most helpful reviewers of
this draft report). In alphabetical order, they are as follows:
Jerry Hausman, Massachusetts Institute of Technology
Robert Haveman, University of Wisconsin
Milton Weinstein, Harvard University
Thomas Wallsten, University of North Carolina at Chapel Hill
v
-------
TABLE [I, QUESTIONNAIRE REVIEWERS AND KEY CONCERNS
Name Affiliation Concern* Version(s) reviewed
Tom WalIston
university of North Carolina
t Psychologist- - Ad vise ""y
Commttte® Member!
• Context of risk.
¦ Design issues
• Risk laddar
Several
Sill Sclwlz*
University of Wyoming
(Economist)
• Direct question v&. bidding
• Probability complexity
• Payment v@hicl«s
September 1383
Robert Mitchell
Resources for the futur#
(Sociologist)
• Context
• Probability complexity
¦ Equity
• Probability design
¦ Risk laddar
September 1983
November 1983
and
February 1984
Milt Weiosteln
Harvard University
( Economist--Advisory
Committee Memb«r)
• Analytical design
• Certainty case
¦ Context
September 1983
Alan Randall and
John Hothft
Univtrsity of Kentucky
(economists)
¦ Bidding
¦ Context
¦ Analytical design
November 1983
George Tolley at #1.
university of ChicaQO
(Economist®)
- Length
» Complexity
¦ Bidding
September 1983
Bob Haveman
University of Wisconsin
(Economist Advisory
CommittM Member)
• Analytical design
• Length
September 1983
Nancy Bockstael
University of Maryland
(Economist)
¦ Context
• Analytical design
September 1983
Dick Kulka
Research Triangle Institute
(Psychologist)
• Length
• Context
Several
Garri# Kingsbury
RTi CCtiemical £ngir>«#r)
* Health affacts--laconical issues
February 1984
David Harrison
Harvard University
f Economist)
* Averting cost
• Hypothetical vs. actual
September 1983
Ron Wyzga
Electric Power Research
institute
-------
The Advisory Committee provided detailed reviews of our proposed question-
naire and critical evaluations of our analysis plan for empirical analysis of the
survey data. Many of their comments sparked ideas that are discussed
throughout the report.
We would also like to thank David Harrison for his contribution. David
reviewed several drafts of the questionnaire and attended some of our focus
group sessions in Boston. Along with James Stock, David also gave us back-
ground and data for his property value analysis that enabled us to do our
comparative analysis in Chapter 15.
Several individuals at Vanderbilt played key support roles. It would not
have been possible for Kerry to complete his work without the continuous
assistance of John Mott and Wei-Wei Kao in helping him to learn and use a new
IBM computer facility introduced at Vanderbilt in September 1984. Long week-
ends and late nights by John at crucial times assured the project would have
the needed computer resources.
The day-to-day administration of the project, budget management, monthiy
and quarterly reports, drafts of chapters, comments and plans, and all of the
correspondence from Vanderbilt would not have been possible without Sue
Piontek. Because Sue handled all of these aspects of the project so well,
Kerry was able to focus primarily on research administration. Steve Smartt
of the Office of Sponsored Research at Vanderbilt also contributed in a signifi-
cant way to ease these administrative burdens.
Several people at RTI assisted us in conducting our research. The qual-
ity of the contingent valuation data is due in large part to Kirk Pate, RTI
Survey Specialist, who worked with us in every aspect of the focus group ses-
sions and in developing the questionnaire. In addition, he conducted the vid-
eotaped interviews, coauthored the interviewer training manual, developed the
overall survey plan, and conducted the interviewer training sessions. Kirk
also supervised the activities of all the interviewers and the assembly of the
questionnaires. Kirk was assisted in these tasks by Annette Born, who super-
vised the day-to-day activities in Boston along with helping in the pretest.
We are grateful to Matthew McGivney of RTI who constructed the SAS
data set and helped perform the means and regression analysis reported in
Chapters 11 and 12 and the contingent ranking analysis reported in Chapter
14. Matt also helped transfer the data for Harrison's hedonic model to SAS
data sets. Glenn Jones of Vanderbilt University also helped in this second
task. David Toy of RTI assisted in the results presented in Chapters 11 and
12, and Lu Lohr of RTI helped by constructing the Census data needed for
the comparative analysts and by organizing the detailed background information
on the survey area.
While the preface to the focus group report identifies the roles that vari-
ous people played in those activities, we would like to add a few special com-
mendations in this report because of the importance of these activities. As
we noted earlier, Kirk Pate's efforts as our unflappable moderator were most
valuable. He always knew at the end of the sessions that we still needed a
-------
survey questionnaire. In addition, Diane Brown, formerly of RTI, now work-
ing for the Power Plant Siting Commission of the State of Maryland, almost
singlehandledly summarized those sessions, helped organize them, and generally
provided good counsel on many issues. Ann Dunson, who has left RTI to start
her own business, helped organize most of our sessions. Finally, we would
like to thank the participants of the sessions who helped us begin to under-
stand how to deal with risk in a survey questionnaire.
We also appreciate both the continuous support and valued counsel of
Tayler Bingham, Head of RTI's Environmental Economics Department. While
Tayler often keeps himself in the background, his help is always highly val-
ued.
Hall Ashmore, Publications Manager In RTI's Center for Economics Re-
search, is primarily responsible for the level of communication, consistency,
and overall form of this report—especially the visual aesthetics of our figures
and tables. Hall has helped to make every chapter more readable and to en-
sure that all the chapters work together in the overall report. We would also
like to thank Hall for his assistance in writing Chapters 3 and 10.
Last, but certainly not least, in our appreciation is Jan Shirley and her
staff of word processing specialists. In working with us over the past 3%
years, they have consistently turned the impossible into the possible. Each
time the scale and complexity of this effort increased, their response grew to
meet it. They continue to be a most valuable part of our research team.
In a project involving multiple locations and almost 2 years of activities,
we would have been scuttled without the help that these many individuals have
given us. Not only these individuals but their families have contributed by
their patience and support when faced with another long working weekend.
In this regard, our wives, Pauline and Shelley, and our children, Timothy,
Shelley, and Anne, have contributed dearly.
We can say without any reservations that we could not have reached this
point in our research without each and every one of you. Had any of the
links in this long and winding human chain failed, we would have been lost.
Thank you.
V. Kerry Smith
Nashville, Tennessee
William H. Desvousges
Research Triangle Park, North Carolina
viii
-------
CONTENTS
Chapter Page
Preface iii
Figures - xv
Tables xix
1 On Valuing Reductions in Hazardous Waste Risks , , , . , 1-1
1.1 Introduction 1-1
1.2 The Importance of Hazardous Waste Regulatory
Policy 1-3
1.3 The Role of Benefits Analysis for Regulatory
Policy ....................... 1-4
1.4 A Central Theme: The Role of Risk.. ....... 1-6
1.5 Risks and Household Decisions ........... 1-9
1.6 Hazardous Waste Risk: Our Framework. ...... 1-12
1.7 Contingent Valuation: A Brief Overview ...... 1-13
1.8 Research Overview 1-16
1.9 Guide to the Report, 1-13
Part I A Conceptual Framework for Measuring the Benefits
of Reducing Hazardous Waste Risks 1-1
2 The Nature of Benefit Analysis in Hazardous Waste
Management 2-1
2.1 Introduction 2-1
2.2 Conventional Benefit Taxonomies. .......... 2-1
2.3 The Treatment of Policy Outputs as Risk Changes. . 2-8
2.4 What is Risk? .................... 2-8
2.5 The Sources of Exposure Risk. ........... 2-10
2.5.1 Six Categories of Functional and Chemical
Characteristics 2-11
2.5.2 Three Exposure Scenarios .......... 2-13
2.6 Summary 2-13
3 Modeling Behavior Under Uncertainty: A Heuristic
Review 3-1
3.1 Introduction 3-1
3.2 The Expected Utility Framework and Contingent
Claims 3-3
3.3 Risk Aversion and Probability Change: State-
Independent Utility Functions ............ 3-10
3.4 The Implications of State-Dependent Utility
Functions ...................... 3-19
ix
-------
CONTENTS (continued)
Chapter Page
3.5 The Performance of the Expected Utility Framework, . 3-29
3.8 Summary . . . 3-33
4 The Role of the Ex Ante and Ex Post Perspectives in
Measuring Welfare Changes Under Uncertainty 4-1
4.1 Introduction 4-1
4.2 Background . . - 4-1
4.3 Ex Ante Versus Ex Post Perspectives ........ 4-8
4.4 User and Intrinsic Values in an Ex Ante Framework;
An Introduction 4-18
4.5 Summary 4-18
5 A Conceptual Framework for Valuing Risk Reductions . . . 5-1
5.1 Introduction .................... 5-1
5.2 Valuing Risk Changes ................ 5-3
5.3 Implementing the Theory; Psychological
Considerations ................... 5-18
5.4 Summary 5-20
6 Ecological and Intrinsic Values Under Uncertainty ..... 8-1
8.1 Introduction .................... 8-1
6.2 Existence and Use Values Under Conditions of
Certainty ...................... 6-2
6.3 Uncertainty of Existence .............. 6-12
6.4 Conclusions ..................... 6-20
Part II Research Design, Questionnaire Development, and the
Survey .......................... 11-1
7 Research Design: The Transition from Theory to
Practice 7-1
7.1 Introduction 7-1
7.2 Guide to the Chapter ................ 7-2
7.3 Overview ...................... 7-3
7.4 User and Intrinsic Values 7-3
7.4.1 Measurement Concerns. ........... 7-5
7.4.2 Sequence Effects and Intrinsic Values .... 7-7
7.5 The Effect of Risk Levels and Changes ....... 7-10
7.5.1 Risk Levels ................. 7-10
7.5.2 The Size of the Risk Change and the Role
of the Conditional Risk ........... 7-13
7.6 Property Rights and Risk Valuation ......... 7-14
7.7 Types of Risks and Risk Attributes ......... 7-18
7.7.1 Types of Risks 7-17
7.7.2 Risk Attributes ............... 7-18
7.7.3 The Role of Differences in the Types of
Risk for the Research Design 7-19
x
-------
CONTENTS (continued)
Chapter . Page
7.8 Context of Risk 7-20
7.9 Risk Outcomes and Endpoints . , . , 7-21
7.10 Contingent Valuation and Eliciting Values of Risks . 7-22
7.10.1 Question Format. 7-23
7.10.2 Perceptions ......... 7-30
7.10.3 The Role of Information ........... 7-32
7.11 The Design for Comparison with indirect Methods . . 7-33
7.12 Research Design; Its integration 7-33
7.13 Implications 7-40
8' Survey Questionnaire Development............. 8-1
8.1 Introduction 8-1
8.2 Overview; A Brief Chronology ........... 8-2
8.3 focus Groups: The Basic Ingredients. ....... 8-3
8.4 Focus Groups: Their Role in Contingent Valuation. . 8-4
8.5 Focus Groups; Their Organization, 8-8
8.6 Focus Group Participants: Their Awareness of the
Hazardous Waste Problem .............. 8-8
8.7 Focus Groups and Questionnaire Development:
Overview and Summary Findings. .......... 8-12
8.7.1 Overview: Findings and Issues in
Questionnaire Development. ......... 8-12
8.7.2 Presentation of Probability 8-14
8.7.3 Perception of Exposure Risk. ........ 8-27
8.7.4 Summary .................. 8-36
8.8 Pretest of Contingent Valuation Survey
Questionnaire .................... 8-38
8.9 Videotaped Interviews 8-43
8.10 The Questionnaire Development Process;
Reflections and Suggestions for Improvement .... 8-46
3 Sampling Plan and Survey Procedures, .......... 3-1
3.1 Introduction .................... 9-1
8.2 The Target Population. ............... 9-1
9.3 The Sampling Plan ................. 9-3
9.3.1 Overview .................. 9-3
9.3.2 Experimental Design Considerations ..... 9-4
3.4 Survey Administration 9-8
9.4.1 Interviewer Training ............ 9-8
9.4.2 Quality Control Procedures ......... 9-11
9.4.3 Data Collection Summary. .......... 9-12
9.4.4 Comparison with Other Contingent
Valuation Studies .............. 9-16
9.4.5 Interviewer Debriefing .- ......... . 9-20
9.5 Summary . . . 9-22
xi
-------
CONTENTS (continued)
Chapter Page
10 Profile: The Survey Area and Its Population ....... 10-1
10.1 Introduction 10-1
10.2 Guide to the Chapter 10-2
10.3 The Survey Area and Population ... ....... 10-2
10.3.1 The Survey Area . 10-2
10.3.2 Socioeconomic Profile , 10-6
10.4 A Survey Focus; Hazardous Wastes in Acton .... 10-9
10.5 Respondent Knowledge and Perceptions of -
Hazardous Wastes and Their Risks. ......... 10-16
10.5.1 Respondent Knowledge ........... 10-17
10.5.2 Respondent Perceptions ........... 10-17
10.5.3 Respondent Awareness of Risk ....... 10-23
10.5.4 Respondent Actions to Reduce Risks. , . , , 10-28
10.6 Summary ...................... 10-32
Part III Preliminary Empirical Analysis .............. 111-1
11 Option Price Results; The Framing of the Commodity
and an Analysis of Means 11-1
11.1 Introduction 11-1
11.2 Guide to the Chapter 11-2
11.3 Framing and Contingent Valuation .......... 11-2
11.3.1 Alternative Bias Classifications ...... 11-3
11.3.2 Context 11-8
11.4 Framing the Commodity: Reductions in Hazardous
Waste Risks ..................... 11-11
11.4.1 Conceptual Linkages. ............ 11-12
11.4.2 Context 11-14
11.4.3 Contingent Commodity Specification ..... 11-18
11.4.4 Elicitation of the Option Price Bids .... 11-28
11.5 Protest Bidders .............. 11-28
11.6 Option Price Bids for Risk Decreases 11-40
11.7 Property Rights and Values of Changes in Risk. . „ 11-51
11.7.1 Framing of the Property Rights Question . . 11-51
11.7.2 Option Price Bids for Risk Decreases .... 11-54
11.S The Certainty Effect 11-59
11.9 Risk Outcomes 11-82
11.10 intrinsic Values 11-64
11.11 Implications 11-89
12 Option Price Results: Preliminary Regression Analyses
Using Unrestricted Models 12-1
12.1 Introduction 12-1
12.2 Guide to the Chapter 12-1
12.3 Simple Models 12-2
12.3.1 The Model's Rationale 12-2
12.3.2 The Model 12-3
xii
-------
CONTENTS (continued)
Chapter page
12.4 Avenues for Adjusting to Risk 12-4
12.5 Health Status , . . 12-7
12.5.1 The Role of Health 12-8
12.5.2 Health Analysis Variables .......... 12-11
12.8 Regression Results 12-13
12.8.1 Option Price Results for Risk Decreases . . . 12-14
12.8.2 Option Price Results for Avoiding Bisk
Increases 12-1?
12,7 Implications 12-21
13 Valuation Estimates for Risk Reductions: Restricted
Models 13-1
13.1 Introduction 13-1
13.2 Guide to the Chapter , . 13-1
13.3 Overview 13-2
13.4 Econometric Qualifications to the Use of Restricted
Models 13-5
13.4.1 Nan protest Zeros 13-5
13.4.2 Functional Form 13-?
13.4.3 Missing Observations , . . „ 13-8
13.4.4 Pooied Samples 13-9
13.5 Restricted Models 13-10
13.5.1 The Model. ................. 13-10
13.5.2 Estimates for Marginal Valuation of
Exposure Risk Reduction .......... 13-14
13.6 Pooled Responses of Risk Change Values 13-21
13.7 Restricted Models; Pooled Sample and GLS
Estimates 13-24
13.8 Modeling the Payment to Avoid a Risk Increase . , , 13-31
13.9 Influential Observations, The Role of Judgment in
Sample Selection and Thick-Tailed Distributions , , . 13-34
13.9.1 Past Practices in Screening Contingent
Valuation Responses for Outliers ...... 13-35
13.9.2 Judging Influential Observations for the
Present Study. 13-38
13.9.3 Some Preliminary Tests for Thick-Tailed
Distributions . . . ' 3-39
13.10 Implications ' 3-44
14 The Use of Contingent Ranking Models to Value
Exposure Risk Reductions: Preliminary Results ...... 14-1
14.1 Introduction 14-1
14.2 The Random Utility Model . . 14-5
14.3 Structure of the Contingent Ranking Questions
and Experimental Design 14-10
14.4 Empirical Findings 14-19
xiii
-------
CONTENTS (continued)
Chapter Page
14.4.1 An Overview of the Nature of the
Rankings Provided 14-20
14.4.2 Some Simple Tests for the Effects of
Exposure and Payment Vectors ....... 14-22
14.4.3 Preliminary Estimates of the Random
Utility Models ............... 14-26
14.5 Comparing Contingent Valuation and Contingent
Ranking Estimates of the Value of Risk Reductions. , 14-34
14.6 Summary ...................... 14-49
15 A. Comparison of Contingent Valuation and Hedonic
Property Value Models for Risk Avoidance ......... 15-1
15.1 Introduction 15-1
15.2 Guide to the Chapter 15-2
15.3 The Role of Judgment in Comparative Studies .... 15-3
15.4 Conceptual Dimensions of the Distance-Risk
Relationship .................... 15-7
15.5 Alternative Avenues for Comparison ......... 15-9
15.5.1 Eliciting the Distances Considered to be
Required for Risk Reductions ........ 15-9
15.5.2 Eliciting Demand-for-Distance Information . . 15-13
15.6 Structure of the Questions and Design for
Comparative Information ............... 15-13
15.7 The Harrison Hedonic Property Value Model. .... 15-17
15.8 Risk Change and Distance .............. 15-25
15.9 The Demand for Distance for Risk Avoidance .... 15-32
15.10 A Comparative Evaluation of the Contingent
Valuation and Hedonic Models ............ 15-35
15.11 Summary ...................... 15-48
16 Policy Implications and Research Agenda. 16-1
16.1 Introduction 16-1
16.2 Guide to the Chapter 16-2
16.3 The Incremental Values for Risk Changes Used
in Policy Analyses ................. 16-2
16.4 Further Research .................. 16-9
16.4.1 Analysis of the Contingent Valuation
Responses. 16-10
16.4.2 Analysis of the Contingent Ranking
Responses. ................. 16-13
16.4.3 Comparison of Hedonic and Contingent
Valuation Responses ............. 16-14
16.4.4 Analysis of Wage Compensation for Job
Related Risks ................ 16-15
16.4.5 The Determinants of Risk Perceptions .... 16-16
16.5 Summary ...................... 16-17
17 References ........................ 17-1
xiv
-------
FIGURES
Number Page
1-1 Number of establishments generating each major waste
group 1-5
1-2 Effects and responses to a hazardous waste regulatory
action 1-8
1-3 Multimarket framework, 1-11
1-4 Study overview 1-17
2-1 The conventional benefits taxonomy adopted for
hazardous waste management 2-3
3-1 Illustration of Von Neumann-Morgenstern indifference
curve 3-8
3-2 Illustration of Arrow-Pratt measure of risk aversion . . . 3-13
3-3 illustration of change in probability for Von Neumann-
Morgenstern indifference curve 3-16
3-4 Measuring risk aversion in absence of actuarially
fair markets 3-18
3-5 The distinction between income certainty and utility
certainty loci with state-dependent preferences ....... 3-23
3-6 Illustration of Kami's measure of risk aversion with
state-dependent preferences « . , . . 3-27
4-1 Graham willingness-to-pay locus and option value ..... 4-14
5-1 Graham locus with change in probabilities ......... 5-3
8-1 Option price and expected use value with risk aversion . . 6-17
7-1 Overview of the origins of the research design for
valuing reductions in hazardous waste risks ........ 7-4
7-2 Effects of instrument--distribution of option price for a
change In water quality from beatable to fishable,
protest bids excluded . 7-27
xv
-------
FIGURES (continued)
Number Page
7-3 Block diagram of experimental design 7-35
7-4 Overview of questionnaire versions by experimental
design component and question format 7-36
8-1 Sample questions used in focus group discussions ..... 8-7
8-2 Probability circles with various combinations for risk of
exposure and effect .................... 8-17
8-3 Card in tabular form to present probability and explain
simple risks . 8-19
8-4 Circles used for probability presentation ......... 8-21
8-5 Sample cards A through C used for probability
presentation 8-24
8-8 Two cards (A-l and C-1) with final format ........ 8-28
8-7 Initial risk ladder including exposure to three kinds of
hazardous waste among risks from other events ...... 8-29
8-8 Revised risk ladder separating occupational risks from
other events and introducing breaks in ladder. ...... 8-31
8-9 Card attempting to tie risk ladder to probability circles . 8-33
8-10 Final version of risk ladder incorporating suggestions
from participants ..................... 8-37
9-1 Map of survey area 3-2
3-2 Matrix, of planned and actual observations for each cell
of the experimental design 9-6
9-3 Interviewer training agenda 3-9
10-1 Survey area. ....................... 10-3
10-2 Final version of risk ladder incorporating suggestions
from participants ..................... 10-26
11-1 Classifications of potential biases in contingent
valuation ......................... 11-4
11-2 Hazardous waste information card ............. 11-17
xvi
-------
FIGURES (continued)
Number Page
11-3 Risk circles 11-20
11-4 Payment vehicle card ................... 11-21
11-5 Description of hypothetical situation ........... 11-25
11-8 Direct question version of research design for valuing
reduction in risk . . 11-41
11-7 Mean option price bids for initial change in risk
exposure 11-44
11-8 Mean option price bids for second change in the risk
of exposure 11-45
11-9 Overview of questionnaire versions by experimental
design component and question format ........... 11-82
12-1 Potential avenues for adjusting to risk exposure ..... 12-5
12-2 Overview of health information .............. 12-9
13-1 Examples of low probability risk cards: design point 7 . . 13-18
13-2 Examples of low probability risk cards: design point 8 . . 13-19
14-1 Modified risk cards for contingent ranking questions-
design point R-1 ..................... 14-13
14-2 Outline of the design for the contingent ranking
component of the survey ................. 14-14
14-3 Description of a hypothetical situation .......... 14-18
15-1 Map of survey area showing industrial sites with landfills
containing hazardous wastes ................ 15-22
xvii
-------
TABLES
Number Page
1-1 Summary of the Objectives and Structure of the
Research 1-20
2-1 Scenario A; Growing Water Contamination and Human
Exposure ......................... 2-14
2-2 Scenario B; Long-Term Effects on Aquatic Ecosystems . . 2-16
2-3 Scenario C; Short-Term Acute Ecological Impacts . , , , , 2-18
7-1 Starting Point Bias: The Results 7-25
7-2 Willingness-to-Pay for Beatable Water Quality ....... 7-29
8-1 Focus Group Summary. . . 8-6
8-2 Focus Group Profile: Participant Awareness of the
Hazardous Waste Problem 8-9
3-1 Sample Sizes 3-7
9-2 Enumeration Results 9-13
9-3 Interview Results „ . . 9-14
9-4 Enumeration Results--Mitchell-Carson [ 1984] ........ 9-17
3-5 Interview Results--Mitchell-Carson [1984] 9-19
10-1 Composition of 1977 Civilian Labor Force for Selected
U.S. Cities , 10-5
10-2 News Summary: Greater Boston Area Communities
Experiencing Problems with Hazardous Wastes ....... 10-7
10-3 Characteristics of the Target Population and the Sample . . 10-8
10-4 Summary of Major Environmental Pollution and Hazardous
Waste Contamination Episodes in Acton. .......... 10-11
10-5 Contaminants found in Acton Water Supply and Their
Potential Health Effects 10-13
xix
-------
TABLES (continued)
Number Page
10-6 Number and Percentage of Total Respondents Who Had
Recently Read or Heard About Hazardous Wastes. ..... 10-18
10-7 Frequency with Which Respondents Had Recently Read
or Heard About Hazardous Wastes. 10-18
10-8 Subject of Hazardous Waste Information Recently -Read
or Heard by Respondents. 10-19
10-9 Respondent Harmfulness Ratings of Environmental
Pollution Sources ..................... 10-21
10-10 Respondent Effectiveness Ratings of Organizations that
Deal with Hazardous Wastes ................ 10-22
10-11 Respondent-Related Annual Risks of Death from Selected
Causes Using Risk Ladder. ................ 10-24
10-12 Respondent-Rated Changes of Exposure Through Typical
Exposure Pathways .................... 10-27
10-13 Number and Percentage of Total Respondents with
Particular Cause of Death in Mind, for Willingness-to-
Pay Bid .......................... 10-23
10-14 Current Respondent Actions to Reduce Risk of Exposure
to Hazardous Wastes, 10-29
10-15 Total Respondent Expenditures on Water Filters During
Last 5 Years ....................... 10-30
10-16 Total Respondent Expenditures on Bottled Water During
Last 5 Years ....................... 10-30
10-17 Total Public Meetings Attended by Respondents During
Last 5 Years 10-31
11-1 The Introduction to Hazardous Waste as a Risk. 11-19
11-2 Profile of Respondents with Revised Bids 11-29
11-3 Frequency of Reasons for Zero Bids by Level of Risk,
Decrease Direct Question Format. 11-31
11-4 Frequency of Reasons for Zero Bids by Level of Risk,
Increase Direct and Ranking Question Formats ....... 11-34
XX
-------
TABLES (continued)
Number
11-5 Profile of Protest Bidders Direct and Ranking Question
Format, Outliers Included ....... . . . 11-35
11-8 Logit Models for Protest Versus Nonprotest Bidders
for Risk Decreases and Risk Increases 11-39
11-7 Summary Statistics of Option Price Bids for Risk
Reductions, Direct Question Format, Protest Bids
Included and Excluded, Outliers I ncluded ........ 11-42
11-8 Mean Willingness to Pay for First Risk Reduction
Relative to Mean Willingness to Pay for Second ...... 11-4?
11-3 Analysis of Variance of Option Price Bids for Risk
Reductions, Protest Bids Excluded, Outliers Included . . . 11-49
11-10 Summary Statistics of Option Price Bids for Risk
Reductions, Direct Question Format, Protest Bids
Included and Excluded, Outliers Included ......... 11-50
11-11 Hypothetical Situation for Risk Increase ......... 11-53
11-12 Summary Statistics of Option Price Bids to Avoid Risk
Increases, Direct and Ranking Question Format, Protest
Bids Included and Excluded, Outliers Included ...... 11-58
11-13 Summary Statistics of Option Price Bids to Avoid Risk
Increases, Direct and Ranking Question Format, Town
Council Responsibility Versus Government Responsibility
for Risk Changes, Protest Bids Excluded, Outliers
Included 11-57
11-14 Summary Statistics of Option Price Bids for Changes
in Risk, Risk Decreases Compared to Risk Increases,
Direct Question Format, Protest Bids Excluded,
Outliers Included 11-58
11-15 Summary Statistics of Option Price Bids for Risk
Reductions to Zero, Protest Bids Included,
Outliers Included . . 11-61
11-16 Summary Statistics of the Change in Option Price
Bids Given a Specific Illness, Protest Bids
Excluded, Outliers Included 11-83
11-17 Summary Statistics of Option Price Bids for a Risk
Decrease Given a Specific Illness, Protest Bids
Excluded, Outliers Included 11-85
xxi
-------
TABLES (continued)
Number Page
11-18 Summary Statistics of Option Price Bids for Intrinsic
Values (Risk Reductions to Critters), All Bidders , . , , , 11-68
12-1 Self-Assessed Health Status 12-10
12-2 Quantitative indicators of Health Status ......... 12-10
12-3 Incidence of Six Diseases Among Sample Respondents . . , 12-12
12-4 Models for Option Prices for Risk Reductions; Common
Sample 12-15
12-5 Definition of Variables 12-18
12-6 Model for Option Prices for Exposure Risk Reduction--
Common Sample . 12-18
12-7 Models for Option Prices for Avoiding Risk Increases . . , 12-20
13-1 Models for Marginal Valuation of Exposure Risk
Reductions: Common Sample 13-15
13-2 Definition of Variables 13-18
13-3 Models for Marginal Valuation of Exposure Risk Reduction:
Specific Sample 13-22
13-4 Ordinary Least-Squares Estimates of Marginal Valuation
Models: Pooled Sample 13-25
13-5 Generalized Least-Squares Estimates of Marginal
Valuation Models 13-28
13-6 Marginal Valuation to Avoid Risk Increases 13-33
13-7 Tests for Thick-Tailed Distributions with Levels of
Valuation Bids 13-42
13-8 Tests for Thick-Tailed Distributions with Transformed
Estimates of Marginal Valuations 13-42
14-1 Frequency of Card Chosen First by Version. ....... 14-21
14-2 Ranking Permutations Chosen, by Version, ........ 14-23
14-3 Tests Concerning the Independence of Version
Administered and Card Chosen First. 14-24
xxii
-------
TABLES (continued)
Number Page
14-4 Tests Concerning the Underlying Distribution of Card
Chosen First, by Version 14-25
14-5 Baste Model for the Random Utility Model with the
Ranked Logit Estimator, by Version--Interaction
with Payment Measure 14-28
14-6 Basic Model for the Random Utility Model with the
Ranked Logit Estimator, by Version--Interaction
with Exposure Risk Measures 14-29
14-7 Basic Model for the Random Utility Model with the
Ranked Logit Estimator, Using the Full Sample ...... 14-31
14-8 Estimated Conditional Probability of a Payment-Exposure
Risk Combination Ranking as First 14-33
14-9 Selected Results for the Random Utility Model with
the Ranked Logit Estimator by Version .......... 14-35
14-10 Selected Results for the Random Utility Model with the
Ranked Logit Estimator by Version ............ 14-36
14-11 Selected Results for the Random Utility Model with the
Ranked Logit Estimator by Version 14-37
14-12 Selected Results for the Random Utility Model with the
Ranked Logit Estimator by Version ............ 14-38
14-13 Selected Results for the Random Utility Model with the
Ranked Logit Estimator, Full Sample ............ 14-33
14-14 Average Valuation of Exposure Risk Change by Town . . . 14-44
14-15 Average Valuation Estimates for Risk Reductions Using
Contingent Valuation and Contingent Ranking Estimates , . 14-45
14-18 Comparison of Contingent Valuation and Contingent
Ranking Valuation Estimates Using Individual Responses , . 14-48
15-1 Results of the Revised Desvousges, Smith, and
McGivney Comparative Study. 15-5
15-2 Description of Selected Variables in Harrison Property
Value Data Set ...................... 15-19
15-3 Hazardous Waste Sites in the Boston SMSA Identified
Before 1982 15-21
xx i n
-------
TABLES (continued)
Number Page
15-4 Hedonic Property Value Function; Initial Version ..... 15-23
15-5 Average Distance Responses, Distance Risk Change
Responses, and Actual Distances by Town. ........ 15-28
15-6 Mean Distance Responses Grouped by the Size of the
Risk Change ........................ 15-31
15-7 Demand-for-Distance Models ................ 15-33
15-8 A Comparison of the Region I Survey Responses and
Harrison Data by Town 15-39
15-9 Predicted and Actual Distance to Hazardous Waste
Sites by Town ...................... 15-42
15-10 Features of the Models for Predicting Distance from
Hazardous Waste Sites ................... 15-43
15-11 Overall Means for Predicted and Actual Distances ..... 15-45
15-12 A Comparison of Actual and Predicted Distances ...... 15-46
16-1 Labor Market Estimates of Value of Unit of Risk
Reduction 16-4
16-2 Implicit Values for Risk Change, Excluding Protest
Bids 16-7
xxiv
-------
CHAPTER 1
ON VALUING REDUCTIONS IN HAZARDOUS WASTE RISKS
1.1 INTRODUCTION
This chapter introduces our draft interim report on using the contingent
valuation approach to measure the benefits of reducing hazardous waste risks.
This approach involves using survey techniques to elicit people's expressed
preferences, or intended behavior, to estimate the value of reducing these
risks.
The research described in this report was conducted in response to our
two main objectives:
To develop a framework for using the contingent vaiuation
approach to measure the benefits to individuals from reductions
in hazardous waste risks.
To design a framework for comparing a hedonic property value
model* for benefit measurement with contingent valuation when
a risk change is the source of the benefits.
To meet the first objective, our contingent valuation analysis explicitly
recognizes the difficulties posed by investigating individual behavior under un-
certainty. As a first step, we began our conceptual analysis. To complement
this effort, and before conducting the contingent valuation interviews, our re-
search activities focused on improving our understanding of the interview tech-
niques and questions that communicate concepts involving risk. These activities
involved using focus groups, a detailed pretest, and videotaped interviews
and progressively revising the questionnaire and vehicles used to explain risk.f
*ln this approach, values are indirectly inferred from the residential loca-
tion decisions of the household.
tGiven the scope of this effort, a separate report of these activities was
prepared with partial support from this cooperative agreement (see Desvcusges
et al. [1984a, b] for a more complete summary of these activities) Chapter 8
in this report summarizes the process used to develop the questionnaire, but
should not be considered a complete description of these activities.
1-1
-------
Based on these efforts, we conducted a contingent valuation survey in the
Boston area during the Spring and early Summer of 1984, This report pre-
sents preliminary results from the empirical analysis of the valuation responses
elicited in these interviews.
To meet the second main objective, we also formed a joint effort with David
Harrison (then of Harvard University/ now associated with Dun & Bradstreet)
to acquire information consistent with a hedonic property value analysis involv-
ing hazardous-waste-related risks that he was completing -with support from
the U.S. Environmental Protection Agency (EPA) under Cooperative Agreement
No. CR-809702-01. Harrison's independently developed hedonic model is the
indirect method for measuring the marginal value of risk changes to which we
compare our contingent valuation approach. Harrison's method also provides
some of the necessary information for using part of the contingent valuation
survey in comparing the two approaches. This report presents the preliminary
results from this comparison.
In addition to our two overall objectives, the research has many specific
objectives, which are identified and discussed in the chapters that follow.
Among these specific objectives are measuring both use and intrinsic values
for risk changes and examining the influence of the attributes of risk, risk
endpoints, and risk outcomes on individuals' values for risk reductions. Addi-
tionally, our research examines the importance of assigning different property
rights to risk levels. Finally, our research compares alternative question for-
mats for eliciting individuals' values of reductions in hazardous waste risks.
The report provides substantial support for using contingent valuation to
elicit values for reducing hazardous waste risks. The overall quality of field-
work, the relatively low number of protest responses, the generally high levels
of statistical significance of valuation response means, and the good perform-
ance of our "restrictive" models lend credence to this conclusion. However,
the specific estimates must be regarded as very preliminary. Indeed, this
report is best viewed as structuring an agenda for future research activities
that either may yield more definitive estimates of the values for risk reductions
or suggest reasons that general conclusions on the nature of these values can-
not be drawn from the methods and information in our survey results.
1-2
-------
To introduce the report, the following sections provide an economic per-
spective for viewing regulatory policies involving hazardous wastes. Specific-
ally, Section 1.2 highlights the legislative mandates that suggest the importance
for hazardous waste environmental policies. Section 1.3 discusses the roie of
benefit analysis for the regulatory policies resulting from that legislative man-
date. Section 1.4 describes the outcomes of these regulatory policies as reduc-
tions in the risk of exposure to hazardous wastes for households and the eco-
system. Section 1.5 provides a general economic framework for viewing a
household's decisions involving risk. Section 1.6 describes the more restric-
tive conceptual framework that underpins our contingent valuation analysis for
measuring the benefits of reducing hazardous waste risks. Section 1.7 pro-
vides a brief overview of the contingent valuation approach, which is one of
the primary focuses of our research activities. Section 1.8 presents an over-
view of our overall research design, including our research objectives and the
activities completed. Finally, Section 1.9 presents a guide to the report.
1.2 THE IMPORTANCE OF HAZARDOUS WASTE REGULATORY POLICY
Hazardous waste regulations constitute one of the decade's most pressing
environmental policymaking challenges. Local, State, regional, and several
Federal agencies are already participating in a regulatory process that will
ultimately encompass the generation, transportation, storage, and disposal of
hazardous wastes. Despite this wide range of activity, however, our primary
focus is on the most influential regulatory element--the hazardous waste regu-
latory actions of EPA.
Congress has mandated EPA's involvement by passing the Safe Drinking
Water Act of 1974, the Resource Conservation and Recovery Act (RCR A) of
1976, and the Comprehensive Environmental Response, Compensation and Lia-
bility Act (CERCLA) of 1980, also known as the "Superfund" Act. The Safe
Drinking Water Act provides for general protection against a variety of organic
and inorganic contaminants and also protects specific aquifers. RCRA contains
a wide range of regulatory mandates involving all facets of the hazardous waste
problem. The Superfund Act requires a comprehensive "cleanup" of unregu-
lated, abandoned hazardous wastes dumps.
1-3
-------
The importance of EPA's role in hazardous waste policy has been directly
stated in the recent reauthorization of RCRA. In the Hazardous and Solid
Waste Amendments of 1984, Congress has required EPA to take significant steps
to reduce the likelihood of exposures to hazardous wastes. For example, the
Congress has called for the prohibition of the land disposal of any hazardous
wa tes where such disposal cannot be shown to be protective of human health
and the environment. If implemented according to its narrowest interpretation,
this prohibition will impose costs on the society that could- be in the billions
of dollars annually. Clearly, these legislative actions imply that EPA can be
expected to play a central role in future hazardous waste regulatory actions.
To identify substances whose transportation, treatment, storage, or dis-
posal might increase mortality or serious illness or pose a hazard to human
health or the environment, EPA has defined hazardous waste as any solid waste
that is ignitable, corrosive, reactive, or toxic. This definition currently in-
volves some 400 chemicals and 85 waste processes, but, as yet, the magnitude
of the hazards is uncertain.
An important, and often confusing, aspect of this definition is the differ-
ence between a hazardous substance and a hazardous waste. The two terms
are not synonymous even though they may involve the same substance, e.g.,
chromium. A hazardous substance becomes a hazardous waste only after it is
discarded or, in economic terms, becomes a residual of some production proc-
ess . Figure 1-1 shows the distribution of various types of wastes and the
number of generators from each waste type. Present estimates are that 85
percent of these wastes are from the manufacturing sector, with the chemical
and related processes, metal-related products, and electrical equipment indus-
tries accounting from the majority of the wastes generated (see Westat [1984]).
Thus, hazardous waste legislation and the resulting regulatory policies ulti-
mately affect sectors that are important components of the overall economy.
1.3 THE ROLE OF BENEFITS ANALYSIS FOR REGULATORY POLICY
In implementing some of the regulatory actions that stem from the Con-
gressional mandates, EPA will be subject to the provisions of Executive Order
12291. This order requires that agencies conduct regulatory impact analyses
of major regulations and of precedent-establishing regulations. Specifically,
1-4
-------
14,088 Total Hazardous Waste Generatora
8,117 D001 Sgnitable Wastes
4,705 D002 Corrosive Wastes
999 D003 Reactive Wastes8
3,923 D004—D0017
E.P. Toxic Wastes
7,180 F001—F005
Spent Halogenated and Nonhalogenated
Solvents
2,309
Foiifi—FA1Q Electroplating and Coating Wastewater Treatment Sludges
and Cyanide-bearing Bath Solutions and Sludges
1,439
K001—K106 Listed Industry Wastes from Specific Sources
1,438
P001—P123 Acutely Hazardous Wastes
4,062 U001— U247
Off-Spaeiflcation or Discarded Commercial Chemical
Products and Manufacturing Intermediates
2,105
State Regulated Hazardous Wastes
1,720
Self-defined Hazardous Wastes
I I __r-
25% 50% 75%
Percent of All Hazardous Waste Generators
0* 10%
100%
•Confidence interval exceeds ± 25% at the 3S% Confidence Level,
Source*. Wastat [1884J,.
Figure 1-1. Number of establishments generating each major waste group.
1-5
-------
r
the Executive Order calls for a consideration of the benefits and cost of regu-
latory actions and their alternatives. The research described in this report
relates to the tasks associated with measuring the benefits of regulatory actions
that reduce the risk of exposure to hazardous wastes.
The principles of benefit-cost analysis provide only a guide for decision-
making, not a rule. They do not provide the final answer for the policymaker,
who must consider other issues such as the distribution of benefits and costs,
the impacts on small business, or the equity implications of policies that dif-
ferentially affect the risks experienced by different groups in the population.
(See Desvousges and Smith [1983] for an overview of benefits analysis.) How-
ever, identifying, classifying, clarifying, and, where feasible, monetizing the
likely outcomes of proposed regulatory actions significantly enhance the abi lity
of the policymaker to respond to the regulatory mandates of this legislation.
1.4 A CENTRAL THEME: THE ROLE OF RISK
If there is a central theme to the legislative mandates for regulating the
management of hazardous wastes disposal practices, it is one of risk. As for-
mer EPA administrator William Ruckelshaus stated, the problems involving risk
confront EPA with one of its most difficult challenges--a challenge that will
require improvements in EPA's decisionmaking process for risk issues (often
called risk management), in the scientific measurement of risk (similarly, called
risk assessment), and in the communication of risk-related issues to the public.
in turn, the scope of these challenges will require integrated research efforts
encompassing a variety of disciplines, ranging, for example, from engineering
and toxicology, which make technical measures and assessments of risk, to
psychology and economics, which predict and evaluate perceptions of and
behavioral responses to risk.
The concept of risk has multiple meanings. In some disciplines, it is syn-
onymous with the probability of some injury or health effect (e.g., cancer or
heart attack). In economics, it can imply the variability of investment out-
comes in formal models of economic decisionmaking under uncertainty. (See
mith [1984a] for a discussion of these points.) This report uses the term
i sk to imply the chance that a detrimental event will happen. (Chapter 2
.provides a comprehensive discussion of our definition of risk and compares it
with other frequently used definitions. )
1-6
-------
In this research, reductions in the risk of exposure to hazardous waste
play a fundamental role. We view these risk reductions as the primary policy
outcomes, or effects, of regulations on the management of hazardous wastes.
Our framework considers these risks from hazardous wastes as consisting of
two parts--an exposure risk and a conditional risk of dying if exposed to haz-
ardous wastes. This distinction is fundamental to our research design. More-
over, we have assumed that regulations affect only the risks of exposure and
not the conditional risk. Finally, our focus had been almost exclusively on
mortality as the outcome and not morbidity.*
To illustrate the role of reductions in hazardous waste risks in our re-
search, Figure 1-2 shows one example of linkages between a regulatory action,
its effects, and a household's behavioral responses. In this example, the
regulatory action changes the types of disposal practices that are allowed for
a hazardous waste. Specifically, the action might eliminate land disposal as
an alternative for liquid wastes containing cadmium. The action changes the
risk of contamination by cadmium for the affected environmental media--e.g.,
groundwater and surface water. By lowering the risk of contamination for
groundwater and surface water, the ecological habitats that are affected by
these media--e.g. , plants, fish, and wildlife that live in an ecosystem near a
recharge zone for an aquifer--experience a lower risk of exposure to cadmium.
Equally important, households are affected by a lowered risk of exposure
through the drinking water or some other pathway. In evaluating the prospec-
tive welfare gains from such policies, the task of a benefits assessment is to
measure the value that the household places on the risk changes as a result
of the regulatory action.
The processes underlying these linkages are considerably more complex
than we have described. The extent of this complexity is not fully known as
there is an inadequate understanding of the technical, environmental, and
behavioral processes that are at work. Nevertheless, such an outline does
*We recognize that changes in morbidity risks also may be very important
effects from hazardous waste regulations, but found it necessary to narrow
the emphasis to mortality to make the scope of the research more manageable.
If our approach proves to be useful for valuing changes in mortality risks,
the morbidity component could be added in future research efforts.
1-7
-------
Hazardous Wast©
Regulatory Action(s)
Technical Effects
of Hazardous Waste "
Regulatory Action(s)
Behavioral Effects
of Hazardous Waste ~
Regulatory Action(s)
Figure 1-2. Effects and responses to a hazardous waste regulatory action.
1-8
-------
enable us to develop a general basis for describing the effects of regulatory
actions as changes in risks.
In contrast, much of the research in areas involving risk has concentrated
on the outcomes--e.q., cancer cases avoided, reductions in restricted activity
days. This focus on outcomes is essentially an ex post view of the benefits
of reducing risks. As developed in Chapter 4, our conceptual analysis sug-
gests that differences in these valuation perspectives--this ex post approach
versus an ex ante perspective—can result in substantial differences in values
placed on changes in risk.
1.5 RISKS AND HOUSEHOLD DECISIONS
Individuals make decisions involving risk every day. For example, these
decisions may involve planning purchases for durable goods-~such as an auto-
mobile—with limited knowledge of their future income and use patterns for the
goods planned for purchase. In addition, the implicit value of the automobile
can be affected by circumstances and actions that are within the household's
control as well as those that are not, ft dramatic increase in the price of oil,
as occurred twice in the 1970s, can change the relative values of large versus
small automobiles. Such changes are outside the household's control; thus,
uncertainty over the price of gasoline affects household choices both in the
purchase decision (as a yes/no choice) and in the type of vehicle selected.
Other choices of the household—such as residential location — can also be im-
portant to the value of the sevices provided. Economic models of these deci-
sions routinely assume that individuals acquire information and formulate plans
based on that information.
A household's opportunities for adjusting to uncertainty affect its planned
behavior. For example, paying a higher price for a fuel-efficient car is one
way to provide for the present cost and the uncertain future costs of using
the automobile (i.e. , the price of gasoline). Maintenance contracts that protect
against the car's failing are another. The first adjustment opportunity is an
example in which a single payment--! .e., the premium in price for the vehicle-
is paid regardless of the future price of gasoline. The second case is an ex-
ample of differential payments. That Is, the household purchases the mainte-
nance contract at a price that constitutes the full cost if the automobile does
not experience problems. However, with failures in performance covered in
1-9
-------
the contract, the costs of repaii—and hence the actual cost of the contract--
are subsidized. This latter adjustment is the basic nature of insurance, ft
is a claim to a level of wealth that is tied to a particular outcome that may or
may not occur. Thus, the presence of a market for contingent claims, or in-
surance, provides one way for a household to adjust to the presence of uncer-
tainty. Our conceptual analysis developed in Part I suggests that opportunities
for adjusting to risk will affect the household's values for a risk reduction.
Clearly, modeling household decisions involving risk is a complex task.
In this report we have not developed a comprehensive framework for modeling
all such household decisions. Nonetheless, it is possible to describe the ele-
ments of a simplified view of how such a framework ultimately might be struc-
tured . Figure 1-3 is an example of how one might view the households' deci-
sions regarding risk. At the center of the framework is the household, which
is exogenously faced with some risk of dying in a given year by virtue of its
genetic endowment. It also experiences risks through its occupational choices,
the location of its residence, and its purchases and use of goods and services.
Each of these boxes has two arrows indicating that these hazards can be volun-
tarily accepted, to the extent there is sufficient information to perceive them.
In addition, there is another set of sources for risk that are imposed on the
household by other factors--the actions of other individuals or firms, policies
of any level of government, or nature itself. These are in some respects sim-
ilar to genetic risks in that the household usually has no basis for direct con-
trol of them. We have designated these risks with single arrows to suggest
that, for the most part, they are involuntary. This does not imply the house-
hold cannot take action to avoid them or mitigate their impacts; rather, it im-
plies there are few (if any) perceived mechanisms for the household to change
them directly.
The risk of exposure to hazardous wastes can be experienced as both a
voluntary and an involuntary risk.* However, these risks are generally
thought to be involuntary and experienced through the location of the house-
hold's residence, which includes its environmental conditions and drinking
*The information about the risk can also be a determinant of whether it
is a voluntary or involuntary risk.
1-10
-------
Figure 1-3. Multimarket framework.
-------
water.* Residents In Seymour, Indiana, who lived quite close to the aban-
doned Seymour recycling site were exposed to a variety of wastes through air,
groundwater, and surface water media. While this Superfund site is likely an
extreme example, it does suggest the types of exposures that could occur as
a result of residential location.
Many of the features of this household decision framework are poorly un-
derstood. For example, most measures of risk aversion are defined without
regard to the attributes of the risk. Equally important, models of household
behavior only recently have moved beyond partial equilibrium models to de-
scribe how the household adjusts to environmental amenities (see Roback
[1982], Bartik and Smith [forthcoming] and Hoehn, Berger, and Blomquist
[1984]). In addition, psychologists also have pointed out the difficulty that
people experience in processing information involving risk (see Kahneman [ 1984]
and Tversky and Kahneman [1982]). Nevertheless, our multiple source ap-
proach provides a general view of hazardous waste risks as a part of a larger
picture. This view seems consistent with a number of analysts who use it to
explain the greater current concerns for risk when the apparent overall risks
to individual well-being are lower than anytime in the past. (See Douglas and
Wildavsky [1983].)
1.6 HAZARDOUS WASTE RISK: OUR FRAMEWORK
To begin developing a framework for viewing hazardous waste risks, our
conceptual analysis uses the conventional expected utility framework as a
starting point. (Part 11 presents this analysis.) Originally introduced by
Von Neumann and Morgenstern [1947], expected utility implicitly assumes that
the household can quantify the probabilities associated with the uncertain
events it faces. In accordance with certain basic axioms, the household plans
its decisions to maximize the weighted average of the values of the conse-
quences, or its expected utility.
Researchers have raised a number of questions concerning the plausibility
of this model as a framework for describing and predicting behavior under
*One example of an occupational exposure was provided by a participant
in one of our pre-survey focus group discussions who felt that he had been
exposed as a member of a state highway department road crew cleaning up
polychlorinated byphenols (PCBs) that had been discarded along the roadside.
1-12
-------
uncertainty (see Schoemaker [1382]). Our conceptual analysis suggests that
amending the framework to allow the household's utility to depend on the state
of the world that actually occurs may explain some of the inconsistencies that
have been found in previous research.* for example, under this view, an
individual's utility function will differ depending on whether or not exposure
to hazardous wastes occurs.
An important implication of our conceptual analysis is that the appropriate
basis for valuing reductions in the risks of exposure to hazardous wastes de-
pends upon the opportunities available to the individual for adjusting to risk.
Under certain circumstances, this measure of value will be the option price,
or constant payment irrespective of the outcome at risk, for the specified
change in the likelihood of the detrimental event.
Our research design also recognizes that not all risks are the same. The
literature in psychology and, to a lesser extent, in economics has begun the
process of distinguishing types of risks. With this identification of types
comes a corresponding need to identify how they are different—in effect, to
enumerate their attributes. Although our conceptual analysis does not expli-
citly include the attributes of risk in its description of household behavior,
the design for our empirical analysis provides preliminary information on how
some of these risk attributes might affect households' valuation responses.
1.7 CONTINGENT VALUATION: A BRIEF OVERVIEW
Contingent valuation is the use of survey methods to elicit individuals'
values for improvements in environmental quality, such as reductions in haz-
ardous waste risks. These values are elicited for specific hypothetical changes
in environmental quality that are described in the survey questionnaire. The
use of surveys to elicit behavioral information is widespread in psychology,
sociology, and market research, and use of the contingent valuation approach
to value improvements in environmental quality has generated a decade of
experimental and field research. Even in this relatively short period, how-
ever, the approach has grown more sophisticated, improving how it defines
*The terms household and individual are used interchangeably throughout
this report. It is possible to develop models demonstrating that households
as collections of individuals behave as if guided by a single utility maximizing
economic agent. See Becker [1981] for examples and further citations.
1-13
-------
the objectives of the survey, how it structures, orders, arid asks the ques-
tions to elicit respondent valuations, and how it chooses appropriate samples
of respondents. Of course, the cornerstone of contingent valuation is the sur-
vey questionnaire, which must
"Frame" the commodity--!.e., describe believable and under-
standable terms in the regulatory effects that the respondent
must have.
Establish a "market" context for the commodity that effectively
describes the conditions under which it must be valued.
Effectively elicit respondent values for the commodity.
The need for considering the contingent valuation approach for valuing
hazardous waste risk changes stems from the lack of any organized market in
which the changes would be valued. In the absence of markets, economists
have used other approaches besides contingent valuation. Presently, we are
unsure of their relevance for valuing hazardous waste risk changes. For ex-
ample, the travel cost approach using the implicit price that people are willing
to pay to visit a recreation site may not be appropriate because few recreation
sites are likely to be affected by hazardous wastes. In addition, the early
results with the property value studies seem to have too much noise to deter-
mine the effects of hazardous waste risks on property values. Thus, asking
people directly in a survey may offer the only alternative.
The central question facing our research is "Can contingent valuation be
used to value reductions in hazardous waste risks?" A long and formidable
list of reasons has been given as to why contingent valuation cannot provide
accurate estimates of values. From the psychologist, the reasons that contin-
gent valuation cannot be used include the following:
People's values for commodities like hazardous waste risks are
labi'e or poorly formed (Slovic, Fischhoff, and Lichtenstein
[1982])
People's preferences will be very sensitive to how questions to
elicit values are framed (Tversky and Kahneman [1381])
People will be unable to process information regarding low prob-
ability events.
1-14
-------
The economist's list includes the psychologist's concerns and adds the follow-
ing :
People will be unable to comprehend the commodity to be valued
because it has no market equivalent (Cummings, Brooksbire,
and Schulze [1984])
People will give unreliable answers because the questions are
hypothetical (Bishop and Heberlein [1984])
People will give a response based on attitudes and not behavior
(Bishop and Heberlein (1984])
People are not familiar with the range of their preferences that
involve hazardous waste risks (Freeman [1984b])
These are important concerns. This report attempts to provide the infor-
mation necessary to address a large number of them. However, in most cases,
there is no unambiguous standard that can be used. Rather, the reader must
weigh the information provided and decide whether or not contingent valuation
can be used to measure the benefits of reducing hazardous waste risks.
We suggest that the economists and psychologists are basically saying
the same things but are using a different vocabulary. The crux of the matter
is effectively framing the commodity. In this regard, we have adapted several
techniques from, psychological and market research --e. g., focus groups, video-
taped interviews, and extensive pretests—to evaluate the effects of different
frames on respondents and to develop our final questionnaire. This detailed
report is our way of letting the reader judge for himself about the overall suc-
cess of our efforts.
In view of the number and types of issues concerning contingent valua-
tion, some perspective on the approach may be helpful in trying to evaluate
it. Important considerations include the following:
Contingent valuation elicits responses directly from people — fre-
quently a random sample chosen from some population. Con-
tingent valuation allows the researchei—-through the question-
naire—to elicit information both on values and on the reasons
for the values provided.
Contingent valuation offers the opportunity to tailor questions
to the issue at hand, ft also has the ability to structure an
experimental design for testing specific concerns that may be
relevant to a valuation estimate.
1-15
-------
Contingent valuation research can exert control over the sam-
pling and survey procedures used to collect data and thereby
provide information designed for the task at hand.
Contingent valuation can yieid insights into people's ability to
perform tasks that implicitly are required of them by the indi-
rect or market based approaches.
These attributes strongly suggest it may be useful in understanding people's
preferences for changes in hazardous waste risks. And while the specific esti-
mates in this report are preliminary for many reasons, we have concluded that
the overall prospects for using contingent valuation to value risk changes is
quite good--good not only in terms of the response rates, the rates of people
rejecting the commodity, and the estimated mean valuations that are consistent-
ly significant, but also in terms of the performance of our more in-depth
regression analysis using the restrictive models and the plausibility of the con-
tingent ranking analysis. The reference operating conditions for the accuracy
of contingent valuation developed in Cummings, Brookshire, and Schulze [1984]
would have led us to expect a less optimistic prognosis. At this stage, a num-
ber of issues will require further investigation to understand their full implica-
tions. Nevertheless, in our judgment--and it is only that—contingent valuation
can yield meaningful economic information. Ultimately, the reader will have to
draw his own conclusion based on his interpretation of the information provided
in this report.
1.8 RESEARCH OVERVIEW
Our primary objectives, which relate to the task of valuing risk changes,
have provided the main guideposts for our research. Figure 1-4 presents an
overview of our research to attain these objectives. Both our objectives and
the subsequent activities follow directly from our assumption that risk changes
are delivered by regulatory actions involving hazardous wastes.
As shown in Figure 1-4, the types of values to be measured are important
to these research objectives. Our research considers two types of values:
Use values, which accrue to households as a reduction in their
risk of exposure and possible premature death from hazardous
wastes
Intrins c values, which accrue to households from knowing that
the risk of exposure to hazardous wastes has been reduced for
plants, wildlife, and animals.
1-16
-------
Figure 1-4. Study overview,
1-17
-------
Our research provides a conceptual framework for measuring both of these
types of values, and our contingent valuation survey contains questions to
elicit both. However, only the use values are relevant to our comparison ob-
jective because the hedonic property value model is capable of measuring only
use values that would correspond to reductions in hazardous wast# risks.
As also shown in Figure 1-4, two Important research issues stem from
our study objectives:
How should the contingent commodity--hazardous waste risk
reductions—be framed?
How can the available approaches for valuing risk reductions
be linked?
To shed some tight on these two research issues, we undertook several re-
search activities shown in the large box in the middle of Figure 1-4, including
the early phase of our conceptual analysis and the efforts to understand how
people would respond to a contingent valuation questionnaire involving hazard-
ous waste risks.
An early but important research activity was the series of focus group
sessions—and other subsequent questionnaire development activities—that fed
to the formulation of our final research design. The key features of this de-
sign include the following:
A risk ladder to elicit risk perception information prior to the
framing of the contingent commodity
The separation of risks experienced by individuals into an ex-
posure risk and the conditional risk of death given exposure
as well as the use of three risk circles to describe the exposure
and conditional risks and their effect on the joint probability
as part of the framing of the commodity
The use of direct question and contingent ranking question for-
mats to elicit the values for risk reductions
The introduction of an experimental design to test for the effect
of risk levels and property rights for valuation responses
The use of alternative framing to evaluate the influence of dif-
ferent risk outcomes — e.g., how death might occur and a risk
of severe birth defects
The use of two alternative methods for linking risk and dis-
tance to a hazardous waste facility.
1-18
-------
These activities led to a final survey questionnaire that was administered
by a staff of interviewers to a stratified random sample of 953 households in
suburban Boston that resulted in 609 completed interviews. Almost 87 percent
of enumerated households completed the interviews. Only 3 interviews out of
the 612 were broken off after initiation.
With these data, and the model and results acquired for the property val-
ue analysis, we have initiated the preliminary empirical analysis that is pre-
sented in Part 111 of this report. The activities included the-following:
Examination of protest responses
An analysis of mean option price responses
Multivariate analysis of the option price responses
Estimation of contingent ranking models and a preliminary com-
parison with the direct question responses
Development of a framework for comparing contingent valuation
and hedonic models including an initial comparison.
These findings are all preliminary because of a substantial number of issues
that are not included in this report due to time and resource constraints.
Table 1-1 summarizes the type of issues that were considered in the empirical
analysis and shows the location of each in the report.
1.9 GUIDE TO THE REPORT
For the reader's convenience, this draft interim report is divided into
three parts. These three parts and the chapters they encompass are as fol-
lows:
Part I -- A Conceptual framework for Measuring the Bene-
fits of Reducing Hazardous Waste Risks
Chapter 2 -- The Nature of Benefits Analysis in Hazardous
Waste Management
Chapter 3 -- Modeling Behavior Under Uncertainty: A
Heuristic Review
Chapter 4 -- The Role of the Ex Ante and Ex Post Perspec-
tives in Measuring Welfare Changes Under
Uncertainty
Chapter 5 -- A Conceptual Framework, for Valuing Risk
Reductions
1-19
-------
TABLE 1-1, SUMMARY OF THE OBJECTIVES AND STRUCTURE OF THE RESEARCH
Objective
Type of risk
Concept of value
Direction of
risk change
Measurement method
Experimental design
Results
Valuation of
risk changes
Exposure to
hazardous
wastes
Ex ante value
to individual
Existence value
Decreases and
increases.
Decreases only
Direct question for
hypothetical situation.
Contingent ranking
for hypothetical cases
Direct question
Exposure risk and
conditional probability
Exposure risk and payment
Chapter 11
Chapter 12
Chapter 13
Chapter 14
Influence of
personal char-
acteristics for
risk valuation
Exposure
to hazard-
ous waste
Job risk
Ex ante value
Decreases arid
increases
Report on actual condi-
tions and attitudes
Representative sample
of households in sub-
urban Boston (with
oversampling of Ac ton)
Chapter 11
Chapter 12
Chapter 13
Comparative
evaluation of
methods
Exposure
to hazard-
ous waste
Ex ante value
Decreases
Direct question for
hypothetical question and
hedonic properly value.
~ ™ ~
Chapter IS
Job risk
Ex ante value
Increases
Direct question For
hypothetical question and
hedonic wage model.
"""" "*
Not in
Phase t
report
Evaluation of
attributes of
risk
Fatal accident
on the job
Ex ante value
to individual
Increases only
Direct question for
hypothetical cases.
Not in
Phase 1
report
Information
on wastes
Exposure to
hazardous
wastes
$
Direct question of
individuals and
review of newspapers
Chapter 10
Perception
of risk
Fatality due
to auto acci-
dent, heart
disease, air
pollution,
hazardous
waste
On-the-job
risk of death
Direct question of
perceptions.
Different types of risk.
Chapter 10
Not in Phase 1
report
Role of avert-
ing cost or
atmg
activities on
risk valuation
—.--T~Ru — —*r- -_2C
Exposure lo
hazardous
waste
Ex ante value
to individual
Report on household's
actual activities.
Representative sample
of households in
suburban Boston {with
over-sampling of Acton}
Not in
Phase 1
report
-------
Chapter 8 -- Ecological and Intrinsic Values Under
Uncertainty
Part II -- Research Design, Questionnaire Development,
the Survey
and
Chapter 7
Chapter 8
Chapter 9
Chapter 10
Research Design:
to Practice
The Transition from Theory
Survey Questionnaire Development
Sampling Plan and Survey Procedures
Profile: The Survey Area and Its Population
Part III -- Preliminary Empirical Analysis
Chapter 11 --
Chapter 12 --
Chapter 13 --
Chapter 14 --
Chapter 15 --
Option Price Results: The Framing of the
Commodity and an Analysis of Weans
Option Price Results: Preliminary Regression
Analyses Using Unrestricted Models
Valuation Estimates for Risk Reductions:
Using Restricted Models
The Use of Contingent Ranking Models to
Value Exposure Risk Reductions: Preliminary
Results
A Comparison of Contingent Valuation and
Hedcnic Property Value Models for Risk
Avoidance
Chapter 18 -- Policy Implications and Research Agenda
Part I --Chapters 2 through 6--describes the conceptual framework we
developed for assessing the benefits from regulations governing hazardous
wastes. Part 11--Chapters 7 through lO--explains how we implemented our
conceptual framework by developing and administering a contingent valuation
survey. Part III--Chapters 11 through 16—presents preliminary findings of
our empirical analyses of the survey data. Also for the reader's convenience,
each part is preceded by an introduction that explains the objectives and
research activities associated with each and outlines the purpose, scope, and
contents of each chapter. Chapter 1? contains references we have cited in the
text.
1-21
-------
BART i
A CONCEPTUAL FRAMEWORK FOR MEASURING THE BENEFITS
OF REDUCING HAZARDOUS WASTE RISKS
-------
r
PART 1
A CONCEPTUAL FRAMEWORK FOR MEASURING THE BENEFITS
OF REDUCING HAZARDOUS WASTE RISKS
Part I of this draft interim report describes how we developed a concep-
tual framework for assessing the benefits expected to accompany regulations
governing hazardous wastes. In particular, Part I consists of the following
five chapters:
Chapter 2 - The Nature of Benefit Analysis in Hazardous Waste
Management
Chapter 3 - Modeling Behavior Under Uncertainty: A Heuristic
Review
Chapter 4 - The Role of the Ex Ante and „ : Post Perspectives in
Measuring Welfare Changes L nie Uncertainty
Chapter 5 - A Conceptual Framework for Valuing Risk Reductions
Chapter 6 - Ecological and Intrinsic Values Under Uncertainty
While our intention is not to evaluate the benefits arising from specific regula-
tory actions, the purpose of the framework is to provide a basis for describing
how such analyses might be conducted. As briefly outlined in the following
paragraphs, there are five important elements in the proposed framework.
First, the analysis assumes that regulations for hazardous wastes reduce
the probability that an individual will experience some type of adverse effect
from the unintended release of wastes to the environment. This assumption
seems consistent with the U.S. Environmental Protection Agency (EPA) regula-
tory evaluation policies such as the RCRA Risk/Cost Model and the Liner Loca-
tion Model, Because release of the wastes can lead to risks to human health
and to ecological systems, individuals may place a positive value on policies
that reduce these risks. A conceptual framework for describing this valuation
process is a necessary first step before empirical analysis can be undertaken.
1-1
-------
Second, the assumption that regulations change the risk of exposure ex-
perienced by individuals suggests the need to develop a conceptual basis for
valuing these risk changes based on a model of individual behavior under un-
certainty. This analysis assumes that the expected utility model can provide
the starting point for describing individuals' responses to risk. Consequently,
it enables us to define an individual's values for changes in risk. That is,
the values for reducing hazardous waste risks will be based on the expected
utility that the individual anticipates from a regulatory act-ion. For example,
more stringent containment of hazardous wastes in land-based disposal sites
might be assumed to reduce exposure risks by some amount. This action, in
turn, reduces the likelihood of detrimental events (e.g., exposure and some
health effect) and thereby increases the expected utility to be experienced by
the affected individuals. The monetary value of the risk change could be
measured by following the Hicksian analogy for the case of certainty.
Third, once the expected utility framework is used as the basis for valu-
ing risk changes, we have accepted an ex ante perspective for welfare analy-
sis. That is, we are maintaining that the relevant benefit measure is based
on how the individual's planned activities change with risk changes. As we
develop later, the perspective for measurement is important for evaluating de-
cisions made under uncertainty, especially if those decisions affect the nature
of the uncertainty itself.
Fourth, to define how much the individual would be willing to pay for
the risk change, we must define how these payments would be made and the
opportunities available to the individual for adjustment to changes in risk. In
effect, we must define the institutions that constrain how an individual can
plan his expenditures given uncertainty as to future states of the world.
Finally, the last key element in our conceptual framework is the motivation
for valuing the risk reduction. We noted earlier that a pattern of exposure
implies a subsequent set of risks that may involve detrimental health effects
for the individual or impacts on specific ecological systems. Each of these
types of events will affect individuals differently. These different effects can
lead to different types of values or benefits for policies that regulate the dis-
posal of hazardous wastes.
1-2
-------
Clearfy, each of these elements involves complex problems. Moreover, in
some cases the literature is not fully developed with respect to the specific
issues associated with extending benefits analysis to deal with risk changes.
It is therefore unrealistic to expect that we can provide a comprehensive analy-
sis of all of the issues in a few chapters, Thus our objectives have a more
limited scope:
Review the elements involved in our proposed formulation of
the problem
Relate them to past efforts to define benefit concepts under
uncertainty
Describe the association between the behavioral relationships In
our framework and the findings of psychologists and decision
scientists involved in the study of individual behavior in the
presence of risk
Explain the statistical hypotheses and more informal empirical
results that we expect should follow from our analysis of indi-
vidual behavior.
Chapter 2 introduces our conceptual framework by discussing types of
benefits in a conventional taxonomy and the valuation problem for risk changes.
It uses a set of scenarios to describe how events involving hazardous wastes
"fit into" our framework.
Chapter 3 reviews an economic approach for describing individual behavior
under uncertainty using a state preference characterization of how individuals
plan their actions when the future states of nature are uncertain. The Key
elements in the model for our description of the values of risk reduction are
the specification of preferences, the description of the adjustments available
to the individual for responding to risk, and the characterization of individual
attitudes toward risk.
Chapter 4 compares the conventional approach to benefit analysis with
the framework implied by an ex ante analysis of individuals' valuations of risk
reductions. This comparison uses the planned expenditure function introduced
in Chapter 3 as the basis for classifying different types of benefits.
Chapter 5 describes the specific implications of our model for an individ-
ual valuation of the risk reductions associated with regulations on the disposal
of hazardous wastes and how we might expect these values to be affected by
the type of risks experienced.
1-3
-------
Chapter 6 completes the conceptual analysis by focusing on ecological and
intrinsic values in the context of uncertainty, including a reconsideration of
the conventional concepts of existence arid option values. It extends the analy-
sis of Chapters 3, 4, and 5 to the value of reductions in risk to environmen-
tal and ecological resources.
1-4
-------
CHAPTER 2
THE NATURE OF BENEFIT ANALYSIS IN
HAZARDOUS WASTE MANAGEMENT
2.1 INTRODUCTION
As the first component in Part I, this chapter addresses several of the
important elements in our conceptual framework for assessing the benefits ex-
pected to accompany regulations governing the management of hazardous
wastes. Given the likely generic effects of hazardous wastes in the environ-
ment, the following sections describe the types of benefits that might accrue
from regulating the management of hazardous wastes and the types of economic
agents to whom those benefits might accrue. In particular, Section 2-2 pre-
sents a conventional taxonomy of benefits that serves as a starting point for
our analyses, Section 2.3 outlines our treatment of policy outcomes as changes
in the risk of exposure to hazardous wastes and compares this approach to
previous approaches for benefit analysis, and Section 2.4 introduces our notion
of risk as synonymous with the probability of a well-defined detrimental event
(e.g. , death). Section 2.5 describes representative scenarios to suggest the
generic sources of exposure risk and to provide examples of typical contamina-
tion incidents. Section 2,6 briefly summarizes the chapter's main points.
2.2 CONVENTIONAL BENEFIT TAXONOMIES
Analysts have used a variety of classification schemes to describe the
components of the total benefits of a policy action. In the early literature on
benefit-cost analysis, the most widely used taxonomy distinguished between
the benefits associated with private market transactions and public allocations--
i.e., the provision of goods and services which could be purchased in markets
(e.g. , hydroelectric power)--and the benefits associated with goods or services
that did not exchange on such markets (e.g., recreation). The benefits asso-
ciated with the second type of commodity were often assumed unquantifiable
and were usually labeled as the intangible component of benefits. The history
2-1
-------
of methods development for benefits assessment of environmental resources has
seen a progressive advance in our ability to measure these intangible benefits.
Thus, taxonomies that distinguish between the benefits associated with market
goods and intangibles have become less useful over time.
The recent literature on environmental benefit estimation has tended to
follow a general classification scheme originally suggested by Krutilla [1967],
This framework separates user and nonuser sources of benefit from environ-
mental resources. Mitchell and Carson [1981] refined this Initial proposal for
the case of water quality benefits in an effort to understand what would be
measured by the various approaches to benefit analysis. Their taxonomy has
provided the basis for several recent efforts to estimate and to distinguish
the individual components of the benefits individuals realize from environmental
resources.*
Figure 2-1 presents an example of this type of benefits taxonomy. It has
been simplified from the form presented in Desvousges, Smith, and McGivney
[1983] but contains many of the same elements. In this format, the mechan-
isms leading to the beneficial effects experienced by individuals are more spe-
cifically identified. Within the category of direct benefits, a distinction is
drawn between use and nonuse benefits. However, several aspects of these
terms require further discussion. Throughout this report, use and user bene-
fits are considered synonyms. Since there can be subtle distinctions between
the two, it is important to describe what they will mean here. Use benefits
arise because of the active consumption of the services of a resource. As Fig-
ure 2-1 indicates, this can be through clean air's generating improved health
or clean water's allowing game fishing. In all cases, use benefits require the
active involvement of the individual as a user of the services of the resource.
If the definition of user is narrowly interpreted, we might be tempted to
conclude that economic agents termed users are precluded from having nonuse
benefits. However, this is not the case. To appreciate why users may also
* Desvousges, Smith, and McGivney [1983] revised the Mitchell-Carson
framework and used it in estimating the components of water quality benefits
associated with a specific resource — the Monongahela River. More recently,
Fisher and Raucher (1934] have used the framework to appraise the relative
magnitude of nonuser (or intrinsic) benefits and user benefits based on recent
empirical studies that have included both sources of individual values.
2-2
-------
Potential
Benefits of
Hazardous
Waste
Management
Regulations
Indirect
Benefits
from
Effects on
Markets or
Production
improved Agricultural Production
Improved Forestry Production
Improved Fishery Production
Direct
Benefits
from
Effects on
individuals'
Utility
Use
Improved Human Health Conditions
Recreation and Other Uses of Environment
Intrinsic
Option Value:
Possible Future Use
Existence—Mo Use
Figure 2-1 The conventional benefits taxonomy adopted for hazardous waste management regulations.
-------
have nonuse benefits, it is helpful to consider how the resource being valued
contributes to the utility of an individual. When it requires some type of
active experience involving the resource's services, the resulting increment to
utility is a use benefit and the individual experiencing it is a user. However,
this same individual may also derive an increment to utility with no action by
simply knowing that a resource has been enhanced or increased in some way.
This change in utility is a nonuse benefit because the individual does not
actively acquire the services of the improved resource. • Knowledge of the
improvement itself was sufficient to enhance utility.
Figure 2-1 also can be used to show the major channels through which
hazardous materials may enter the environment and affect human welfare. This
figure distinguishes between the effects on production and market values that
affect people's utility indirectly and the direct effects on individuals' utility.
Production or market values arise when some attribute of the ecosystem is an
argument in the production and cost functions for a marketed good. For exam-
ple, if the presence of hazardous wastes in the environment results in a lower
level of an ecosystem attribute, the economic productivity of the ecosystem
would decrease, causing an increase in the cost of producing the marketed
good. These changes in turn would result in changes in market quantities,
product prices, factor prices, rents, and/or profits. Standard economic models
can be used to obtain measures of the economic value of changes in the pro-
ductivity of managed and commercially exploited ecosystems. (See Freeman
[forthcoming a] for a review of these models.) In the case of production and
market values, hazardous wastes in the environment affect individuals' utility
only indirectly by changing the prices of goods they purchase with their
incomes.
In addition to these indirect effects, however, the figure indicates that
hazardous materials can affect individuals directly by altering the level of some
argument in their utility function. For example, if utility depends in part on
health status, exposure to a toxic material through environmental pathways
can lead to lower health status and, therefore, lower utility. Also, if some
attribute of the environment or an ecosystem (e.g., the number of different
species) is an argument in individuals' utility functions, then hazardous mater-
ials can affect utility by altering the level of that attribute.
2-4
-------
Another key distinction in the direct benefits lies between those associated
with in situ use of the environment and those associated with nonuse or intrin-
sic values. As noted above, the in situ use of the environment is an activity
that includes the scarce resources of the individual, including, but not limited
to, time. For example, the individual may have to incur time and other costs
to travel to the site of the ecosystem to engage in some activity. Nonuse or
intrinsic values, on the other hand, are defined as those benefits or welfare
gains to individuals that arise from ecosystem changes independently of any
direct use of the ecosystem. The figure further divides intrinsic values into
pure existence and option values related to some uncertainty concerning future
demands or the availability of the system, for possible use.
The concept of pure existence value was apparently first suggested by
Krutilla 11967] and was further discussed in KrutilSa and Fisher [1975, p, 124}.
Weisbrod [1964] first introduced the term option value in the literature of
benefit-cost analysis 21 years ago. Option value is said to arise either when
an individual is uncertain whether he might demand a good in the future or if
he is faced with uncertainty in the future supply or availability of that good.
Weisbrod apparently viewed the existence of positive option value as intuitively
obvious. But, as subsequent analysis has shown, option value, as convention-
ally defined, can be either positive or negative depending upon the particular
circumstances [Schmaiensee, 1972; Bishop, 1982; Freeman, 1984a],
However, there is a basic inconsistency in this and earlier taxonomies.
They combine two distinct perspectives for welfare analysis—the ex ante and
the ex post frameworks for defining values. The concept of option value con-
nects the two frameworks. Rather than a separate component of benefits, op-
tion value is the result of these different perspectives for welfare concepts.
Consequently, a more consistent taxonomy would identify the particular valua-
tion perspective instead of mixing ex ante and ex post. We will return to this
more general framework in Chapter 4. Therefore, we present the taxonomy
in Figure 2-1 as a starting point for viewing benefits from risk changes. In
particular, benefits analyses for policies that involve changes in risks will re-
quire a different orientation of the selected welfare measures. For example,
consider how we have proposed to describe "the services" delivered by regula-
2-5
-------
tions governing the disposal of hazardous wastes as the reductions in the risk
of exposure to hazardous wastes. Using this present taxonomy, these risk
reductions yield a use benefit when the household's planned consumption
choices of all commodities in all possible states of nature are changed. That
is, any action that causes a change in these planned choices generates a posi-
tive or negative use value.
Use value accrues to the household, but it requires only a change in
planned consumption or activities and not an active involvement. Moreover,
use value involving risk changes can arise only in an ex ante valuation per-
spective, If an ex post perspective were employed, the benefit is no longer
the use benefit from the risk change. In an ex post framework, the value
would stem from the outcome and not the risk change. Thus, the presence of
valuation under uncertainty reveals the need of a new taxonomy for benefits
analysis that also distinguishes between the valuation perspective-*-ex ante ver-
sus ex post—and the nature of the commodity—certain or uncertain.
2.3 THE TREATMENT OF POLICY OUTPUTS AS RISK CHANGES
The conventional practice in environmental benefit analysis maintains that
policy actions lead to changes in either the quantity or the quality of the ser-
vices provided by an environmental resource. - These changes were assumed
to be known with certainty. Thus, benefit concepts were defined based on
how the environmental resource was assumed to affect individual preferences.
By contrast, our analysis of policies related to the disposal of hazardous
wastes treats therm as changing the likelihood an individual will be exposed to
these wastes. It seems reasonable to inquire into the rationale for making this
distinction.
Given our current state of knowledge, it seems reasonable that there is
no aspect of environmental quality that we can assume is available with certain-
ty. The observed level of air or water quality at each time and in each loca-
tion has a significant stochastic element determining its value. This is true
for a number of reasons. One of the simplest to explain concerns the environ-
mental quality-weather interaction. Weather patterns affect the ambient con-
centrations of pollutants and these patterns are best treated as realizations of
2-6
-------
a stochastic process. Equally important, our current knowledge of the rela-
tionship between the activities under policy control--e. g. , emissions of residu-
als into the atmosphere or water courses — is imperfect. The character of the
air diffusion system governing the residuals--) .e., the ambient air quality
associated with each location or the absorptive capacity of each river or lake--
determines the relationship between measures of environmental quality that are
relevant to individuals' behavior and the patterns of emissions of residuals,
Since it is the latter that is affected by policy, this case atso has a significant
amount of uncertainty in the connection between what policy actions can do
and what is "delivered" to the individual.
What is at issue is the degree of uncertainty. Current benefits assess-
ment practices have implicitly maintained that the random influences and associ-
ated uncertainty are small enough that individual behavior can be described
as if it were in response to certain changes in the environmental resources
under study. Of course, this is an assumption--one that may well be inappro-
priate for some circumstances. However, what is important for our purposes
(i.e., in defining the individual benefits associated with a policy action) is not
the random components connecting residual emissions with ambient quality but,
rather, the influence of these sources of uncertainty on individual behavior.*
In contrast to the assumptions underlying the policies in the Clean Air
and Clean Water Acts directed at the conventional air and water pollutants,
there are significant questions concerning whether any level of exposure to
hazardous wastes can be said to be free of risk, f Moreover, it is not clear
that there is a continuous relationship between the level of exposure and the
impacts on the individual. Rather, a discrete framework has often been se-
lected in describing the implications of hazardous wastes for individuals with
any level of exposures potentially leading to detrimental outcomes. In the case
*There is, of course, a separate issue as to how to treat estimation un-
certainty in benefit-cost analysis. In this case, we assume that individuals
can be described as making choices with certainty, but we as analysts observe
these decisions and understand their motivations imperfectly.
tThis is part of the motivation for the definition of hazardous air pollut-
ants in the Clean Air Act.
2-7
-------
of air and water pollution, the close-effects relationships have generally been
assumed to be continuous.*
Finally, it should also be acknowledged that our understanding of disposal
technologies (and their effectiveness) and of the implications of a wide array
of hazardous wastes for ecological systems in general and for human health in
particular is quite limited. Consequently, a framework that recognizes these
uncertainties explicitly and acknowledges that individuals will respond to them
was judged to be necessary for this case.
2,4 WHAT IS RISK?
The term risk has been used in a number of different ways in policy anal-
ysis, A wide variety of definitions can be found in the literature on risk
assessment and risk management. Equally important, in economics, risk is
often associated with that portion of the uncertain outcomes facing an economic
agent that cannot be diversified away (i.e., insured against using market op-
portunities). To this point and throughout this report, the term risk is used
in a narrow definition. It is considered synonymous with the probability of a
specific detrimental event.
The definition of a particular event at risk and the characterization of
the decision problem provide the mechanisms for incorporating some of the fac-
tors discussed by a number of analysts as important to explaining individuals'
responses to risk. For example, Crouch and Wilson [1982] define risk as a
composite of the probability of an adverse event and the severity of the event.
According to their framework, the risk facing an individual can be reduced
either by reducing the probability of the event or by lessening the magnitude
or severity of the event involved. Similarly, in discussing the problems asso-
ciated with judging acceptable levels of risk, Fischhoff et al. f 1981 J defined
risk as the probability of a more specific outcome--reduced human health and
death. They also acknowledge that the cause of the risk can be important to
its perceived severity to the individual. Both of these discussions of risk
*Of course, in implementing these models, distinctions are drawn between
chronic and acute health effects, materials damages, and aesthetics. In some
of the individual sources of benefits, the empirical models have been developed
as if there were thresholds below which no effect would be experienced.
2-3
-------
have been based on the results of experimental analyses of risk-taking behav-
ior along with actual observations of the decisions of individuals in accepting
specific risks. Starr [1969] appears to have been the first to consider this
latter approach to compare risks in order to provide an indirect basis for iden-
tifying the characteristics of risks that influence individuals' willingness to
accept them. While the studies following this tradition have been rather crude,
they are nonetheless suggestive of a general issue that seems to emerge from
both the experimental (field and laboratory) and indirect approaches to under-
standing behavior under uncertainty. It is best summarized by suggesting
that there is a need for a type of hedonic function to describe risks. In
effect, an individual's appraisal of the subjective level of risk may weli depend
on the characteristics of that risk. Many psychological studies have contrib-
uted to identifying some of them. They would include (using primarily Litai's
[1980] terms); volition, severity, origin, effect manifestation, exposure pat-
tern, controllability, familiarity, ambiguity, and necessity.
What is really at issue in modeling individual behavior in response to risk
is how we choose to reflect these characteristics in describing how individuals
make specific decisions under uncertainty. As Arrow [1974] observed some
time ago, the expected utility framework separates the tasks of risk perception
and preference formation. For a state-independent specification of preferences,
this separation is especially clear; with state-dependent specifications additional
information describing the source of the state dependency is needed to maintain
the separation.
To begin operationalizing a hedonic view of the types of risks as they
are perceived by individuals, there are two distinct modeling strategies. The
first, which maintains the separation of preferences and perceptions, proposes
what are often ad hoc ruies for describing how one or more attributes of risk
would affect the perceived risk level. This perceived risk is then used in an
expected utility model to describe behavior. In effect, optimal choices are
separated from risk perception decisions. This is the approach implicitly used
by Kahneman and Tversky [1979] with prospect theory and by a wide variety
of other proposed alternatives to the expected utility framework (see Scnoe-
maker [1982]). Hogarth and Kunreuther's [1984] analysis of the role of ambi-
guity in risk perception is another interesting example of this approach.
2-9
-------
The second approach would require a behavioral model of individual deci-
sionmaking under uncertainty that includes an explanation of risk perception
decisions. This second strategy is an especially difficult one. It would des-
cribe the hedonic function for risk as an outcome of the optimizing decisions
of households in relationship to the range of alternative sources of risk and
thereby remove the separation of risk perception and preferences that is cen-
tral to the expected utility model,*
At present, there are a few attempts that are moving in the direction of
developing such a model, but none of the frameworks offers a complete analysis
of individual decisionmaking (see Bell [1982], and Loomes and Sugden [1982,
1983j as examples). Nor will our conceptual analysis attempt to develop such
a general framework. Rather, we have selected a more conventional expected
utility model, which allows for state-dependent preferences, and then, in our
empirical analysis, we control the attributes of the risks presented to individ-
uals. By presenting two different types of risks (each carefully controlled
through the descriptions given to our survey respondents), the empirical
analysis proposes to add to the information available on the the role of the
attributes of risk on individual behavior, but not to develop a framework that
would deal with these attributes in a general way. The primary reason for
our discussion of these issues at the outset of the analysts is to acknowledge
that the attributes and context of the risks are important to individual behav-
ior. Consequently, valuation estimates of risk changes for certain types of
risks in specific contexts may not be relevant to comparable (in numerical
terms) risk reductions of other types in other settings,
2.5 THE SOURCES OF EXPOSURE RISK
The basic premise that provides the fink between our conceptual analysis
of how households value hazardous waste management policies and changes in
*lt is important to note that in the hedonic models used in economics the
market plays a crucial role in converting the hedonic function into a technical
function for the individual, thereby providing a similar type of separation as
to what has been used in expected utility analysis. While the hedonic price
function is an equilibrium relationship, no one individual can affect it. Conse-
quently, it is treated as a given for any single individual's decisions and
choices are constrained by it. The risk perception process does not seem to
have a comparable institution exerting discipline on the decisions of the house-
hold involved in appraising risks.
2-10
-------
those policies is the assumption that those policy changes lead to risk -educ-
tions. More stringent regulations for the disposal of hazardous wastes reduce
the risk of exposure to those wastes for individuals. Consequently, the values
attributed to the risk reductions become the consumers' valuation of the policy
changes. In this section we discuss whether this view of the problem is rea-
sonable and its implications for the interpretation of our valuation estimates
for policy decisions. This is accomplished by first describing hazardous mater-
ials, particularly the characteristics of those materials that are likely to be
important to any evaluation of the impacts of exposures for ecological systems
and human welfare. Following this discussion, we describe some examples or
scenarios of how hazardous substances might enter the environment. These
examples are then related to the types of exposure risks we have sought to
mode! in our conceptual analysis and to estimate values for in our empirical
work.
2.5.1 Six Categories of Functional and Chemical Characteristics
Hazardous materials can be placed in one of several categories based on
their toxicity and degree of persistence in the environment. In a recent study
of instances of environmental contamination by hazardous materials, the U.S.
Environmental Protection Agency [1980, p, vi] offered a classification of haz-
ardous substances with six categories reflecting the functional characteristics
of substances in commerce and industry and their chemical characteristics.
The following is a brief discussion of the most likely major environmental im-
pacts and fates of each of these categories.
Solvents and Related Organics. This category includes such substances
as benzene, trichloroethylene, chloroform, and toluene. Many of these sub-
stances are acutely toxic in high doses to humans and other organisms On
the other hand, most of these substances disperse rapidly in the environment
and are subject to breakdown to relatively innocuous substances by a variety
of chemical and biological processes. Accordingly, they have relatively short
half lives in the environment. Some of these substances are known o- sus-
pected human or animal carcinogens and thus present a potential threat to hu-
man health, especially from long-term exposures at low levels, Bui due to the
short half fives of these substances, such long-term chronic exposures are not
likely except in the case of contamination of biologically inactive groundwater
2-11
-------
aquifers or in the case of biogenic sources such as chlorination of drinking
water containing naturally occurring organic compounds.
Polychlorinated and Polybrominated Biphenyls. Polychiorinated biphenyls
(PCBs) and polybrominated biphenyls (PBBs) are not readily degraded in the
environment. PCBs are known to be widely dispersed throughout the environ-
ment, and detectable amounts of PCBs are present in the atmosphere around
the earth, in the water column and sediments, and in the tissues of a variety
of organisms (National Academy of Sciences (1979]). PCBs can cause a variety
of adverse effects on nonhuman species and have been classified as a possible
human carcinogen (I nternational Agency for Research on Cancer [1979] ).
Pesticides. This is a heterogeneous category in terms of environmental
impacts and persistence. Some types of pesticides--e. g. , the organophos-
phates--are acutely toxic but degrade quickly in the environment under most
conditions and are not subject to bioaccumulation. On the other hand, the
chlorinated hydrocarbon pesticides have long half lives in the environment and
are subject to bioaccumulation. Long-term exposures to these substances and
some of their degradation products are known to have adverse effects on non-
human species even at low levels. And several of these substances are sus-
pected human carcinogens.
Inorganic Chemicals. This category includes such things as ammonia,
cyanide, and various acids and bases. While many of these substances may
be highly toxic and/or corrosive, they tend to have short half lives in the
environment because of processes such as oxidation (e.g. , as for cyanide) or
neutralization.
Heavy Metals. Examples of this category include mercury, lead, chromi-
um, and cadmium. Heavy metals are obviously persistent in the environment.
But they may become immobilized in sediments. Not all chemical forms of heavy
metal compounds are subject to bioconcentration. Some compounds are known
to be toxic at relatively low doses over long periods of time. And some are
known or suspected carcinogens.
Waste Oils and Grease. Some components of waste oils and grease may
be toxic and/or carcinogenic. But most of the components of waste oil and
grease are biodegradable and have relatively short half lives in the environ-
ment. Waste oils are often contaminated with heavy metals and persistent
organic compounds such as PCBs.
2-12
-------
In terms of enviromental impacts, the categories of hazardous materials
described here differ primarily with respect to two character!sties--the nature
of their toxicity to humans arid other organisms (acute or chronic toxicity)
and their degree of environmental persistence (highly persistent or relatively
short half lives). Furthermore, those substances that are acutely toxic also
tend to have short environmental half lives, while those substances that are
toxic in long-term doses (some of which are known or suspected carcinogens)
also tend to be highly persistent in the environment. For this reason, two
types of scenarios are offered in the following subsection. One type involves
large quantities of acutely toxic substances with short environmental ha if lives,
e.g., organic solvents, some forms of pesticides, and such inorganic chemicals
as cyanide and acids. The other type involves lower quantities of environmen-
tally persistent and chronically toxic substances, such as PCBs, some forms
of pesticides, and heavy metals.
2.5.2 Three Exposure Scenar ios
To provide a more tangible connection between the ways in which hazard-
ous wastes might enter the environment and the role of management policies in
affecting these events, we have constructed three alternative scenarios of pos-
sible hazardous waste spills or uncontrolled releases and the patterns of health
and ecological impacts likely to be associated with them. It should be noted
that the scenario's described here are not meant to reflect all possible signifi-
cant events and ecological end points. Rather, they are meant to represent
the more typical or more likely events involving hazardous wastes and events
for which significant health and ecological and intrinsic effects are likely.
For each case the events are treated as random' occurrences. The principal
purpose of hazardous waste regulations is to reduce the probability of such
events. Thus, these scenarios provide the basis for describing the ways in
which risk reductions might arise from regulations on the disposal of hazardous
wastes.
Table 2-1 provides a summary description of Scenario A: Groundwater
Contamination and Human Exposure. In this scenario acutely or chronically
toxic, substances--e. g., PCBs, chlorinated hydrocarbon pesticides, heavy
metals, organic solvents, or acids--are released from a poorly designed or un-
regulated surface or subsurface storage land disposal site. If these materials
2-13
-------
TABLE 2-1, SCENARIO A: GROWING WATER
CONTAMINATION AND HUMAN EXPOSURE
Substances: PCBs, chlorinated hydrocarbon pesticides, heavy metals, organ-
ic solvents
Event: Unregulated land storage or disposal leads to more or less con-
tinuous leaching of materials through the soil to a groundwater
aquifer used as a source of drinking water.
Impact; Human exposure to toxic materials with the probability of ad-
verse health effects being an increasing function of the accumu-
lated dose for many substances.
Forms of Economic Damages
1. Production/Market Values: Increased cost of treatment or
finding alternative municipal water supplies, once contam-
ination is detected.
2. Use Values: Poor health and increased probability of fatal
disease.
Examples; As reported in U.S. Environmental Protection Agency [1380]:
1. Occidental Chemical Corp., Lathrop, California, 1980 (p. 3)
2. Rocky Mountain Arsenal, Colorado (p. 7)
3. McKin Site, Gray, Maine (p. 14)
4. Hooker Chemtcal, Muskegan Michigan, 1979 (p. 18)
5. St. Louis Park, Minnesota ( 3. 20)
6. Jackson Township, New Jersey (p. 26)
2-14
-------
reach a significant aquifer, they have the potential for contaminating municipal
water supplies and private wells. Thus, there is a probability of human expo-
sure to toxic materials that may cause cancer, mutations, and other adverse
health effects. The probability of a specific adverse health effect given expo-
sure depends upon the potency or toxicity of the substances, the dose received
by each individual, and the genetic endowment and health of the individual
exposed. For carcinogens, the probability of an effect given exposure is an
increasing function of the accumulated dose.
Regulations governing the design of storage and disposal sites serve to
decrease the probability of exposure. Furthermore, regulations establishing
groundwater monitoring programs serve to decrease the expected time interval
between the onset of exposure to contaminated water and the time of detection
at which point avoidance actions can be taken. Thus, regulations of the sec-
ond type may serve to reduce the probability of an effect given an exposure
(depending on the substances involved). The combined effect is to reduce
the probability of an effect--!. e., to reduce the risk of adverse health effects
associated with exposure to hazardous wastes through the contam ra on of
groundwater.
A second scenario describing long-term effects on aquatic ecosystems is
outlined in Table 2-2. This scenario also begins with the unintended release
of materials from storage or improper disposal. In this scenario these materials
reach surface water systems where, because of their lack of biodegradability
and persistence in the environment, they become widely dispersed. Many of
these substances enter the food chain, which is likely to lead to reductions in
the populations of sensitive species and their predators--e. g., fish and fish-
eating mammals and birds such as ospreys and eagles.. Also, accumulation of
these substances in body tissues could render some species of fish unsuitable
for human consumption.
In this scenario it seems reasonable to reflect on several reasons that
individuals might value a risk change. First, contamination of surface waters
increases the probability of exposure by increasing the potential pathways.
Direct exposure through the water itself or "indirect" exposure through the
effects of these substances in the food chain are two cases. Equally important,
the contamination of fish or reduction in their populations could result in a loss
2-15
-------
TABLE 2-2. SCENARIO B: LONG-TERM EFFECTS
ON AQUATIC ECOSYSTEMS
Substances: Polychlorinated biphenyls (PCBs), chlorinated hydrocarbon pes-
ticides, or heavy metals such as mercury, lead, or cadmium.
Event: Unregulated disposal or the breakdown of a poorly designed
disposal site leads to the more or less continuous release of
the substance into the environment. As a result of environmen-
tal transport via runoff, leaching, or migration through soils,
the substance reaches surface water systems. As a result,
the substance achieves wide distribution throughout the aquatic
ecosystem.
Impact: The accumulation of the substance in the food chain is likely
to lead to reductions in the populations of sensitive species and
their predators.
Forms of Economic Damages
1. Production/Market Values: Reduced productivity and har-
vests of commercial fish species; loss of marketability of
fish because of tissue contamination.
2. Use Values: Lost recreation opportunities because of lower
populations of fish, water fowl, etc. Risk to human health
through direct or indirect (food chain) exposures,
3. Nonuse/lntrinsic values: Losses due to increased threats
to endangered species and fragile and/or unique ecosys-
tems .
Examples: As reported in U.S. Environmental Protection Agency {1980):
1. Hooker Chemical, Montague Plant, Muskegan, Michigan, 1979
(p. 18)
2. Waste industries, Inc. , New Hanover County, 1980 (p. 29)
3. ABM Wade, Pennsylvania (p. 35)
4. Taft Forge, Inc., Howell, Michigan (p. 125)
2-16
-------
of use values associated with recreational activities related to that wildlife.
Reductions in the populations of water fowl due to direct toxicity or changes in
the food chain could affect recreational hunting and viewing uses. There could
also be increased risk of the loss of amenity values to the extent that reduc-
tions in the populations of nongame species and the mammals and birds that
feed on them--e. g. , otters, seats, loons, ospreys--reduce the opportunities
for wildlife observation. Also, there could be existence and other intrinsic
values associated with avoiding threats to the populations of species of aesthetic
or emotional significance such as eagles, loons, or seals.
Scenario C, summarized in Table 2-3, also focuses on ecological effects.
Because of their short-term nature, the effects associated with this scenario
are not likely to increase the risk of losing the services that would be associ-
ated with a form of existence values. The substances involved--organic sol-
vents, acids, etc.--are either biodegradable or neutralized rather quickly in
the environment. Thus, although there may be severe reductions in biological
productivity and in populations of sensitive species, once the materials are
dispersed, populations are restored through recolonization and in-migration.
Of course, it is conceivable that for some substances, there could be long-term
ecological effects as well as short-term Impacts.*
As has been noted above, all three of these scenarios have a common
structure in that there is a set of adverse effects that might occur. The
probability of their occurrence depends in large part on the probability of the
release of the substances to the environment. The conceptual analysis in
Chapters 3 through 5 and all but one of the survey questions have been based
on cases resembling Scenario A, Thus, the risks are treated in our model
and described in the survey questionnaire as being experienced by the mem-
bers of a household as a result of land-based disposal of hazardous substances.
There are two aspects of the description in the survey questionnaire that
are important to these scenarios. First, the nature of the exposure to hazard-
*There is also a probability of human exposure and adverse health effects
associated with poorly regulated concineration of hazardous wastes and subse-
quent airborne emissions. And human exposure could occur through contami-
nation of soils and subsequent absorption through the skin or ingestion. The
formal structure of these alternative scenarios is essentially the same as the
all too common groundwater contamination scenario.
2-1?
-------
TABLE 2-3. SCENARIO C: SHORT-TERM
ACUTE ECOLOGICAL IMPACTS
Substances: Organic solvent, acid, or inorganic toxic such as cyanide.
Such substances are acutely toxic but have relatively short en-
vironmental half lives.
Event: An accidental spill or breach from a poorly designed contain-
ment such as r ' on. The substance quickly spreads to near-
by streams or lake^
Impact; Heavy losses of aquatic organisms including fish, and possible
losses of fish-eating species. Because of dilution, neutraliza-
tion, and/or biodegradation, concentrations of the substance in
the environment fall to background levels relatively rapidly.
Species population recover through recolonization and irt-migra-
tion,
Forms of Economic Damages
1. Use Values: Activities such as sports fishing and boating
are adversely affected until the toxic materials are dis-
persed or neutralized and the populations of the target
species restore themselves.
2. Nonuse/ lntrinsic Values: Not likely to be significant.
Examples: As reported in U.S. Environmental Protection Agency [1980]:
1. Kernersville, North Carolina, Reservoir (p. 27)
2. Byron, Illinois (p. 18).
2-18
-------
ous wastes is not defined in physical terms but rather in terms of the outcome
it implies. Exposure is described as sufficient to lead to death through an
initially undefined health effect. The risk is described as the risk the "inter-
viewed individual (or other member of the household) wilt die in 30 years as a
result of exposure. So the presentation maintains that the exposure level (an
issue discussed in Scenario A) would be sufficient to impose a risk of the
health effect. After questions associated with household exposures were dis-
cussed, our survey questionnaire did attempt to determine whether individuals
had additional willingness to regulate the disposal of hazardous wastes to lower
risks experienced by fish, wildlife, and plants. This is the second aspect of
our framework that relates to the scenarios. That is, we attempted to capture
just the ecological effects that were highlighted in Scenario B without a full
description of the mechanism that leads to the risks to these species. It was
done in a way that attempts to isolate each motivation for valuing a risk
change, but did not explicitly distinguish user and nonuser motives for valuing
the wildlife.
2.6 SUMMARY
This chapter has provided a brief overview of the organization of our
conceptual analysis of individual decisionmaking under uncertainty and its rela-
tionship to the issues we suggest are important to valuing regulations govern-
ing the management of hazardous wastes. These regulations are treated as
reducing the risk of exposure to hazardous substances. Risk in our analysts
is treated as synonymous with the probability of a welt-defined event. The
event is an exposure to these materials that is sufficient to lead to a second
stage risk of death. The second risk is explained to be the result of the in-
dividual's health and heredity.
Finally, to explain how our analysis can be related to specific events and
the associated policy actions involving hazardous wastes, we presented three
examples of how hazardous wastes might enter the environment and described
the scenario that is most closely aligned with the implicit circumstances under-
lying our conceptual and empirical analyses.
2-19
-------
CHAPTER 3
MODELING BEHAVIOR UNDER UNCERTAINTY: A HEURISTIC REVIEW
3.1 INTRODUCTION
Models of how individuals respond to uncertainty have been an important
part of the theoretical contributions to microeconomics in the last 50 years.
Today, interest in the results of these theoretical advances and in the pros-
pects for extensions of this work seems especially great. There are at least
two factors contributing to this interest. First, and perhaps most impcrtant
to the objectives of this research, there has been a growing public concern
over risks that are imposed on households without their consent. These risks
can arise from the actions of firms, other households, or the public sector.
They take many forms — ranging from what is perceived as inadequate testing
of new products to insufficient safety provisions in new technologies. More-
over, in some cases, there may be the perception that past decisions were
based on incomplete information or failed to give appropriate attention to the
future risks accompanying specific actions. The past disposal practices for
hazardous wastes are a good example of discussions in this last category.
The "surprises" associated with past disposal practices in a large number of
cases such as the widely publicized examples of Love Canal, New York; times
Beach, Missouri; or Newark, New Jersey contribute to this perception. As a
consequence, despite what many observers have suggested is a relatively low
risk environment in the United States,* public policy has increasingly focused
*One of the most widely cited studies identifying this seemingly contradic-
tory behavior is associated with Douglas and Wildavsky [1982] . More recently,
Slovic [1984] has noted that recent poiIs of corporate executives, members of
the banking and investment community, members of Congress and their aides,
Federal regulators, and the general public seem to suggest that "regardless
of whether things actually are riskier, most people think they are now more
risky." (Slovic [1984], p. 2).
3-1
-------
on the risk implications of new technologies arid the role of social regulation,
and public policies in general, for risk management.*
At the same time, there has been growing concern over the validity of
economic models for describing individual behavior under uncertainty. Experi-
mental tests of the assumptions and predictions of the most widely used frame-
work for modeling behavior under uncertainty — the expected utility model - -
appear to have found important violations of the model's assumptions and incon-
sistencies with its predictions (see Schoemaker {1980] and Machina 11383}).
These results have, in turn, stimulated research, both theoretical and empiri-
cal, to attempt to evaluate the plausibility of these findings and to understand
the reasons for them. To date, there has not been a clear reconciliation of
the available experimental evidence with the predictions of conventional eco-
nomic models of individual decisionmaking. The available alternatives to the
expected utility model all suffer from significant limitations that restrict (or
preclude) any one of them from serving as an effective basis for empirical
analyses of individual behavior.^
This chapter cannot do justice to the research in both of these areas.
Summaries have already occupied several overview volumes of varying technical
detail. + Our conceptual analysis will largely accept,, as a maintained hypothe-
sis, the expected utility model as a description of individual behavior under
uncertainty. While the empirical analysis is based on this conceptual frame-
work, it has been designed to allow consideration of the relevance of the ex-
pected utility model for explaining individuals' responses to risk. Chapter 7
describes the relationship between our conceptual and the empirical analyses.
The objective of this chapter is to explain the overall features of the expected
utility model and to discuss in more detail several specific aspects of the
framework that will be particularly relevant to our empirical analysis. More-
*See Lave [1981] and Huber [1383, 1984} for discussions of risk manage-
ment issues in a policy context.
^See Weinstein and Quinn [1983a] for a good overview of some aspects of
these models and their limitations.
tSee Hey [ 1979], Machina [1983], or Schoemaker [1980] for discussions of
the theory and of limitations of the expected utility framework. See Fischhoff
et al. [1981], Viscusi [1983], Lave [1981], or Crouch and Wilson [1982] for
discussions of aspects of the treatment of risk in public policy decisions.
3-2
-------
over, in developing this review we will consider, briefly, some of the dif f xul -
ties raised with the expected utility framework in relationship to the objectives
of our analysis.
Section 3.2 begins this overview with a review of the assumptions of the
model, an outline of the state-preference approach for describing it, and a
discussion of the implications of the treatment of contingent claims as claims
to commodities versus claims to income. In Section 3.3, the expected utility
model is used together with the assumption of state-independent utility func-
tions to describe the implications of changes in the probability of detrimental
events and the measurement of risk aversion. Section 3.4 describes the ra-
tionale for state-dependent preferences and reconsiders several of these issues
using this specification for consumer preferences. Section 3.5 discusses some
of the limitations of the expected utility model with special attention to the
issues of potential relevance to our empirical analysis; Section 3.6 summarizes
the chapter.
3.2 THE EXPECTED UTILITY FRAMEWORK AND CONTINGENT CLAIMS
Two conceptualizations of the process of individual choice under uncer-
tainty have been frequently used in economics. In the one we shall use--the
state-preference approach—the objects of choice are redefined from what is
assumed in conventional descriptions of consumer choice under certainty. They
become contingent commodity (or income) claims. This means that they are
entitlements to goods or services under specified states of nature. If it is
assumed that an individual is uncertain over the state of nature that will be
realized at the time his consumption decisions must be made, then this frame-
work describes the individual as planning consumption choices contingent jpon
which state of nature is realized. These plans are formalized through the
selection of contingent commodities. These commodity claims are valid only if
the state that is part of their description is real ized. Thus, for example, an
individual cannot exercise a claim to financial resources in the event of a dis-
abling injury unless the injury is experienced.
An alternative description of behavior under uncertainty defines the ob-
jects of choice as specific parameters describing the probability distributions
for something of interest to the individual, such as a commodity or income.
What is important is that the specified source of uncertainty and the descrip-
3-3
-------
tion of the features of that uncertainty affect the utility realized by the indi-
vidual. for example, an individual might be selecting actions that would
change the mean income or its variance. The state-preference approach is
more general than this parametric formulation of the problem. Of course, as
we might expect, there are assumptions that can be imposed on the state-
preference formulation to reduce it to this more restrictive approach. How-
ever, for our purposes, these assumptions are too restrictive. Therefore,
since our analysis will use the first format, no attempt is made here to sum-
marize the second.
The origin of the contingent claims approach is usually associated with
Von Neumann and Morgenstern (1947], who deduced maximization of expected
utility as the type of behavior implied by a set of assumptions on the features
of decisionmaking under uncertainty. These assumptions are usually described
with some variation on three axioms: transitivity of preferences over lotteries
(or prospects), continuity of preferences over lotteries, and the independence
axiom. A prospect or lottery involves a listing of the outcomes in each state
of nature and a specification of the probabilities for each state. Thus, if a
prospect, A, involves two states, and if state one yields Wr with probability
p, and state two yields W2 with probability (1-p), then prospect A would be
described as follows:
Prospect A = (Wj, W2, p, (1-p)) .
Transitivity implies that if prospect A is preferred to prospect B, and
prospect B is preferred to prospect C, then A will also be preferred to C.
Continuity is also similar to the assumption in conventional models without un-
certainty, If a sequence of prospects converges to a given prospect, then
the utility generated by the sequence wilt converge to that generated by the
given prospect.*
Independence implies that if an individual prefers prospect A to B, then
this preference should not be affected by whether the choice of A over B is
in simple terms or if it is as a possible prize in a compound lottery (i.e., the
choice of a lottery involving A and another event C versus 8 and C where
*See Machina [1383], pp. 5-7, for a more complete description of these
conditions.
3-4
-------
the odds of A and B are the same in the two compound lotteries). The impor-
tance of the independence assumption is that it gives the expected utility
framework its empirical content by restricting the form of the preference func-
tional (described over the distributions of lotteries) to be linear in the proba-
bilities- -or the expected utility.
With this background we can turn to the form of the model usually pre-
sented in describing specific decision problems. To start the process, assume
we have a utility function, U (X1( X2), expressed in terms'of commod!t:es Xx
and X2. Since we are dealing with contingent claims, to describe the choice
process we must distinguish claims by commodities and states. If there are N
commodities and S states, we wouid consider N'S contingent claims for the
commodities. The conditions governing the availability of any good can be
different depending upon the state of the world one is considering. In our
example, if there are two states of the world, then there must be four contin-
gent claims for a complete description of all possibilities.*
If we assume the probabilities of states one and two are p and (1 -p),
respectively, then the Von Neumann-Morgenstern utility function, V(.), will
be given as;
V (Xu, Xl2, X21 , X 2 2 / P* 1~p) ~ P U(xilf X ! 2)
(3.1)
+ (1-p) U (X21, X22) ,
where
X.. = contingent claim to commodity j in state L
To describe how the representative individual responds to uncertainty we
must specify the constraints imposed on his maximization of Equation (3.1).
Before doing this, however, we should note that concavity of U(. ) assures
that V(. ) will be concave. Therefore, Von Neumann-Morgen stern indifference
curves will resemble ordinary indifference curves as in Figure 3-1. This indif-
ference map is drawn holding one of the commodities constant in both states
(i.e., X12 and X22) and allowing the state of nature to vary for the other.
The slope of the indifference curve is given as:
*We have not discussed the potential role of securities markets as a basis
for reducing the number of markets from N S to N + S. See Arrow [1964]
and Nagatani [1975] for discussion of these cases.
3-5
-------
of Xi in Stats 2
Figure 3-1, Illustration of Von Neumann-Morgenstern indifference curve.
3-6
-------
d xtl
(1 p) ^2 1 ' ^22 )
^ ^21
(3.2)
This is simply the negative of the ratio of the expected marginal utilities for
the first commodity in each state. Of course, if we held state constant and
considered the slope in terms of Xn and X12, commodities one and two in state
one, then it would be the conventional ratio of the marginal utilities without
probabilities since they would be equal. This feature will be important to un-
derstanding the simplification of this model that has routinely been used in
the literature. The contingent claims are generally described as claims to in-
come and not claims to commodities. Moreover, it has been argued that this
approach can be adopted without a foss in the generality of the conclusions.
We shall argue in what follows that this concfusion is not entirely true. To
do so, however, requires consideration of the constraints to individual c no ice.
Therefore, we must describe what limits the individual's efforts to maximize
expected utility.
The limits arise, as in conventional models of consumer choice, with a
budget constraint. This constraint can be formulated in a variety of ways.
In Equation (3.3) we maintain that the individual has an income level, y, that
constrains the planned choices of contingent commodities:
We could have assumed that the individual was endowed with certain levels of
contingent claims. Given fixed prices for the claims, this would also lead to
a fixed income for planned consumption.
Optimal consumption plans in this framework require that the ratio of
probability-weighted margina! utilities for claims associated with the sam° <¦ :>m-
modity in different states equals the ratio of the relevant prices and that, for
different commodities in the same state, the ratio of the marginal utilities
equals the relevant price ratio, These two cases are given in Equations (3.4)
and (3.5) below;
V ~ SH *11 + s 12 *12 + s21 *21 + s22 *22
(3.3)
where
s.. = price of the contingent claim for commodity j in state i.
3-7
-------
(3,4)
(3.5)
where
MU = marginal utility of X,,,
A*. II
i| 1
It is important to recognize that these prices are for contingent claims,
not for the commodities themselves. Two aspects of this distinction will affect
the interpretation of our analysis. The first concerns the relationship between
the prices for claims and the prices of commodities and the probabilities of
states of nature. The second concerns the relationship between what might
be designated the ex ante relative prices of the commodities (not the claims)
and the ex post prices of these commodities. The first of these issues is dis-
cussed below. The second relates to an exchange between Arrow [1375] and
Nagatani [1975] and the extent to which a full set of markets for contingent
claims would lead to an ex post efficient allocation of resources. Since this
second issue is not directly relevant to our analysis, it will not be developed
further here.*
Clearly, the interpretation of these first-order conditions depends on the
relationship between the prices of contingent commodity claims and the proba-
bilities for each state. If, for example, we assume that there exists a set of
ex ante prices for the commodities Xj. and X2 involved in these claims and that
the prices for the contingent claims to them are simply the probability-weighted
counterparts to these prices, then we can see directly an explanation for the
simplifications used in most contingent claims models. With these assumptions,
the prices for contingent claims would be defined as follows:
*As Nagatani [1975] noted, it is important to recognize the prospects for
differences in the ex ante and the ex post prices that might influence planning
for purchases of contingent claims. If we are to assume individuals know the
ex post prices, we must consider the mechanisms that permit this information
to be realized ex ante. We wilt not consider these issues at this point but
will return to the distinction between ex ante and ex post behavior in discus-
sion of the appropriate basis for welfare measures associated with policies that
affect risk in Chapter 4.
3-8
-------
S12 - P t2
S11 = P tj
S21 = (1-p) tt
s22 ~ (1"P) *2
(3.6)
where
t. = the ex ante price for the jth commodity.
With substitutions it can be demonstrated that the allocation of resources to
claims for commodities In a given state is determined by the same conditions
as in the certainty case—the marginal rate of substitution between the two
goods equals the ex ante price ratio. To the extent ex ante commodity prices
are the same as ex post commodity prices, then we can expect that the planned
consumption choices under fair markets for contingent claims will lead to the
same relative consumption incentives as the ex post relative prices would.
Moreover, the optimal allocation among claims to the same commodity for differ-
ent states of the world implies that the relevant marginal utilities will be equal-
ized .
These results and the assumptions associated with them provide the basis
for understanding the implications of defining contingent claims in terms of
income rather than in terms of commodities. In contrast to discussions of un-
certainty used to develop models for analyzing option value (see, for example,
Hartman and Plummer [1981], Plummer and Hartman [forthcoming], or Freeman
[1984a]), our analysis has not "attached" the uncertainty to a specific variable
affecting preferences. Rather, the individual is assumed to be uncertain over
the state of nature but to have access to a complete set of markets for contin-
gent claims. If these markets are actuarially fair, and If it is reasonable to
assume ex ante commodity prices are equal to ex post prices, then it is clear
that there is no need to distinguish commodities in evaluating an individual's
allocation, of resources with state-independent preferences. Indeed, we could
further relax the assumptions specified above by allowing the prices of contin-
gent claims to be a product of a function of the probabilities and the ex ante
prices for each good. Provided this function was the same for all commodity
claims associated with a common state of nature, the marginal conditions gov-
erning their selections would be equivalent to the certainty case. Of course,
selections of a commodity claim differentiated across the states of nature would
3-9
-------
be affected. These effects could be as easily described using income claims
in lieu of commodity claims. Consequently, in nearly ail work with contingent
claims models, it has been assumed that the utility function is actually an in-
direct utility function expressed in terms of income and commodity prices.
Ex ante selections of commodities are assumed to be identical to ex post selec-
tions. Uncertainty and the ability to diversify risk were assumed to affect
the allocations to income claims in each state. Since commodity choices are
conditional upon available income, it was felt this simplification did not affect
the description of individual behavior. We shall argue that, with state-depend-
ent preferences, these same arguments do limit the relevance of the analysis.
3,3 RISK AVERSION AND PROBABILITY CHANGE: STATE-INDEPENDENT
UTILITY FUNCTIONS
Risk aversion is associated with a concave utility function,* It is often
convenient to have an index of the degree of risk aversion. One of the most
popular of these measures is associated with the work of Arrow {1385] and
Pratt [1964], ft has been described by absolute and relative measures of risk
"
aversion. The first, absolute risk aversion, can be defined in terms of the
change in the marginal utility of income. Using the arguments discussed in
Section 3,2, we replace our utility function with the corresponding indirect
utility function and assume the prices of commodities are held constant. Let
designate Che state-independent, indirect utility function associated with
claims to income in state i, y.. Since the prices of commodities are held con-
stant, they have been omitted for simplicity in exposition. The Arrow-Pratt
index of absolute risk aversion, 6, is given in Equation (3.?), with the elasti-
city formu I at ion--relative risk aversion, r--given in Equation (3.8):
etVj ) =
d^n
dyf
_
dy,
(3.7)
*The classic paper on the implications of risk aversion for behavior is
Friedman and Savage [1948]. More recently, discussions of the concept of
option value have debated the appropriate definition of risk aversion in the
evaluation of the sign of option value. See, for example, Schmalensee [19721
and Bohm [1975].
3-10
-------
1 d2p
~
r(y, ) — y, ©
-------
p fj(y+a) + (1-p) M(y ' ^j?|) ~ M(y) + a P
4 . a n 4
dy dy
(3.11b)
1 2 « U 1 , D , 2 d M .
+ 2 a P + 2 ( V? ' 82 " 1 '
where
e and t = the remainders fo the Taylor series expansions.
Simplifying terms in Equation (3,11b) and setting Equation (3.11a) equal to
Equation (3.11b) (as implied by our definition of n in Equation (3.9) and Equa-
tion (3,10) ), we have:
»j(y) - ^7= +
p a2 + (1-p)
te):
Further simplification yields
. 5MH= 1 M!m Var (A)
dy 2 dy2 var K'} '
(3.12)
(3.13)
where
Var (A) = variance of a (i.e., Var (A) = E( A2) - (E( A))2).
E( A) = 0 by definition of the prospect as actuarially fair.
Therefore, the risk premium is a multiple of the variance associated with the
uncertain prospect:
n ~ 2 Var *
(3.14)
The role of n in the shape of the utility function can also be illustrated
with Von Neumann-Morgenstern indifference curves as in Figure 3-2. Let y
designate the constant income starting point for evaluating the degree of risk
aversion. It is given as point A on the 48° line with expected utility given a
V2. The uncertain prospect described earlier is illustrated by point B, a
movement along the line designating state-dependent payments that are actu-
arially equivalent to y. The expected utility of B, V x, is less than at point
A. To determine n, we need only consider the maximum constant payment
(regardless of state) that would leave the individual indifferent to being at
point B. This is given by the intersection of Vj with the 45° line, as at point
C. Clearly, the flatter the indifference curve, the smaller the difference be-
3-12
-------
Figure 3-2, Illustration of Arrow-Pratt measure of risk aversion.
3-13
-------
tween C and A. For given Var( A), then, large values of 0 (yp will be asso-
ciated with utility functions that are "more concave" and exhibit greater risk
aversion. Smaller values of 0 (y.) will be associated with smaller degrees of
risk aversion.
Before discussing another aspect of the expected utility model with state-
independent preferences, it is important to note that our definition of the
measure for the degree of risk aversion was able, because of the specification
of the utility function, to avoid an important issue. This issue concerns the
reference point and institutional framework assumed to be relevant in the mea-
surement of the individual's degree of risk aversion.
in the case of state-independent preferences, the point of constant and
equal income claims across states will correspond to the optimal selection made
by an individual facing actuarially fair markets for contingent claims. This
result is readily derived by considering maximization of the Von Neumann-
Morgenstern utility function subject to a constraint on purchased of income
claims (in a two-state framework), as in Equation (3.15) below:
L = V(yj, y2) + A [y - rx y2 - r2 y2) , (3.15)
where
v(yt, y2) = p m(yi) + (i-p> M(y2)
r. = price for ith contingent claim for income.
The first order conditions imply that the ratio of expected marginal utili-
ties of income claims will equal the price ratio for those claims, as in Equation
{3.16):
MiLii
—5V-, = ? . (3.16)
(1-P)itoa
Actuarially fair markets imply ^ . Therefore, claims will be allocated to
r r 2 1-p
equalize the marginal utilities realized in each state. Since the utility function
is state independent, we can expect equality of total utility and of claims to
income. Thus, the 45° line is simultaneously the locus of income certainty,
utility certainty, and equality of marginal utilities. Selection of equal claims
3-14
-------
to income in deriving a relationship between the curvature of the utility func-
tion and the risk premium has rather special implications that we return to
below,
As indicated earlier, our objective is to develop a conceptual framework
and empirical estimates for a change in the probability of exposure to hazard-
ous wastes. Thus, it is important to consider how a change in the probability
of a state would affect an individual's planned consumption choices. In the
state-independent case, this is easily described. The Von" Neumann-Morgen-
stern utility function is a probability-weighted function of the utilities realized
in each state. A change in the probability of any one state changes all of
the weights (since these weights always sum to unity by definition). Thus
the indifference map must shift as the probability changes.
For the state-independent specification, the indifference curves will pivot
about the 45° line. Along this line the claims to income will be equal. Thus,
the slope of the indifference curves are given by the probability ratio. The
marginal utilities are equal at the point of income equality (see Equation (3.2)
for the case of commodity claims or Equation (3.16) for that of claims to in-
come). Figure 3-3 illustrates the process graphically. A movement from Vx
to Vi is associated with a flattening of the slope-~a decrease in the probability
of state one and increase of that for state two.
Changes in p affect the risk experienced by the individual. This is easi-
ly established by considering the change in the risk premium, as in Equation
(3,17) below:
3^ _ 1 a/. . \ 3 Var (A) /-in \
ap - 2 8(yi) • (3-17)
The results are not as clearcut when state-dependent preferences are
used to describe the individual's responses. In this case, the measure of risk
aversion will be seen to also be affected by changes in the probability. It is
also important to recognize that the measure of risk itself depends on what is
assumed about the opportunities available for adjustment to risk. The most
direct way of illustrating the importance of this point is to note that we need
not initiate our evaluation of an actuarially -fair gamble at a point along the
income certainty locus.
3-15
-------
V2"
Figure 3-3, Illustration of change in probability on
Von Neumann-Morgenstern indifference curve.
3-16
-------
There are actually two distinct issues being raised in these comments.
First, by starting the analysis along the income certainty focus, we implicitly
assume that the gamble offered is the only risk facing the individual. Second,
our analysis of the risk premia assumes the individual has the equivalent of
access to actuarially fair markets. It is difficult to illustrate both points on a
single diagram. The reason is that both the indifference curves and the po-
tential budget constraints (in the case of actuarially fair markets) change in
response to changes in the probabilities of states of the nature. To attempt
to describe the effects of these problems we have made several simplifying
assumptions. The individual is uncertain about which of two states of nature
will prevail. The likelihood of each state corresponds to the odds of a hypo-
thetical fair gamble that will be presented to him. In the absence of the gam-
ble, but with access to fair markets for contingent claims, the individual would
-P
select a point along a budget constraint with slope and the position would
be affected by the income assumed to be available for contingent claims. Since
the utility functions are state independent, this selection would He along the
45° line and would yield a starting point for evaluating the effects of the fair
gamble that is equivalent to the case of certainty. However, if the individual's
choices for adjustment to the first type of risk are not actuarially fair (i.e.,
his budget constraint does not have slope y^p), then the starting point is not
along the 45° line. For example, using Figure 3-4, if the individual faces a
budget constraint given by the line labeled T (with slope - 1-1), point A
r 2
would be selected as the constrained expected utility maximizing choice of
planned claims. It would not correspond to equal allocation of the planned
budget to income claims for each state.
Starting at this reference point and evaluating the second source of un-
certainty, the fair gamble ieads to the potential for several different measures
of the risk premium. One could, for example, consider following the Arrow-
Pratt logic by asking what is the maximum amount the individual wouid pay
regardless of the state (i.e., state-independent payment) rather than experi-
-p
ence the gamble. The fair gamble is given by the line with slope through
A to B. With expected utility held at the level given by V x, we could consider
equal payments from the point A to reach expected utility V1. Constructing
a line through A parallel to*the 45° line, we can determine the risk premia by
3-17
-------
Figure 3-4. Measuring risk aversion in absence of actuarially fair markets.
3-18
-------
the difference" in the coordinates of A and C, or ft in the diagram. This will
differ from the value implied by initiating the analysis on the 45° line. Alter-
natively, we could consider adjustment to the expected utility implied by the
game (i.e., Vl) using the contingents claims markets assumed to be available
to the individual. This would amount to comparing the coordinates of the tan-
gencies of T with V2 versus R with V! and would imply state-dependent risk
premia.
It should be acknowledged that in order to place both-risks on the same
diagram we have made quite restrictive simplifying assumptions. These
assumptions can have important implications for the relationship between risk
premia and the Arrow-Pratt measures of risk aversion (see for example, Kihl-
strom, Romer, and Williams [1981] and Pratt [1382]).
The objective of this section has been to describe conventional measures
of risk aversion using a state-independent specification for individual utility.
This analysis suggested that the conventional analysis of risk premia as mea-
sures of an individual's degree of risk aversion may well be quite sensitive to
the characterization of the individual decision process. Assumptions of only
one risk, actuarially fair markets, and the income certainty reference point all
influence the measures of risk premia derived from these models. This implies
that when one evaluates the relevance of the expected utility framework in
real world circumstances, the limitations imposed by these assumptions may
well be as important to the observed performance of the framework as the ex-
pected utility model itself. Indeed, when the assumption of state independence
of preferences is relaxed, the problem of gauging the size of the risk premia
expected to be associated with risk adverse preferences becomes even more
difficult.
3,4 THE IMPLICATIONS OF STATE-DEPENDENT UTILITY FUNCTIONS
A state-dependent description of preferences has often been regarded as
a controversial, if not an inconsistent, specification for an individual's prefer-
ences, Early discussion of this possibility by Malinvaud [1972] seemed to imply
that the specification simply reflected an inadequate specification of the model.
The recent literature has seen a change in attitude, with growing acceptance
of arguments made by Cook and Graham [1977] and Arrow [1974] (earlier in a
somewhat obscure source) on the potential importance of the state-dependent
3-19
-------
specification. * White Arrow's arguments focused on the modeling of insurance
decisions involving health risks, we will argue that they have relevance to a
wide range of risks. In justifying the state-dependent specification for his
example, Arrow observes that
income is not the only uncertainty, especially in the context of health
insurance, and only under special and unrealistic circumstances can
it be held that the other uncertainties have income equivalents.
Put loosely, the marginal utility of income will in general depend
not only on the amount of income but also on the state of the indi-
vidual or, more generally, on the state of the world. (Arrow [1974]
p, 2, emphasis added)
Arrow also suggests that a state-dependent formulation can be derived from
an axiomatic framework provided that we maintain that there are effects to an
individual of being in a state that do not correspond to or translate into de-
cisions on purchases of goods or services or other types of income allocations.
Thus, an individual's utility is affected by the state of nature, but there is
no explicit relationship between how it is affected and changes in market-based
economic activities. The state-dependent formulation implies that some conse-
quences are not only impossible but irrelevant to some states of the world.
As a result, the axiom (often postulated in the conventional framework) sug-
gesting that a consequence under any one state of the world is possible under
any other cannot be accommodated with the state-dependent specification.
The use of a state-dependent formulation has a number of implications.
Three are of direct relevance to our analysis:
1. Under a state-dependent specification, planned consumption
activities will be distinct from ex post consumption choices even
if the individual is assumed to face complete and actuarially
fair markets for contingent claims with ex ante and ex post
commodity prices equal. t
*More recently, state-dependent utility functions have received consider-
able attention. See Kami [1983a, b], Kami, Schmeidler, and Vind [1983], and
Dionne and Eickhoudt [1983).
tThis conclusion is simply an alternative statement of the fact that, with
state-dependent preferences, the ex ante or planned expenditure function will
not necessarily equal the expected value of the ex post expenditure functions
associated with consumption choices made under each state of nature.
3-20
-------
2. Measurement of risk aversion in a state-dependent framework
is a more complex and arbitrary process when the measures
are required to have a relationship with a risk premium.
3. Violations to the behavior implied by the expected utility frame-
work can largely be explained within a state-dependent specifi-
cation for utility.
We will consider the first two of these implications in this section and return
to the last as part of the brief overview, in the next section, of some of the
violations to the expected utility framework encountered in studies of individual
behavior under experimental conditions.
To consider the first implication of the state-dependent formulation, we
must return to the specification of individual choice developed in Section 3,2
and modify Equation (3.1) to reflect the state dependency as given below:
V (Xlw X,
2 ' X211 X22 / P t (1-p)> = P Uj, (X1 j , X12)
(3.IS)
+ (1-p) U2 (X21, x22) .
The subscript to each utility function indicates different preferences for com-
modities one and two. These differences can arise, as Arrow | 1974] and Cook
and Graham [1977] have suggested, through some omitted factor that affects
preferences and is conveyed "outside the available markets" with the state of
nature (i.e., U. (X. , X } might equal UCX.-, X.„, z.)). For our purposes,
I II I c J I I cL i
this source of state dependency need not be specified at this stage of our
analysis. However, it will be more important to the planning for our empirical
results, indeed, state dependency can be regarded as a reflection of our ig-
norance of the factors that influence individual utility. Therefore, by adopting
this specification to model decisions under uncertainty, we are acknowledging
that there are aspects of the events at risk (or the risk itself) that affect
individual well being and that we cannot identify. Given our incomplete infor-
mation, it is prudent to assume that state of the world can matter to an indi-
vidual's utility.
Repeating the constrained maximization of Equation (3.19) subject to a
budget constraint as defined with Equations (3.3) and (3.8) yields require-
3-21
-------
merits for equality between the marginal rates of substitution and relative
prices for the two commodities in each state and equality of the marginal utility
for each commodity overstates as in Equation (3.20) below:
MU * = MU2
*11 *21
(3.20)
MU * = MUA
*12 *22
However, these conditions do not imply that the levels of contingent claims
for each commodity will be equalized across states. Accordingly, the allocation
of income to each state will differ. Thus, the locus of income certainty and
that for utility certainty will diverge, as illustrated in Figure 3-5. As Arrow's
justification for state-dependent preferences suggested, the nature of the
change in the marginal utility of income across states will determine where the
optimal allocation of income among contingent claims will be in relationship to
these foci. Indeed, this relationship forms the basis for the Cook-Graham
classification of commodities into irreplaceable (both normal or inferior) and
replaceable.
Moreover, these considerations have direct implications for the use of the
expected utility framework. Consider, for example, the expenditure function
that woufd describe an individual's planned consumption of the two commodities
as price, income, and probabilities changed. This function is defined, for
the two-state, two-commodity case, by Equation (3.21):
Minimize E. ~ s^ 2X 12 ^21^21 s 2 2 X 2 2
(3.21)
subject to V = pUjCXn, X12) + (1-p)U2(X21 ,X22) ¦
The planned expenditure function would then be given as follows:
E = E(slu s12, s21, s22; p, (1-p), V) . (3.22)
With the state-independent specifications, we assume .) ~ U2(. ). Moreover,
with actuarially fair markets for contingent claims, Equation (3.8) would
describe the relationship between the prices of claims and the probabilities.
3-22
-------
Figure 3-8, The distinction between income certainty and utility
certainty loci with state-dependent preferences.
-------
Taken together, they imply that it - (Vp)t2 " AC1-p>3fr 1 0
(e) f£ = V - pU(X11(X12 ) - (1-p)U(X25 ,X22) = 0''
then (a) and (c) together with (b) and (d) imply that Xxx = X2i and X12 =
X22. Consequently, the problem can be reduced to an equivalent statement
omitting distinctions for states of nature,
^This requires that ex ante and ex post prices will be the same.
3-24
-------
when these expenditure functions are different, the measures of changes in
well-being they imply will be different. Consequently, we will return to the
specific implications of these distinctions in Chapter 4,
However, when we relax the assumption of state-independent preferences,
this reduction of the planned expenditures to the ex post function is not possi-
ble. This is easily established by considering the change In planned expendi-
tures with a change in the probability of state one under the two cases.
We can, without loss of generality, use claims to income In the state-
independent case. Maximization of expected utility subject to fair markets will
imply that ^ (yi) = (y2), that u(yt) - (y2), and, therefore, that yi = y2.
The change in the ex ante expenditure function is given in Equation (3,23)
below:
|| = " X21+ % *22) . (3.24)
9Xn
If we treat claims to income as akin to Hicksian composite commodities (i.e.,
with fixed prices t, and t2), we can rewrite Equation (3.24) as follows;
If = H2isEl + (y' -yz) - (3'25)
3y»
~The total differential for the Von Neumann-Mor gen stern utility function
in this case is given as:
dv = M(yi) - m(y2> + p M'Cyi) ^ + d-p) m'(yz) ^ .
Since m<¥i) = h(Y2)' and jj'(yi) = m'(Y2> yi = Y2- 1° addition, the constancy
of expected utility implies that
p = - (-|.p) §Y2
P aP 11 Pi aP ¦
3-25
-------
where the marginal utility of income for state one is given as follows:
9U 1
iMl = 3Xn
3y, t:
The measurement of risk aversion with state-dependent preferences must
also be distinguished from the state-independent case. Of course, it should
be acknowledged that the difficulty arises in relating measures of the concavity
of each state's utility function to a single risk premium (or to a set of risk
premia). It is always possible to measure risk aversion in terms of the degree
of concavity of each state's utility function. What is at issue is developing a
numerical index of how concavity would affect a risk averse individual's re-
quired risk premium when confronted with an actuarially fair gamble.
Two aspects of the extension to the state-dependent case are important.
First, as we noted for the state-independent case, we must define the appro-
priate reference point. Second, it is important to consider the role of the
institutions available for diversifying risk in judging an individual's risk premi-
um, However, that analysis did not pursue the full implications of institutions
for risk premium. Indeed, the Arrow-Pratt definition imposes an institution
(or payment mechanism) by assuming constant payments across states.
This requirement is not essential to the characterization of the individual's
attitude toward risk. Before developing this argument in detail, consider
Kami's [1983a] approach for measuring risk aversion with state-dependent
preferences. He observed that, in the state-independent case, the coincidence
of the income and utility certainty loci together with the locus of equilibrium
selections of claims to income in the presence of actuarially fair markets elimi-
nates a difficult choice--what is the relevant reference point and how should
it affect the adjustments assumed possible for the individual?
Kami [1983a] argues for the locus of claims to income that will assure
equality of the marginal utilities of income. In evaluating an actuarially fair
gamble, regardless of the starting point, his analysts would measure risk
premia in terms of points equivalent to the starting and ending positions in
terms of expected utility but lying on the locus of equal marginal utilities.
Figure 3-6 illustrates his case, where A designates the individual's initial
3-26
-------
3-27
-------
endowment. Regardless of the size of the change that is offered as a fair
gamble, Kami's proposal focuses on the most preferred point along AA'--the
point B where expected utility would be greatest. At point C, the individual
realizes the same expected utility as A. The maximum amount that would be
paid for these fair prices corresponds in this setting to state differentiated
payments--/!! in state one and n2 in state two (with n2 t in general).
This analysis could also have considered measuring risk in terms of the
utility certainty locus, where there is no risk experienced. If D D' designates
this locus in Figure 3-6, a different set of premia would have been relevant--
namely, 52 and 6X. Which choice is best depends on how the risk measure is
to be used. And it is for this reason that our discussion of the role of differ-
ent opportunities for diversifying risk must be considered in appraising the
degree of risk aversion of an individual. Simply stated, an individual's aver-
sion to the risk introduced by an uncertain situation wilt depend on his exist-
ing opportunities to adjust to risk.
The definition of the Arrow-Pratt measure selects as a reference point
the riskfess locus of equal total utility in each state for the case of state-
independent utility functions. This is only relevant if this point characterizes
an individual's initial position. Moreover, the definition of the risk premia
maintains state-independent payments. For the case of state-independent util-
ity with actuarially fair adjustment opportunities, this point will be the selec-
tion. However., it will not be the selection if opportunities for adjustment do
not imply constant risk premia regardless of state. Consider the case given
in Figure 3-4. If the individual were allowed to adjust based on the existing
prices of claims (i.e., and r2), the expected utility equivalent of B selected
would have been G, not C. Our measure of risk aversion would depend upon
how we treated the state-differentiated premia implied by G. Similarly in the
state-dependent case, the Kami reference set could be redefined to be the
locus of income claims where MU .j/MU ^ - r^r2 and risk premia measured
with reference to it. This approach would also yield state-differentiated
premia, and its implications would depend on how they were weighted in deriv-
ing a composite index of risk.
At first this may seem to simply add to the confusion associated with
characterizing the degree of risk aversion. We think this is an inappropriate
3-28
-------
interpretation. The Arrow-Pratt concept of a risk premium is simply one way
of characterizing an individual's valuation of avoiding a risky situation. By
demonstrating that it is sensitive to the specification used to characterize indi-
vidual preferences arid the opportunities available for adjusting to risk (i.e.,
the nature of markets for contingent claims), this section provides the basis
for our conclusion, in Chapter 4, that the valuations of risk changes derived
from the planned expenditure function wilt themselves depend on what is
assumed about the individual's opportunities to adjust. The problems associ-
ated with defining an index of risk aversion with state-dependent preferences
are reduced in this case because the changes in state-dependent payments are
combined using prices of claims to form the change in planned expenditures.
Thus, these results are a tangible reflection of the point made by Cook and
Graham [1977]: the valuation of a change in p depends on the individual's
existing opportunities to adjust to risk. We have simply generalized this argu-
ment to acknowledge its importance to the characterization of an individual's
risk aversion.
3,5 THE PERFORMANCE OF THE EXPECTED UTILITY FRAMEWORK
For the most part, evaluations of the expected utility framework, based
on laboratory experiments, have questioned its relevance to real-world deci-
sions . Indeed, Schoemaker's [1382], recent review article concluded its apprai-
sal by noting that
As a descriptive model seeking insight into how decisions are made,
expected utility theory falls on at least three counts. First, people
do not structure problems as holistically and comprehensively as
expected utility theory suggests. Second, they do not process
information, especially probabilities, according to the expected utility
role. Finally, expected utility theory, as an "as if" model, poorly
predicts choice in laboratory situations. Hence, it is doubtful that
the expected utility theory should or could serve as a general
descriptive model. (Schoemaker [1982], p. 552)
A comparably pessimistic view of the prospects for the expected uti lity
framework can be found in Slovic and Lichtenstein's [1983) interpretative eval-
uation of the evidence on the extent of preference reversals in the literature.
They concluded by calling for a radical modification in the expected utility
framework. More specifically, they observed that
3-29
-------
This review has attempted to show how preference reversals fit into
a larger picture of information-processing effects, that as a whole,
pose a collective challenge to preference theories for exceeding that
from reversals alone. These effects seem unlikely to disappear, even
under rigorous scrutiny. Moreover, anything less than a radical
modification of traditional theories is unlikely to accommodate these
phenomena, (Slovic and Lichtenstein [1983], p. 603)
These are two of a number of examples of the criticisms of the expected
utility framework that could be cited based on the contradictions to it observed
in laboratory experiments involving individual decisionmaking under uncertain-
ty. While the typical advocate for the expected utility model can always argue
that the laboratory is not the real world and that a contradictory performance
pattern in the former does not necessarily imply the same for the latter, this
position has nonetheless become an increasingly difficult one to adopt. Indeed,
as Machina [1983] has observed in discussing a similar criticism of the results
from experiments that did not involve real money,
if the primary defense of the expected utility mode as a real world
descriptive model rests on the presumed "rationality" of the typica
economic agent, it seems odd to then assert that such agents are
not rational or competent enough to correctly state how they would
behave in some simple proposed choice situations, (Machina [1983],
p. 90)
There has been a growing tendency to argue that analyses involving the
expected utility model are purely theoretical and to "apologize" for its use in
empirical analysis. Since our theoretical analysis as well as our specification
of hypotheses and models for analysis with the survey results will begin from
conceptual analysis of individual behavior based on the expected utility frame-
work, it is important to review these experimental violations and to consrder
how they might influence our research efforts.
Machina [19831 has recently prepared a detailed state-of-the-art appraisal
of the theory of individual behavior involving risk including a careful appraisal
of this experimental work. Our review will be based in large part on his
work, supplemented by the earfier work of Schoemaker [1980, 1982], Slovic
and Lichtenstein [1983], and others who are identified as they become rele-
vant ,
Evaluations of the expected utility model have tended to focus on two
axioms — the independence axiom, the most specific assumption in terms of its
3-30
-------
effect on the model—and the transitivity axiom. The examinations of independ-
ence have found a wide variety of conditions that appear to lead to violations.
For example, a common outcome is that consistent ratios of probabilities across
pairs of prospects can lead to violations of choices that would have been ex-
pected based on the expected utility framework. Equally important, experimen-
tal behavior exhibits responses that are overly sensitive to changes in low
probabilities for extreme or outlying events, a result that is inconsistent with
the independence axiom. The violation of the independent condition is impor-
tant because it is crucial to our ability, using expected theory, to recover
the Von Neumann-Morgenstern utility function. Under this theory, all legiti-
mate recovery methods should yield the same utility function up to a positive
linear transformation. Yet, a number of studies have found contradictions to
this result. Equally important, violations have been found with the transitivity
assumption, with the most important of these involving preference reversals.
These reversals arise in the rankings individuals assign to risky prospects in
comparison to the certainty equivalents they describe to be relevant to those
prospects. Clearly, these findings are important for any attempt to measure
empirically the values of reducing hazardous waste risks.
The most relevant question for benefits measurement is: How does one
proceed in light of the empirical evidence for expected utility theory? Rather
than consider the specific details of each of the types of experiments that have
led to this questioning of the expected utility framework, we accept findings
at "face value." Yet, we do not conclude that the framework is irrelevant for
describing behavior under uncertainty. There are several reasons for this
conclusion. First, all of the models have been based on a state-independent
specification for the utility function. Once this assumption is changed, predic-
tions concerning real world responses become much more difficult. Arrow
[1974] anticipated this conclusion in his discussion of the role of state depend-
ency of the utility for the insurance decisions of households. Indeed, he ob-
served that the state-dependent specification for utility posed significant prob-
lems for the behavioral interpretation of probability. Specifically, he noted
that
The expected utility theorem or hypothesis, especially in conjunction
with the Bayesian concept of subjective probability, implies the
meaningful separation of tastes (as represented by the utility func-
3-31
-------
tion) from beliefs (as represented by probabilities). But in the form
(3) [the state-dependent specification], this separation is no longer
operational... No set of observations can distinguish the probabili-
ties from the utilities. (Arrow [1974], p. 6, n . 1, )*
Unfortunately, this potential explanation for observed violations has not
been considered in any of the literature that is critical of the expected utility
framework. ^ This is surprising since all of the proposed alternatives address
the violations by specifying decision frameworks in which the separation of
tastes and probabilities is also not really preserved.
Of course, we would not want to "over interpret" the importance of state
dependency for the simple laboratory setting. In most of these cases, poten-
tial sources of state dependency for individuals' utility functions cannot be
identified. Winning or losing is not a sufficient explanation. Thus, state de-
pendency offers, in our judgment, a more plausible explanation for violations
observed in real world settings.
There are alternative explanations short of the "radical modifications"
called for by Slovic and Lichtenstein [1983]. Indeed, they build upon much
of the research in psychology on risk-taking behavior. This work has tended
to call for models that replace either or both of the tastes/probability formation
dimensions of the expected utility framework. For example, the assunot'on of
cognitive limitations to decisionmaking has often led to the acceptance of a
model that assumes bounded rationality to describe individual behavior, Indi-
viduals are postulated as using simplified models of complex processes or deci-
sion circumstances and as acting according to those models. These frameworks
could be consistent with probability assessments as they have been used in
the expected utility framework. Clearly, this interpretation is consistent with
the recent results of Viscusi and O'Connor [1984] with respect to compensating
wage differentials and job risks.
*Arrow attributes this insight to an unpublished paper by Herman Rubin,
t
Machina's [1982] work comes closest to identifying this point. His paper
on the expected utility framework without the independence assumption analy-
ses behavior in response to distributions of probability mass defined over pay-
offs rather than states and does recognize that the assumption of the equiva-
lent of a state dependency would be one means of establishing consistency be-
tween the violations to the independence axiom and a reformulated expected
utility hypothesis.
3-32
-------
Alternatively, we might assume individuals utilize Judgmental heuristics
(T versky and Kahneman [1974]) to appraise probabilities. In this case,
assessments of probabilities for specific events or outcomes depends on how
representative the event is, the availability of information, and use of the most
familiar aspect of the event to form an initial judgment with subsequent adjust-
ment based on its other aspects.*
Both of these views of individual decisionmaking are part of a process of
trying to model how individuals process information. In our view, this proc-
essing need not be inconsistent with the expected utility framework. Analysts
may simply have done a poor job at communicating the experimental conditions
or the problems at risk with respondents. In the real world, some situations
involve repeated experience with uncertain phenomenon. With repeated trials,
individuals are likely to improve their ability to form assessments of probabili-
ties, and we would therefore expect this experience to influence the results
of evaluations of the expected utility model. By contrast, in the context of
laboratory experiments, the same opportunities for learning are usually riot
available. Consequently, the analyst must communicate the information to par-
ticipants to assure that this information can be acted upon in ways that are
comparable to decisions in the real world.
To deal with this requirement in our own analysis, we report in Chapter 8
the results of an extensive set of discussion, or focus group, sessions con-
ducted as a part of the process of questionnaire development. These activities
were used to determine the wording, the methods for explaining probabilities,
and the events at risk. They build on the experience of psychologists in their
attempts to model decisonmaking under uncertainty but do not dismiss the ex-
pected utility framework. They were an essential dimension of the research
design and were required to respond to these violations to the expected utility
framework.
3.5 SUMMARY
This chapter has provided a brief review of the state-preference approach
to modeling individual behavior under uncertainty. This framework offers the
*See Wallsten 11980] for a comparative analysis of the psychological ap-
proaches to decisionmaking under uncertainty.
3-33
-------
most flexible approach for evaluating the role of risk in decisionmaking. States
of nature can be defined to conform to the specific features of each problem.
The individual is viewed as planning consumption contingent upon which of
these states is realized. The constraints to these plans define the opportuni-
ties the individual has to adjust to the risk posed by uncertainty in the state
of nature.
Our review has largely focused on the expected utility model to describe
individual behavior. Within this framework we have considered the implications
of how individual preferences and constraints are defined. Since most of the
literature has adopted a state-independent specification of preferences and im-
plicitly adopted the case of actuarially fair markets as a reference point, the
analysis in this chapter has considered the effects of modifications in this
assumption for the expected utility description of behavior.
Two points are especially relevant to the empirical analysis reported later
in this volume. First, conventional measures of the extent of risk aversion
have been based on state-independent preferences and have implicitly main-
tained a specific institutional mechanism for adjustment to risk. When both
state-independent preferences and fair markets are maintained, the restrictive
nature assumptions used to define the Arrow-Pratt index of risk aversion is
not as easily identified. However, once each of these assumptions is relaxed,
the definition of risk premia associated with actuarially fair gambles will depend
on what is treated as the reference point and the opportunities the individual
is assumed to have available for adjusting to risk.
Second, the violations to the expected utility framework's assumptions
found in experimental studies do not necessarily imply it is an inappropriate
basis for organizing our survey results. We have argued that these findings
imply the approach used in our questionnaire and survey must reflect an
understanding of how to communicate risk and changes in risk to individuals.
Moreover, these findings provide support for the adoption of a state-dependent
specification of preferences. All of the tests of axioms of the expected utility
framework have maintained a state-independent specification for individual pref-
erences. As Arrow observed, state dependency, by eliminating the separation
of tastes and beliefs, provides a mechanism for accommodating most of the in-
formal and formal models of individual decisionmaking proposed as alternatives
3-34
-------
to the expected utility framework. Of course, it is simply a reflection of our
ignorance. Until we can specify the factors that lead to state-dependent pref-
erences and identify how they affect the marginal utility of income, we do riot
have a framework that offers sufficient understanding of individual decision-
making to permit predictive evaluation of individual responses to risk. Conse-
quently, one dimension of our empirical analysis will be to identify the attri-
butes fo risk that might affect how individual preferences vary with the states
of nature associated with our problem—the management of the disposal of haz-
ardous wastes.
3-35
-------
CHAPTER 4
THE ROLE OF THE EX ANTE AND EX POST PERSPECTIVES IN
MEASURING WELFARE CHANGES UNDER UNCERTAINTY
4.1 INTRODUCTION
The need to value reductions in risk, which result from regulators ac-
tions pertaining to hazardous wastes, has significant implications for trying to
use conventional benefits measurement practices. In an attempt to address
the most important of these implications, this chapter reconsiders the con-
ventional practices of benefits measurement in the absence of uncertainty
and then addresses the role of analytical perspective--!.e,, ex ante versus
ex post--for the process of measuring changes in welfare under uncertainty.
Section 4.2 outlines the ways in which the effects of policy actions have been
described in the past and how these effects would differ under conditions of
uncertainty. Section 4.3 explores the applicability of the ex ante and ex post
analytical perspectives for measuring the welfare changes from policy actions
governing hazardous waste management. Section 4.4 defines use and intrinsic
values within the ex ante framework, and Section 4.5 summarizes the implica-
tions of the ex ante framework for the definition of valuation concepts for risk
reductions and further research on welfare measurement in the presence of
uncertainty.
4.2 BACKGROUND
As it has been developed in applied welfare economics, the theory and
practice of benefit measurement has largely been concerned with valuing goods
or services under conditions of certainty. In evaluating the benefits associated
with a regulation or other policy action, the practice has been to relate the
action involved to some change in the prices facing individual households (and
firms) or to the quantities of goods or services they consume under defined
4-1
-------
conditions of access.* Once these changes are described, valuation measures
can be defined in a Marshallian (i.e., holding household income constant) or
Hicksian (i.e., holding household utility constant) framework. As a rule, the
household has been assumed to select its consumption choices in a world of
certainty, and the policy change itself is assumed to be certain. Of course,
most of the specific applications of benefits analysis have acknowledged the
difficulties associated with translating the specific policy decision into an im-
plied price or quantity change. In some cases, these difficulties have led to
efforts to define a range of scenarios in an effort to capture the uncertainties
inherent in describing the intervening mechanism that connects the policy to
the outcome. Indeed, the use of scenarios designed in this way has come to
be accepted as one way of reflecting the implications of this form of uncertainty
for the results of benefit-cost analyses.
This kind of uncertainty might be called planner's or policy uncertainty.
Analytical models and concepts such as statistical decision theory, the value of
information, and quasi-option value have been developed to help policy analysts
deal with this type of uncertainty. In particular, these methods attempt to de-
velop approaches to organize information and decision rules that explicitly take
account of the uncertainty concerning the magnitudes of variables relevant to
their decisions (e.g., benefits and costs). However, none of these analytical
tools recognizes that individuals will modify their decisions in the presence of
uncertainty. Consequently, uncertainty facing economic agents can affect how
they will value the services (or price change) delivered by a policy.
In this chapter, we develop a working description of individual uncer-
tainty—that form of uncertainty faced by individuals who are users or poten-
tial users of an environmental resource. For example, individual users of a
*The term conditions of access in this context refers to how the individual
is allowed to use a resource. Where private firms do not decide the amounts
to be available (e.g., a government agency providing "protection" from risks
of exposure to hazardous wastes through regulations), conditions of access
can involve nonprice rationing conditions and/or uncertainty over the levels
available for any specified set of terms. This means that an individual might
be viewed as bidding for an improved likelihood of realizing some desirable
slate. Maier [1934] has recently demonstrated that a change in the probability
of some desired state characterized in terms of having more of some environ-
mental amenity (or other commodity) can be treated as equivalent to a change
in the quantity of the amenity with unchanged odds.
4-2
-------
contaminated groundwater aquifer may face a higher probability of developing
cancer. Similarly, individuals may also be uncertain whether a particular
unique and irreplaceable environmental resource will be available for their use
at some future date, or they may be uncertain whether they will actually want
to use that resource in the future.
To consider the implications of this type of uncertainty, we need to spe-
cify the model describing how household choices are made in the presence of
the uncertainty of interest and then evaluate the welfare -implications of the
proposed changes within it. That is, the analytical problem is to explicitly
incorporate uncertainty in the models of individual choice and to deduce the
implications of that uncertainty for the measurement of welfare changes associ-
ated with policies dealing with environmental resources. The resulting benefit
measures wili differ from those derived under the assumption of certainty.
As we noted in the preface to Part I, we have adopted the expected util-
ity framework as the basis for describing individual choice under uncertainty.
Based on this framework, Chapter 5 develops the specific benefit measures
proposed for valuing reductions in the risks of exposure to hazardous wastes.
However, before proceeding to a discussion of these proposed benefit measures,
it is important first to consider the implications of individual uncertainty for
existing benefit analysis and the analytical perspective used in benefit analysis.
Most conventional analyses have described policies in terms of changes in
either prices or quantities (because these have been the basis used to define
the available welfare measures). Using this type of approach to incorporate
individual uncertainty would require that we specify a model of individual deci-
sionmaking under uncertainty to describe how individuals would value some
changes in the price or quantity of an environmental resource under these
circumstances and then use it to evaluate the policy-induced change. Alterna-
tively, we might describe the policy as changing the nature of the uncertainty
itself, , In this case, we would be focusing on a set of new parameters that
are added to the exogenous factors that affect household behavior with the
introduction of uncertainty —namely, the probabilities of the states of nature.
These probabilities could reflect the individual's uncertainty as to the vector
of prices in alternative states of nature, the incomes to be received in alterna-
tive states, or the magnitude of some other state variable describing conditions
4-3
-------
that are important to an individual's utility, including health status, availability
of an environmental resource, and so forth.*
The second aspect of this reconsideration concerns the measure of indi-
vidual welfare—what we have designated the "perspective for welfare analysis/1
t
ex ante or ex post? An ex ante perspective is one in which we view the in-
dividuals as planning actions that he would take contingent upon the state of
the world. It may be tempting to suggest that, in an ex ante framework, a
measure of the change in an individual's welfare as a result of a policy increas-
ing the likelihood of some desirable outcome is the individuai's willingness to
pay for the change before the uncertainty over states of nature is resolved.
Indeed, several studies have used this convention (e.g., see Jones-Lee
[1974]), However, this definition makes a subtle assumption. When we con-
sider an individual's planned consumption, we define those plans for each state
of nature. That is, consumption choices are described as contingent in that
they suggest what the individual would plan to do as if the state of nature
were realized. By specifying a constant willingness to pay, we are implicitly
assuming it will be made irrespective of the state of nature. Therefore, this
very definition includes an assumption about the mechanisms constraining how
an individual can plan. If plans involve contingent consumption choices, there
is no reason that we cannot define contingent payments--the payments an indi-
vidual would be willing to make in each state of nature if the policy was imple-
mented. Indeed, the appropriate welfare change measure would be the set of
payments with the policy that yielded the same expected utility available with-
out the policy and without the payments. Of course, there is not one such
set of payments but an infinite set of payments. Indeed, this is simply one
description of the Graham [1981] willingness-to-pay locus.
How, then, does one define a welfare change In an ex ante framework?
It would seem that the definition itself requires an assumption with respect to
what characterizes the institutions available for individual adjustments. In
short, what are the constraints to how these payment vectors might be defined?
*This is consistent with the approach adopted by Cook and Graham [1977]
arid more recently proposed by Simmons [1983],
t
This question was first raised in the context of environmental regulation
by Smith and Desvousges [1983],
4-4
-------
Once this is specified, the appropriate welfare measure is the change in the
planned expenditures required to realize a constant expected utility under the
policy compared to the planned expenditures required for that level of utility
without the policy. Clearly, that expenditure function is defined conditional
upon the specification of mechanisms constraining the state-dependent pay-
ments.
By contrast, an ex post definition requires two analyses to define the
willingness to pay. The first considers an individual's willingness to pay for
the outcomes implied by the policy action under each of the possible states of
nature. These evaluations must be conducted with each outcome considered
as a choice in_ the absence of uncertainty. The second step involves the calcu-
lation of the expected value of these individual state-specific welfare measures--
i.e., generally, the expected value of the compensating surplus values associ-
ated with each state.
It should be clear that these two perspectives on the treatment of uncer-
tainty (i.e., ex ante and ex post) need not yield the same valuation estimates,*
One important element affecting differences in these valuation estimates is the
set of opportunities available for diversifying risk. An individual's valuation
of a change in the probability of some adverse event will depend upon the
extent to which the event can be insured against.^ For example, if the events
can be expressed exclusively in financial terms, a risk-adverse individual,
with access to actuarially fair insurance, will Insure until he is indifferent to
the outcome (in ex ante terms). + These two features are further discussed
in Section 4.3 below.
Use values have generally been defined as some form of consumer surplus
(either Marshall's consumer surplus or the Hicksian compensating or equivalent
measures). These are ex post measures of benefits. By contrast, the timeless
*As we show in Section 4.3, it turns out that much of the controversy
over option value can be interpreted as a question of perspective. See Bishop
[1982] and Smith [1983] for recent reviews of the option value problem. The
question of perspective has considerable relevance to the controversy over
estimating the value of "statistical lives." See Ulph [1982] for further dis-
cussion ,
tThis point was clearly demonstrated by Cook and Graham [1977].
+This conclusion assumes state-independent utility functions.
4-5
-------
definition of option price is based on an ex ante perspective. Therefore, one
component of the nonuser benefits, the option value, actually mixes these two
perspectives by decomposing the option price into the expected consumer sur-
plus and the option value. This mixing of perspectives arises because meas-
ures for distinct components of the total benefits derived from environmental
resources have been defined from what are fundamentally different models of
the individual's decision process. Yet the results are treated as if they were
fully compatible. In the next section we present an introduction to the issues
associated with classifying the types of benefits within an ex ante framework.
4.3 EX ANTE VS. EX POST PERSPECTIVES
An ex ante social welfare function makes social welfare a function of the
expected utilities of the individuals in the society, while an ex post social wel-
fare function makes social welfare equal to the expected value of the social
welfares realized in alternative states of nature. The choice of an ex ante
versus an ex post welfare measure involves fundamental questions of welfare
theory—in particular, the role of equity in societal welfare and the way equity
is defined. Broadly speaking, ex ante social welfare functions reflect a social
concern with the equity of opportunity in the expected value sense, while
ex post social welfare functions reflect a concern with the equity of outcomes.
Consider a society that has adopted a social welfare function reflecting
its ethical judgments concerning equity and has undertaken the redistributions
of wealth and/or taxes and transfer payments necessary to achieve a social
welfare maximum at some given point in time. Suppose also that new invest-
ment opportunity is being considered that would alter the distribution of in-
comes and utilities in different ways in various states of nature. If the project
is undertaken, then society will wish to levy taxes and make compensating pay-
ments to restore the optimum distribution of outcomes after the state of nature
has been revealed. The consumer surplus changes provide a basis for deter-
mining the required taxes and compensation, and the expected value of aggre-
gate consumer surplus is an indicator of whether the payments can be made
without making anyone worse off.
Now let us assume that the society has chosen an ex ante social welfare
function. Thus, the focus of attention for benefit-cost analysis is changed to
expected utilities and their monetary equivalents. How are these monetary
4-6
-------
equivalents to be measured? Option price is only one of many possible ways
of defining a monetary equivalent for a change in expected utility. We will
show that the appropriate way of defining the monetary equivalent depends
upon particular circumstances, including the opportunities for diversifying
risks through contingent claims markets and the institutional feasibility of en-
forcing alternative contingent payment schemes.
The expected value of consumer surplus is an ex post measure in that it
focuses on the realized outcomes of policy choices. Evaluating policies in terms
of the planned expenditures required to maintain a constant expected utility,
which may reflect risk aversion, is the basis for ex ante welfare measurement.
Option price is an ex ante measure of the increment associated with planned
expenditures under one institutional framework for making state-specific pay-
ments. In particular, it is that state-independent payment that makes the
expected utility with the policy exactly equal to the expected utility w thout
the policy.
The presence of uncertainty for measuring individual welfare creates dis-
tinctions between ex ante and ex post perspectives analogous to those dis-
cussed by Ulph [1982] and discussed above for the specification of society's
welfare function. To illustrate these differences in specific terms,, consider
the ex post case and the conventional description of consumer choice, where
individual decisions are assumed to be made under conditions of certainty. In
this setting we can describe the individual as minimizing the expenditures made
on all goods and services to realize a given utility level. If X. describes the
consumption of the ith commodity, Pj its price, and U(Xt, X2, .. ¦, Xn) the
individual's utility, then Equation (4.1) defines the expenditure (or cost) func-
tion for the individual:
n , .
E(Pi, P2, Pn, U) - Min [ I P.X. J U = U(Xl# X2 ..., Xp)] . (4.1)
i=1
Two further assumptions must be made. First, we wiil assume that the re-
source is a nonmarketed good, some of whose services are available without
any need to travel or otherwise gain access to them. Second, we will assume
there is at least one observable (and implicitly priced) measure of the use.
4-7
-------
Adding an argument (Q) to U(.) for the resource's contribution to satisfaction
that is disassociated with the individual's use of it satisfies the first require-
ment, and selecting one or more X.'s as measures of use is consistent with
the second.
If there is only one measure of "priced" use, say , then the compen-
sating variation (CV) measure of the value of the site when the level available
of the resource is Q is given in Equation (4.2):
CV = E(P i(Q), P2, • ••, Pn, Q, U) - E (P,, P2, PR, Q, U) . (4.2)
Introducing Q into the utility function together with the specification of a fixed
level of the resource that is available for uses not necessarily reflected in Xt
(namely Q) leads to the expenditure functions in Equation (4.2). Compensat-
ing variation is the difference in the expenditures that would be made at the
"choke," PiCQ) (i.e., the price at which Xr = 0), and those at the actual
price, P^ for a given level of utility, given values for all other prices and
Q. It is the maximum amount an individual would pay for the lower price of
(i.e., from Pj (Q) to Pt), It is important to note that the choke price
for Xt is assumed to also be related to the level of Q.
It is also possible to use the framework to describe other motivations for
valuing the resource, Q--the nonuser or intrinsic values that individuals might
realize as a result of the existence of the resource amenities at particular lev-
els or the increments to these values because of increments to the resources.
This distinction has played an important role in the classification of the bene-
fits associated with changes in environmental resources, and we will return to
it in the next section and in Chapter 6. Our objective here is to compare the
ex ante and ex post perspectives for welfare measurement. Thus, consider
the description of a measure of change in ex ante well-being. Here we wili
also use a different type of expenditure function. In this case, the individual
is viewed as planning consumption so that each commodity is defined as a claim
for consumption of that commodity contingent upon realization of a state of na-
ture. Following the same format developed in Chapter 3, we have the planned
expenditure function (given state-dependent preferences) defined as follows:
4-8
-------
EfPP PP P P p
11' 21' K1' 12' '' " K2' *" •' 1N ' KN'
*
K-1 _
*2, - • -, ?tK.1, 1 - I n.; EU) = (4.3)
K N _ K
Mil [ 2 ^ Pij X{j I EU = In. Uj (Xjr X,2, ..., XiN)] ,
where
X.j = contingent claim to commodity j in state i
P.. = the prices of contingent claims (the s.. in the notation of
' Chapter 3). IJ
Equation (4.3) is simply a generalization of the two-commodity, two-state case
developed in Chapter 3. In this case, there are N commodities and K states.
If we assume a complete "set of markets for contingent claims, there are N• K
such prices.
To define option price in this framework, however, we must make some
additional assumptions. In particular, if we assume that Q enters at least one
of the state-dependent utility functions and that designates the state-
dependent, planned consumption of use of Q, then option price can be defined
in terms of the expenditure function (with Q as an argument) given in Equa-
tion (4.4):
OP = ECO, P12, P22 PK2 '¦¦¦' P1N' P2N""' PKN'
K-1 _
™ 2 7t-, O, EU)
i=1
(4.4) •
E(0, P12, ^22' ' " " f PK2' • * " P1N ' P2N ' PKN '
K-1 _ _
Tt , n , 1 - I n; Q} EU)
1 £ i=1
4-9
-------
Two points should be noted in this definition. A zero value for Q (as in the
first term on the right side of Equation (4.4)) is assumed to imply that no
use of the resource can take place when Q = 0. Q must be positive for use
to take place. In this case, the option price is for a level of Q of the re-
source. Also the price of use in any state, , has been assumed to be zero
in all states. We will argue that this is consistent with the original definition
of the option price. It precludes state-dependent payments for use of the ser-
vices of the resource- However, it should be acknowledged that our original
definition of option price was unclear about the per-unit charges for the use
of the resource. Under one interpretation, only a constant price for X.^ over
all states of nature is required. While this would hold constant the per-unit
charge for across states, planned total expenditures for use would be state
dependent because the planned state-dependent consumption levels could vary.
This would seem to violate the intentions of the original definition of the option
price. Alternatively, it could be suggested that the payment of an option
price was only to ensure access. Therefore, under this view, one would define
the option price as the payment for access. The fact that payments for con-
sumption levels would be state dependent does not in this case affect the con-
stancy of the payment for access and the definition of option price as a state-
ment independent payment for access. Either assumption can be accommodated
in our analysis. Regardless of the view of the process that describes how
the resource is allocated (i.e., one payment for guaranteed access and then
payments for use or simply a payment for use), the basic point of the analysis
remains unchanged-
With this background, it is now possible to use the two types of expendi-
ture functions in the definition of the option value (GV), The conventional
definition for option value is given in Equation (4.5):
K
OV = OP - 1 7t. CV. . (4.5)
i=1 1 '
Substituting from Equation (4.4) for the option price (assuming the price for
planned use is constant at zero), we have
4-10
-------
_ _ K
OV = i (O; P, k; 0; EU) - E (O; P; n; Q, EU) - I n. CV. . (4.6)
i=1 1 1
To reduce the complexity of the notation, we have represented the set of zeros
for the prices of contingent claims to with a single zero, the prices of other
contingent claims with P, and the set of probabilities with h.
We can also replace each of the CV.'s in Equation (4.8) using their defini-
tion in terms of the ex post expenditure function (e.g., Equation (4.2)),
That is, we repeat the process defining the ex post expenditure function with
each state-dependent utility function, derive the corresponding expression for
CV., and substitute each expression in Equation (4.6), Option value, OV, is
now given as Equation (4. 7):
OV = E (O; P, n, O, EU) - E (O; P; n; Q, EU)
(4.7)
K
- 1 71 [E.(PX(Q), P2# P3, Q, 0) - E; (Pl( P2, Q U)j .
i = l ' 1 1
Equation (4.7) illustrates how option value mixes two perspectives for individ-
ual decisionmaking. The first is the planned or ex ante view of consumption,
while the second utilizes the ex post orientation in defining benefits from use
of the resource. This conclusion that option value-mixes perspectives is un-
changed if we assume that consistency requires we assume that the actual price
for use of the services of the resource (i.e., Pi) is nonzero.
To illustrate the importance of the difference, consider an alternative
definition for user values based on the expenditure function associated with
planned consumption. Letting X.^ designate the contingent claim associated
with use in the ith state, the value of planned use would be the difference in
expenditures when the prices of the contingent claims for Xx are at the choke
levels for planned consumption and the expenditures are at existing prices,
as in Equation (4.8):
4-11
-------
CVp ~ ^ (P11 ' P21 ' '"' PK1 ' P12' PKN' n1' rt2' *""'
K_1 _
1 - 1 7i ; Q; EU)
i=1 ' (4,8)
~ E ( ^ , ^21' ¦ ¦ ¦ ' PK1' P12' ' ' ' ' PKN ' ^1' n2' " ' * '
K_1 _
1 - I 7i - Q; EU) .
i=1
With state-dependent preferences we do not, in general, expect that CVp will
equal the expected compensating variation calculated ex post from each utility
K
function (i.e., CV f I n.CV.). The difference between option price and the
P j=y 1
planned consumption value appears to offer yet another potential definition for
option value. However, this interpretation is misleading. CVp describes
planned user values under one set of institutional arrangements for the con-
tingent claims, including those that have been identified to be associated with
use of the resource (i.e., the X.^'s). These institutions are not compatible
with either definition of the option price.
We could, of course, modify our definition of the option price and assume
it was defined with constant, but nonzero, prices per unit of use (i.e., =
constant for alii). Of course, without payment of the option price, no con-
sumption would be possible. In this case, we have a different value for the
option value. It is also important to acknowledge in evaluating this definition
that it appears to be similar to the McConnell [ 1983J definition of existence
value. In comparing these two definitions, option price plays a rote analogous
to the total value of the resource. However, there is a difference that again
reflects the importance of the opportunities for adjustment. Our definition of
CVp allows the prices of claims to to vary with the state of nature. The
definition of option price precludes this possibility since it assumes they are
either all zero or alt constant.
The difference between option price and use value based on the expendi-
ture function derived from planned consumption choices is not an option value.
It is a reflection of both nonuse values and the institutions we assume are
4-12
-------
available to individuals. Equally important, it highlights an additional dimen-
sion of modeling individuals' behavior under uncertainty and the implications
of these models for benefits analysis. Basically, this issue relates to the ref-
erence point at which we initiate the analysis.
Option value, as conventionally defined, compares the option price with a
point on the income certainty iocus that is defined by the expected consumer
surplus. It compares two institutional regimes—one with uncertainty where
the option price is paid and another where there is no uncertainty in the deci-
sion process. Benefits are constant at the expected consumer surplus. This
point is easily seen using Graham's [1981] willingness-to-pay locus. This
function offers an alternative means (to the expenditure functions defined
earlier) for illustrating the implications of institutions for adjustment. Figure
4-1 reproduces Graham's discussion of option price and option value. Option
price is compared with the expected surplus as a certainty concept. The ref-
erence point is one of a certain income given by E(S), or 1 nXVj in terms of
our notation. This may seem to be a natural reference point because it was
the one used in nearly all work following the Arrow-Pratt analysis of risk
aversion. While it may be natural from an analytical perspective, it is not a
natural reflection of the world in which these choices must be made. Graham
[1981] provided a similar argument in his critique of the attention given to
the sign and magnitude of option value. He observed that
Option price ts the appropriate measure of benefits in situations
involving similar individuals and collective risk.
Expected value calculations are appropriate to situations involv-
ing similar individuals and individual risk.
Whether or not option price exceeds the expected value of sur-
plus is largely irrelevant to the evaluation of risky project.
[Graham, 1981, p. 716]
Graham's analysis adopted an ex ante perspective and focused on the types of
institutions available for adjustment to risk. Expected value measures of bene-
fits were specified as appropriate for individual risks because his analysis also
assumed that in these cases there existed actuarially fair insurance. It is im-
portant to recognize that his argument was not advocating the use of the ex-
pected consumer surplus as the benefit measure in this case but, rather, the
4-13
-------
Figure 4-1. Graham willingness-to-pay locus and option value.
-------
expenditures associated with the fair bet point. In the presence of state-
dependent preferences, this point would not fall on either the income certainty
or the utility certainty loci. It would correspond to the expected expenditure
on the contingent claims associated with each commodity and state of nature.
Thus, in developing a benefits taxonomy, it is important to clearly identify
the perspective used in describing individual behavior because it is directly
relevant to any evaluation of actions designed to affect that behavior. This
discussion has suggested that the choice depends on the reference point from
which we start the analysis and the mechanisms we assume are available to
the individual to adjust in response to risk. Once we propose to evaluate a
policy that changes the risk an individual faces, then the ex ante framework
is the appropriate basis for defining the benefits associated with that policy,
since it corresponds to how the individual would have to make the valuation
decision in judging the action in advance. Of course, this Judgment in itself
does not imply the option price is the relevant concept.
Option value is a valuation measure that compares an uncertain situation
with a certain reference point. The reference point is defined in terms of
income certainty and presumably is of interest because of the history of the
development of measures of risk aversion.
Once the ex ante perspective is accepted and it is acknowledged that
individuals' decisions do not take place in circumstances that begin with cer-
tainty, then benefit measures must be defined in terms of our planned expendi-
ture functions. These functions can be defined to reflect all the risks faced
by the individual and the mechanisms available for adjustment. Option price
is seen (as in Equation (4,4)) as one valuation concept based on specific insti-
tutions and risks. Moreover, policies can be considered to change either the
level of availability of the resource (as in our definitions thus far) or the
probabilities of specific states of nature. Moreover, this analysis need not
assume that the risks are limited to those specifically associated with the poli-
cies under study. Risks can be added to existing uncertain income streams.
Consequently, an income certainty reference point may not be of practical rel-
evance, Even for analytical purposes — to measure the extent of risk aversion--
comparison of the risk premia (i.e., the payment over the expected value to
avoid risk) of different individuals need not be equivalent to the ordering im-
4-15
-------
plied by the Arrow-Pratt index of risk aversion when background risk is
present.
4,4 USER AND INTRINSIC VALUES IN AN EX ANTE FRAMEWORK:
AN INTRODUCTION
The analysis of the previous section argues that the relevant basis for
valuation is the planned expenditure function (i.e., Equation (4.3)), In the
process of defining the option price we have introduced two ways in which an
environmental resource can affect planned expenditures—through planned uses
(i.e., the contingent claims for Xt in all states of the world) and through the
availability of the resource itself (i.e., the presence of Q in the expenditure
function). This specification opens the prospect for distinguishing benefits
into categories according to whether they are associated with planned use or
independent of those plans.
To define these components in an ex ante framework with uncertainty is
somewhat more complex than in the certainty case, First, we must identify
the nature of the conditions of access to the resource and the institutions
available for adjustment. Second, we must specify the nature of the change
to be evaluated. Given this information, it is possible to specify the user and
existence or intrinsic components of the value of the policy. Before proceeding
to develop these in specific terms, it is important to highlight a key difference
between this case and classification of use and existence benefits under cer-
tainty. In the certainty case, the use benefits are often capable of being
measured from the actions of individuals. That is, they can be indirectly in-
ferred as a result of the actions of individual economic agents through their
use. By contrast, the existence values are usually not observable, since they
do not involve tangible (or at least observable) actions by these agents. Thus,
we would want to identify the distinction to recognize that benefit estimates
based on use may well understate the full benefits provided by the resource.
The same arguments are more difficult to apply in the ex ante framework,
where all actions are planned. We do not observe the plans. While we shall
argue that these plans may be associated with observable actions in our dis-
cussion of the relationship between estimates of the value of risk based on
hedonic property value studies and the survey-based estimates, the full details
of these plans will not be known. Assumptions must be used to substitute
4-16
-------
for observed behavior in attempting to understand the motivations for behavior.
Consequently, it is not clear that the distinction is as meaningful or desirable.
In effect, if we are to estimate individuals' valuations in an ex ante framework,
we may have to rely on direct survey methods, ft is not clear that we can
successfully elicit individuals' values and request that these values be assigned
to specific motivations. While one can argue that this is an empirical question,
this argument in itself may be misleading. We will not know the true values
for benefits in this ex ante framework; thus, it is not clear that we can judge
whether an analysis leads to a meaningful separation of the two types of bene-
fits.
With this background it is now possible to use the ex ante framework to
propose a general approach for classifying benefits. In Chapter 6 we return
to this classification in relation to the classification of user and existence values
under conditions of certainty. Planned user benefits, PUB, can be defined in
general terms using Equation (4.9):
PUB = E (Pi <0>; P; n~, Q, E™U) - E (Px; P; h; Q, EU) . (4.9)
We have used the same notation for summarizing the prices of contingent claims
(separating the prices of the claims associated with use) from those of other
commodities, and Pj(Q) represents the vector of choke prices--where planned
consumption would be zero in all states. The existence value (planned exist-
ence benefits, PEB) would compare expenditures with no planned use with
those when there was none of the resource available, as in Equation (4.10):
PEB = E (Pi; P, n; O; EU) - E (Px (Q); P, n; Q, EU) . (4.10)
It is important to point out that the assumed relationship between the
measure of planned use and the level of the resource available is crucial
to the interpretation of Equation (4.10). We have assumed that without Q > 0
there can be no consumption of X.^ irrespective of the price. This implies
that
E(P1; P; n; O, EU) = EfP^Q); P; tt; O; EU) . (4.11)
4-17
-------
Use is effectively precluded as it would be if the price were at the level of the
choke price. Thus, FEB could also be written as follows:
PEB = E(P1 (0); P; h; O; EU) - ECP^Q); P; n; Q; EU ) . (4.12)
The sum of PUB and PEB corresponds to a planned counterpart to McConnell's
total resource value- In this framework, however, we can consider a variety
of amendments:
Changing the terms of access to the resource. For example,
these definitions might be recast with an option price that would
hold all P..'s constant (either at a specified value or zero).
This woulc^ imply the option price included both planned user
and existence values.
Changing the character of the description of the way in which
policies affect how individuals gain access to the resource.
This modification might imply a fixed Q but that policies change
the probabilities of access (i.e., the rt.'s). For these cases,
it would also be possible to define use and existence values as
welt as to specify an option price.*
Finally, we can expand the detail in the model by describing
the source of uncertainty (i.e., identifying components to the
rt.'s). Within such a framework it is also possible to consider
additions to risk as a result of changes in one of these compo-
nents and valuation concepts for each type of change.
We return to consider in more detail the measurement of nonuse values
associated with the reduction of risks to ecological systems in Chapter 6. In
that chapter we develop in formal terms the benefit taxonomy under certainty
and use a single institutional framework for adjustment, the option price, to
discuss the measurement of these nonuser benefits.
4.5 SUMMARY
This chapter has discussed the implications of how we define valuation
concepts. In the past, valuation or benefit concepts have mixed benefits de-
fined under conditions of certainty with those defined to arise because of the
*We return to this case in Chapter 6 by considering use and existence
values in a framework where it is only possible to make constant state-inde-
pendent payments for an improvement in the probability of a desirable state.
4-18
-------
existence of uncertainty. Each of the two types of benefit, then, is defined
under a different perspective on valuation. Should we consider valuation in
terms of planned actions or only when these actions are undertaken?
We have argued that when the policies to be evaluated change one or more
aspects of the risks facing an individual, then an ex ante perspective is war-
ranted for welfare analysis. Within an ex ante framework two features are-
especially important to valuation concepts. The first is the reference point.
Does the individual whose valuation is to be defined start from a position that
has no other sources of uncertainty but the risk to be evaluated, or is the
policy induced change simply an effect on one component of a number of risks
faced? The second concerns the ability to diversify the risk. That is, what
opportunities does the individual have to adjust to risk and ameliorate its
effects?
Of course, these features are not independent. Moreover, the resolution
of how one aspect is treated affects the others. For example, the definition
of option value is based on selecting a certainty point for comparsion and on
specifying a particular institutional system on how payments for claims can be
made. Payments must be constant across states and the valuation concepts of
interest, and in the size of the payment in relationship to the expected con-
sumer surplus.
r-
These features can be reflected in a planned expenditure function. Con-
sequently, it is possible to define use and existence values (not option values)
based on how a policy changes parameters important to these planned expendi-
tures. This planned expenditure function allows one to evaluate the effects
of institutions for adjustment as well as the nature of the change—prices, re-
source quality, or likelihood of access. Within each, one can define use and
existence value concepts provided the resource is hypothesized to have two
distinct effects on individual utility through use (that requires existence of 3
positive amount of the resource) and the level (or quality) of the resource
itself.
4-19
-------
CHAPTER 5
A CONCEPTUAL FRAMEWORK FOR VALUING RISK REDUCTIONS
5.1 INTRODUCTION
Chapters 2, 3, and 4 have set the stage for developing a conceptual basis
for valuing the risk reductions that are assumed to accompany increased regu-
lation of hazardous waste disposal. Chapter 2 introduced the overall problem
by describing why the valuation of these regulations must be treated different-
ly from the valuation of many other environmental policies. In the latter cases,
it is often reasonable to maintain that a policy leads to a certain increment in
some desired output--e.g. , cleaner air in a specific region or improved water
quality in a given river or take. In contrast, given the uncertainty that sur-
rounds both the disposal of hazardous substances as well as the ultimate
effects of exposure to them, we cannot assume any policy provides a certainty
of protection. We have argued that at best we can assume policies reduce the
risk (i.e., the probability in our context) of some adverse outcome. Conse-
quently, the development of a set of procedures for valuing policy outcomes
requires specific consideration of how to model individual behavior under un-
certainty. Chapter 3 provides a heuristic review of this literature.
With this description of the problems posed by any attempt to value poli-
cies associated with the disposal of hazardous wastes, and with our acceptance
of the expected utility framework for modeling individual behavior under uncer-
tainty,* one remaining question must be considered before defining the specific
'This approach contrasts with one recently suggested by Weinstein and
Quinn [1983b]. They observe that in light of the contradictions to the expect-
ed utility framework observed in individuals' decisions under uncertainty, it
may wet I be reasonable to inquire as to whether they should be reflected in
normative decision rules. More specifically, they describe this issue as a fun-
damental motivation for their evaluation of the models used to value changes
in the risks to life, noting that
The fundamental question raised in this paper is to what extent the
contextual and psychological attributes of a risky decision have suf-
5-1
-------
valuation concepts. This point has often been overlooked or confused. It
concerns what we describe in Chapter 4 as the perspective for decisionmak-
ing. That is, do we evaluate actions from an ex ante or planning perspective,
or do we use individuals1 values of the ex post outcomes? We have explicitly
argued for an ex ante approach. With this perspective, changes in the proba-
bility of a detrimental event are valued based on the changes in planned
expenditures an individual would undertake to maintain a constant level of ex-
pected utility.
Given this background, it is now possible to proceed to a description of
how this chapter completes the conceptual analysis of one component of an indi-
vidual's valuation of risk changes--the "use" component of these values. This
chapter uses a simple two-state model to describe the specific features of the
planned expenditure function described in Chapter 4 and discusses the impor-
tance of these features to the valuation of risk changes. Our example is now
explicitly tied to the framework we have used to present the risks posed by-
hazardous wastes to individuals in our contingent valuation survey. The chap-
ter also identifies the relationship between the model and what can be expected
in an empirical analysis of individuals' valuations of risk changes.
Section 5.2 describes a simple two-state modal to illustrate the valuation
concepts and the role of the opportunities for adjustment that are available to
the individual in influencing these values. Section 5.3 relates the model's im-
plications to the psychological literature describing how individuals make deci-
sions under uncertainty. This section also reconsiders the review of past
results discussed in Chapter 3 as tests of the expected utility framework and
to determine whether aspects of these findings would help in identifying factors
that have been found (or are thought) to influence individual choice under
uncertainty and should therefore be included in the empirical analysis. Sec-
tion 5,4 provides a brief summary of the chapter.
ficient normative status to justify their formal inclusion in methods
for valuing risk. Stated in terms of environmental decisionmaking,
the question becomes the following: which psychological and contex-
tual concerns do citizens want their decisionmaking agents to reflect
as normative principles in environmental decisionmaking, and which
would they want them to treat as irrationalities,'psychological weak-
nesses, or otherwise unjustifiable perturbations of rational decision-
making. (pp. 2-3)
5-2
-------
5.2 VALUING RISK CHANGES
Following the analysis discussed in Chapters 3 and 4, we maintain that
an individual seeks to maximize expected utility subject to a budget constraint
that describes his opportunities for adjustment. Our descriptions of these
adjustment possibilities will enter the decision problem through the specification
of different definitions of the markets for contingent claims facing the individ-
ual. Given an ex ante perspective for valuation, the relevant conceptual basis
for valuing a risk change is in terms of what we defined in Chapter 4 as the
planned expenditure function. This function defines minimum planned expendi-
tures on contingent claims that would be required to meet a given level of ex-
pected utility. It is a function of the prices for contingent claims, the proba-
bilities of the states of nature that are assumed to be uncertain, and the level
of expected utility that is to be realized. Thus, an individual's valuation of
a risk change, defined using this function, will depend on the nature of the
markets for contingent claims. We noted this point in Chapter 4 and now pro-
pose to use a simple two-state model to illustrate both how these values are
affected by the assumed nature of the markets for contingent claims and, in
turn, what these results imply about the testable hypotheses derived from the
model.
Consider the following planning problem for the representative individual.
There are two possible states of the world. In the first, an individual will
experience a detrimental health effect that could (but need riot) lead to death
for the purpose of our analysis.* The effect is assumed to be associated with
exposure to hazardous wastes; however, exposure does not ensure that the
health effect will be incurred. It introduces the individual to a second stage
lottery, which can be avoided if exposure is avoided. Thus, our analysis
emphasizes the distinction in outcomes by assuming that the probability of the
health effect is zero when the individual is not exposed to the substance.
The health effect leads to preferences that differ depending on whether
or not it is incurred. This follows the state-dependent preference arguments
*1 n the empirical analysts associated with evaluating individuals' valuations
of risk reductions, we consider the effects of a selected set of variations in
this end state- However, the basic scenario used to describe what is at risk
describes the outcome as death after 30 years from the time of the exposure
to the hazardous substances.
5-3
-------
proposed for the case of life-death decisions by Jones-Lee [1974] and Wein-
stein, Shepard, and Pliskin (1980] and described in more general terms by
Cook and Graham [1977]. As we noted in Chapter 3, a state-dependent spec-
ification for utility is simply a means of acknowledging that the marginal utility
of income may be different across the states of nature described in any partic-
ular problem,* To keep notation relatively simple, each state-dependent utility
function is specified to be a function of claims to income in that state. As we
observed in Chapter 3, it is possible to generalize this formulation to identify
claims for individual commodities. This generalization will be important if the
relative prices for these claims across states of nature for a given commodity
bear a different relationship to the relative odds of those states as the com-
modity in question changes. While this can be an important dimension of the
problem in some applications, it was not judged to be important for our dis-
cussion here.
To highlight the two-stage nature of the lottery, we have identified two
probabilities—the likelihood of being exposed to a hazardous waste, defined as
R, and the probability of incurring the detrimental effect once exposed, de-
fined as q. Equation (5.1) defines the expected utility realized from allocating
claims to income, the W.'s between the two states, with state one representing
the case of experiencing a detrimental health effect and state two representing
the case of remaining unaffected:
EU = R[qV1 (wp + (1-q)V2 (Wg)) + (1-R) V2 (Wg) . (5.1)
In this case the health effect can be incurred only through exposure to the
hazardous wastes. Therefore, the specification in Equation (5.1) can be re-
duced to Equation (5.2):
EU = RqV1 (W^ ) + (1-Rq)V2 (Wg) , (5.2)
where
EU = expected utility
V.(. ) = utility realized in state i
W. = contingent claim to income in state i .
*This is also the point of Marshall's [1984] recent discussion of the role
o1 indivisibilities in modeling decisionmaking under uncertainty.
5-4
-------
The individual seeks to allocate total income for planning purposes, E, among
these claims to income in each state to maximize Equation (5.2). We introduce
the role of opportunities for adjustment to uncertainty through the specification
of the prices for these claims to income and the budget constraint. This is
given in general terms in Equation (5,3):
E = r1W1 + r2W2 , (5.3)
where
r. = the price for the claim to income in state i .
The problem can be stated equivatently as one of minimizing the planned
expenditures required to realize a given level of expected utility. This
approach provided the basis for the derivation of the planned expenditure
function in Chapter 4. For our simplified example, the conditions for a con-
strained minimum imply a function defined from the expenditure minimizing de-
mands for claims, as in Equation (5.4):
E[rr r2, Rq, EU] = r^, Rq, EU) + f*2W2^r1' r2' Rq' EU) * ''5'4>
The marginal value of a change in risk is simply the partial derivative of the
expenditure function with respect to the component of the risk that is assumed
to change. Thus, for a change in R, the marginal value, MVR, would be de-
fined as follows:
aw aw
MVR ~ ¥R = rl ™3R + r2 3R ' - (5'S>
where
MVR = marginal value of risk increment .
The principal objective of this section is to demonstrate how these marg-
inal values change with changes in the assumed opportunities available for indi-
vidual adjustment. However, before proceeding to that discussion, it is impor-
tant to relate MVR to the incremental analysis developed in Chapter 4 and to
earlier literature on valuing risk changes.
Consider a discrete change in risk from Rq to R^ (with Rq < R.p. The
value (loss) of the change is defined by Equation (5,6):
5-5
-------
.
rR1
AVR = J || dR = E[rr r2, F^q, EU] - E[ry RQq, EU ] . (5,6)
R0
In this case, the individual must plan to allocate more income to realize the
same level of expected utility. Thus, the change in VR is the maximum amount
he would be willing to pay to avoid the change. This value is completely anal-
ogous to the values defined in Chapter 4, although it, of course, does riot
distinguish a user or existence component. The reason is simple: we have
not provided a basis for the distinction in our description of the choice proc-
ess, If the state-specific preference functions were expanded to define more
specifically the implications of the exposure beyond a simple health effect, then
the total value and use value components specified in Chapter 4 can also be
defined for this case. We return to this issue in the next chapter and discuss
the relationship between use and nonuse, or intrinsic, values of a risk change.
Of course, it should be recognized that all of these classification schemes for
benefit components simply reflect the introduction of additional information into
the choice process.
It should be also acknowledged that this valuation concept is more gen-
eral than what has been used in earlier analyses of the value of risk changes.
For example, Jones-Lee (1974) defined the value as the maximum amount an
individual would pay to realize a reduced probability of a detrimental event.
For a comparable risk change (i.e., Rq to ), his definition would be as fol-
lows in our notation;*
R0qVt (WrP) + (1-R0q)V2 (Wg-P) = R1qV1 (W.,) + (l-f^q^ (W2>, (5,7)
where
P ¦= payment for reduced probability of exposure (with Rrt < R as
before).
This payment, P, was described by Jones-Lee as the Hicksian compensating
variation in wealth. It is riot the compensating variation, but rather the option
price for a change in the probability. The definition is directly comparable
to what Freeman [forthcoming b] has described as the option price correspond-
~See his Equation (10) on p. 839 for his definition of the "compensating
variation" for a risk change.
5-8
-------
tng to a change in the conditions of uncertainty. Moreover,, use of this defini-
tion as the basis for defining individuals' valuations for risk changes is less
general than our formulation because it assumes the individual is unable to
make state-specific payments.
This conclusion is easily appreciated using the Graham (1981] willingness-
to-pay locus. Values are defined with respect to changes in one point on the
Graham willingness-to-pay locus rather than in terms of expenditures as de-
fined in Equation (5.5). Depending on how the locus shifts with a change in
R, we can expect different individual valuations for the risk change. That
is, the valuations in this case are described by measuring how each of the
points of the locus shifts with the change in R. This is the point emphasized
by our planned expenditure function. Since the Graham locus is an alternative
means of describing the effects of the opportunities for adjustment on an indi-
vidual's valuation of a risk change, it provides the basis for a graphical illus-
tration of how opportunities for adjustment affect an individual's valuation of
risk changes,* To illustrate the difference graphically, it is convenient to
consider a small modification in Graham's framework. His locus describes the
alternative set of payments that would be made to realize some desirable access
conditions or level of a commodity that is valued by the individual under one
of his state-dependent preference sets. The locus maintains a constant expec-
ted utility when the favorable access or quantity is realized, but state-
dependent payments must be made with that level realized without making these
payments and without the improved conditions (or increased quantity).
We could easily modify this framework by assuming a given level of ex-
pected utility as our reference point, without specifying where it came from
*lt may be tempting to draw parallels between the relationship of the
planned expenditure function to the Graham willingress-to-pay locus and the
relationship of the expenditure function to the indirect utility function ~.rder
certainty. However, this would be incorrect. Total planned expenditures do
not enter the Graham locus. The Graham' locus is not specific to one statement
of the constraint set facing an individual but, rather, provides the basis for
characterizing all of them and their implications for what total utilities can be
realized. This is one source of error in the recent comment on Graham by
Mendelsohn and Strang [1984]. It should also be noted that Graham's use of
the locus is different than ours. His objective was to describe the vacation
measures for a certainty of supply of a resource in the presence of demand
uncertainty. Ours is to illustrate the implications of institutions for the valu-
ation of risk changes. I
5-7
-------
and consider an individual making state-dependent payments under two sets
of conditions. These payments, taken together with the level of exposure risk
R, maintain the constant expected utility level. The Graham locus incorporates
the change in a single locus, and our proposed modification breaks up the
process. Thus, the Graham payments would be equivalent to the changes (or
increments) in payments under this format.
Consider the change from Rq to again. In this case, we will consider
the change consistent with a risk reduction from to Rq (since it was earlier
assumed that Rq < R^). Equation (5.8) defines the conditions implied by a
change in R from to Rq and Its implications for the definition of two modi-
fied willingness-to-pay loci. The equation on the left of the equality defines
the original level of risk and the payments (y^ and -yg) that would be made
for it, and the equation on the right side of the equality defines the new,
lower risk and the consequent higher payments:
R1qV1 (W1 - + (1-Riq)V2 (W2 - y2> = RQqV ^ (W^ - ^) (5 g)
+ (1"R0q)V2 (W2 - v2) ,
where
R, > R0 .
This case is illustrated in Figure 5-1 by the shift in the Graham locus
from A (with R^ describing the risk of exposure) to B (with Rq describing
the risk of exposure). The option price is the maximum constant payment
(across states) that an individual would be willing to make to realize the lower
risk. In this case it is given by the difference between the intersections of
the two Graham loci corresponding to the left and right sides of Equation (5.8)
with the 45° line.. When the opportunities for adjustment are taken into
account, the model is then explicitly acknowledging the prospects for varying
the payments across the states of nature. If the terms of payment are given
by the slope of TT', then a measure of difference in the implied value of the
change in R is given by the difference in the intersection of these tangents
to the Graham locus (with slope (- jr1)) with the 45° line, EF.* Clearly, the
*EF will describe the change in expenditures norfn.aliz.ed by r1+r2. Thus,
if we assume that Wt and W2 are measured so that r!+r2=1, then it can be
interpreted as the change in expenditures.
5-8
-------
Figure 5-1. Graham locus with change in probabilities.
-------
values can differ from the option price because there is no necessity that the
locus will shift in a parallel fashion. The shift in the locus will depend on
the change in risk and on the change in the marginal utilities of income with
changes in the income realized in each state,
The specific nature of these differences can be developed by using the
planned expenditure function. Each assumption for the relationship between
the relative prices of contingent claims can be used to specify a different set
of opportunities for adjustment (and a different point on the Graham locus).
For example, if we consider the case of actuarially fair prices for claims, this
interpretation would imply that the relative price is equal to the ratio of the
probabilities, as in Equation (5.9):
_L = _5S_ (5 9)
r2 1_Rq '
This value corresponds to what Graham designates the fair bet point and yields
a specific relationship for the marginal value of a risk change,
A specific expression for this marginal value can be derived using Equa-
tions (5.2) and (5,3) together with the first-order conditions used in defining
the expenditure function. As discussed in Chapter 3, in the presence of actu-
arially fair markets, an individual wifl select claims so as to equalize the mar-
ginal utilities across states:
dV dV?
__ = __ (5.10)
Together with the partial differentials of Equations (5.2) and (5.3), this condi-
tion with respect to R can be used to describe the marginal value of a risk
change as follows. The total differentials for Equations (5.2) and (5.3),
assuming dq = 0 (substituting for and for Rq and (1-Rq)), are given
in Equations (5,11) and (5.12):
ap.j* dv aw dV aw
^ * 0 = qlV, (Wl*) - Vg(Wj «)] + Rq --I + (1-Rq) ^ ^
(5.11)
5-10
-------
3W 3W
Tr= * - W2*) + Rq w + (1 -Rq) ^ , (5.12)
where the asterisk designates the expenditure minimizing values.*
Using Equation (5.10) to simplify Equation (5.11) and substituting in Equation
(5.12), we have
aF q [V-(W *) - V (W *)] :
= q(W1* " w2*} + dv ¦ (5-13)
dW1
Since it is reasonable to assume that (W^*) > (W^*), we can conclude
that an individual will value a risk change by more than the expected insur-
ance (q (W^* - W^*)) that would be purchased at actuarially fair rates. This
is a variation on the case described by Cook and Graham [1977]. Under these
3 E 3 ^"E
conditions, we can expect that > 0 and —5 > 0. Thus, the marginal value
3R
of an incremental reduction of, say AR, in the risk of exposure to hazardous
waste will be greater at higher levels of risk. Differentiating Equation (5.13)
with respect to R and simplifying, we have
d2v aw
(V (W *) - V (W *)) y —
3 E dW £ 3R
= ._ , _L_ ¦ (5.14)
9R
This conclusion is readily established once it is recognized that our model
d*Vl
implies that V„ (W,*) > V (W *) and j- < 0.
£ 1 11 dW^
This same conclusion was derived by Jones-Lee [1974] in the case where
an option price was assumed to be. the mechanism for paying for the risk re-
*ln what follows, all derivatives are assumed to be evaluated at the rele-
vant optimal values, depending upon which constrained optimization problem is
discussed.
5-11
-------
duction and in somewhat more general terms by Weinstein, Shepard, and Pliskin
[1980] for the same payment mechanism. Of course, in the case of the option
price, it is a payment made to avoid an increase in the risk of exposure that
could lead to detrimental health effects. Thus, these option price results are
simply special cases of this more general framework.
The rationale for adopting this more general framework follows from the
fact that the mechanisms available to the individual for adjustment affect the
change in the marginal valuation of risk changes with the level of the risk.
32E
That is, we cannot unambiguously sign under alternative assumptions con-
cerning the opportunities for adjustment. To illustrate this conclusion, we
consider two such cases; (1) when relative prices of claims are related to
the likelihood of exposure, but not to the conditional probability of the health
effect given the exposure and (2) when relative prices bear no specific rela-
tionship to either of the probabilities involved.
The first of these cases contrasts with what conventional practice would
define as actuarially fair markets, where the relative prices of claims would
be tailored to the individual's circumstances by adjusting them to reflect the
conditional probability, q. If we assume that q (the conditional probability of
the health effect given exposure to hazardous wastes) reflects an individual's
health and overall heredity, then we might assume the first case gives each
individual a "fair" opportunity to adjust to the risk under policy control but
does not attempt to make distinctions for individual circumstances.*
9 E
Following the same logic outlined earlier to derive the expression for —
under actuarially fair markets, we have for this case
« (V (W ) - V (W ))
Jf = (W, - V ~ 2dVl .
-dW,
*An alternative definition would be to pick a value for q--a threshold--
and define the "fair" opportunities in terms of the joint probabilities at that
value of q. This approach could be considered analogous to the definition of
a sensitive group in the specification of the primary standards for the criteria
air pollutants. See Smith [1984a] for further discussion.
5-12
-------
where
the bar designates the expenditure minimizing values for this defini-
tion of the relative prices of contingent claims,
9 E
It is clear that — > 0. Indeed, increases in risk require greater planned
expenditures to maintain a given utility level. Unfortunately, without specific
assumptions concerning the nature of the utility functions, it is impossible to
establish a precise relationship between the marginal values implied by each
set of institutions.
We can establish the ambiguity in the size of the marginal value of a risk
change with the level of risk by differentiating Equation (5.16) with respect
to R:
ii
3R2
dV1 3W1
$
(5*18)
, _ , d2v aw
W " Vwi'j * ^ ' a-T
dV d V 1
Substituting for in terms of from the necessary conditions for an ex-
penditure minimum for this constraint, we have
/ _ \ dS 9^1
f(Vp (W ) - V (W )) -2 ¦ w
32E = / q - 1 \ aW2 V / dW1 . (5,1?)
9R^ V * 9R
5-13
-------
aw
2
3r '
since
The first term in Equation (5.17) will have a sign opposite to
3 £
q < 1, and Rq < 1. It is clear that 0 since, by the equivalent of Equa-
oK
aw„
3W„
tion (5.11) for this case, Rq + q(l-R) > 0 and W > W_. We would
oH oH I c.
3W2
therefore expect > 0 in this case and, in turn, that, the first term in
Equation (5.17) will be negative. The second term in Equation (5.17) is clear-
ly positive. Since the relative magnitudes of these two terms cannot be gauged
a priori, we cannot suggest how the marginal value of incremental reductions
in the exposure risk will change with the level of that risk.
The same conclusion arises for the case where no specific relationship is
assumed for linking the relative prices of claims and the relative odds of the
two states of nature. Equation (5.18) presents the marginal valuation for this
case, and Equation (5.19) the second partial derivative:
9_E
9R
V2 {VV " V1 (^1}
dW„
(5.18)
where
the tilda (~) designates the expenditure minimizing values for this spe-
cification for the contingent claims markets.
/dV, 9W
8R2
-
'T \dW2
__2
9R
dVj 3WA
cftCj ~8R J
dW„
V2 (W2} " V1
-------
As part of their analysis of individual valuation of risk changes in the
presence of state-dependent preferences, Cook and Graham [1977, p. 152,
n. 18. ] identified three components of the benefits of any risk change: (1) the
pure protection benefit associated with the risk reduction, (2) the value of
moving from an initial wealth distribution to an efficient (or more efficient)
one, and (3) the cost of financing the action inefficiently if the post-invest-
ment distribution of wealth is inefficient.
The theoretical analysis of this section has extended past efforts at defin-
ing a conceptual basis for valuing risk changes to permit an explicit treatment
of the character of the opportunities for adjustment to risk within a framework
that is consistent with this benefit taxonomy. It represents a specific example
of how the properties of the planned expenditure function can be used to con-
sider an individual's valuation of risk changes. Moreover, this modification
leads to a change in one of the more important testable implications of past
research on the valuation of risk changes. That is, it has been suggested
that the individuals' incremental value of a risk change will increase with the
level of the probability of the detrimental event. While this conclusion holds
where individuals have access to actuarially fair markets for contingent claims
or where they must make state-independent payments for the risk change, it
does not necessarily hold in other cases. As a consequence, a failure to ob-
serve an increasing incremental valuation of risk reductions may not imply
rejection of the expected utility model. It can also reflect the individual's per-'
ceived opportunities to undertake state-dependent adjustments in income claims.
Most of the literature in this area (see, e.g., Jones-Lee [1974] and Wein-
stein, Shepard, and Pliskin [1380]) has used the option price as the benefit
concept for defining how individuals would value risk changes, it has not
specifically described the role of institutions in influencing individuals' valua-
tions of changes in the conditions of uncertainty and therefore has not dealt
with the issues that are posed by our more general framework. Of course, in
the final analysis, the importance of this refinement depends on its empirical
relevance. For our purposes, this means that how individuals respond to a
contingent valuation question that elicits a state-independent bid may well be
affected by their ability, or indeed their perceived ability, to make state-
dependent adjustments. That is, if individuals accept the terms of the contin-
5-15
-------
gent valuation questions eliciting a bid for a risk reduction as the only means
available to them for adjusting, then the results of the existing literature on
changes In the incremental option price with the level of risk are relevant.
In effect, option price is the relevant measure of the benefits of reducing haz-
ardous waste risks. However, if individuals perceive themselves as having
the ability to take specific actions that would be the equivalent of state-
dependent payments, option price may not be the relevant measure.
5.3 IMPLEMENTING THE THEORY: PHYSOLOGICAL CONSIDERATIONS
To this point, our analysis has implicitly accepted the expected utility
framework as the basis for describing individual behavior under uncertainty.
White Chapter 3 briefly discussed violations of this framework, it argued that
these violations could be explained by either of two amendments to the frame-
work—the introduction of state-dependent preferences or the recognition that
individuals may adhere to an expected utility model but form their judgments
on the probabilities of state of the world in ways that have not been properly
modeled in past applications of the expected utility framework. These modifica-
tions change what Arrow [1974] refers to as the separation or independence
of the information on risk and that on preferences. As a consequence, it be-
comes impossible, without additional information, to distinguish the reason
(i.e., taste or risk perception) for specific behavior in the presence of uncer-
tainty. Moreover, either explanation is simply an alternative means-of express-
ing the analyst's ignorance of the factors influencing individual choice. In
the first case, state dependency acknowledges that utility (and, in particular,
the marginal utility of income) may vary with the state of the world. It usual-
ly does not offer an explanation of the features of the state or of the individ-
ual that account for the dependency in general terms. While it may be possible
to identify some factors in specific applications, no attempt has been made to
provide a complete or comprehensive descrption that would accommodate ail
cases.
Similarly, in the case of the available alternative explanations for proc-
esses used by individuals to estimate the probabilities of the states of nature/
a number of approaches for information processing were identified as offering
the potential for reconciling observed contradictions with the expected utility
model. However, the frameworks discussed were not part of an integrated
5-18
-------
explanation of how individuals process different types of information or
appraise the likelihood of different types of risk. This is a reflection of the
limited nature of our understanding of behavior under uncertainty.
As we noted at the outset of this report and this chapter, the primary
objective of this research has been to develop estimates of individuals' valua-
tions of reductions in the risk of exposure to hazardous wastes. Since there
are specific aspects of this problem that may affect how individuals respond
to valuation questions, it is important to review the general-features that past
research has indicated may play a role in Individual behavior. While this is
not a substitute for a comprehensive model that describes the role of aM the
features of the circumstances that may influence behavior, It is nonetheless a
complement to our conceptual model. That Is, it serves to indicate the poten-
tial limitations of our framework and to highlight the factors that must be ex-
plicitly considered in implementing it for empirical purposes.
The most important of the factors influencing individual behavior for an
analysis of hazardous waste risks would seem to be what Slovic [1984] refers
to as "dread risk" and "known risk." Our case embodies both. The first
involves the notion that an event is dreaded because it is potentially cata-
strophic. The second quality relates to both perceptions about the individual's
knowledge of the risk and whether the events at risk are delayed in time.
Based on the research of Slovic and his associates, Slovic has suggested that
these factors influence how individuals respond to uncertainty. This would
imply that individuals may well value incremental reductions in the risk of
death from different sources quite differently.* Moreover, based on these
*There is an important Issue that arises in modeling an individual's valua-
tion decisions concerning risk changes. It arises because the models have
routinely assumed the risk of interest is the only one the individual faces, tf
instead the individual faces multiple risks, and policy is intended to change
one of them, then the problem becomes much more complex. Kihlstrom, Romer,
and Williams [1981] offer some initial insights into the general problems raised
by these cases.. They note that even if the risks are independent, ordering
individuals by Arrow-Pratt measures of risk aversion will not necessarily cor-
respond to the ranking implied by the certainty equivalents. Indeed, the whole
problem of characterizing risk aversion and individuals' responses to risk be-
comes more complex in these cases. For example, to the extent sources of
risk are correlated (especially negatively correlated), engaging in some risky
activities may be an approach to risk diversification or have a role akin to
institutions that affect the valuation of risk changes. We have not considered
5-17
-------
considerations, we might expect that an individual would have a greater valuat-
ion of risk reductions where the dread and known factors are present. To
develop some information relevant to the potential effects of these factors we
have designed our survey to include valuation questions for changes in two
types of risk—exposures to hazardous wastes and fatal accidents on the fob.
In addition, in the process of questioning individuals about their valuations of
reductions in the risk of exposure to hazardous wastes, we consider two vari-
ations on the questioning format. First, we investigated whether a change in
the health end state (e.g., whether the cause of death was due to damage to
the body's immune system or whether the risk was associated with birth defects
severe enough to mentally retard or physically handicap children for a lifetime)
would alter the individual's valuation of the risk change. Second, our experi-
mental design allows for consideration of both low levels of risk (where an
individual may well assume our knowledge of processes leading to rate events
is imprecise) and bids for the elimination of risk. To the extent dread and
the imprecision of knowledge of the risk would affect individuals' valuations of
risk reductions, we would expect to see their effects evidenced in responses
to these different elements in our design. Moreover, it was also possible to
gauge the effects of these factors on risk perception by asking each individual
their perceived risk of death from four causes — an automobile accident, heart
disease, a disease caused by air pollution, and a disease caused by exposure
to hazardous wastes. As the discussion of the design and structure of the
questionnaire in Chapter 8 describes in more detail, these questions were posed
before any valuation questions and provide the basis for evaluating how risk
perceptions vary for these different types of risks.
In addition to these factors, the controllability and voluntariness of risk
have been found to be important elements in psychological studies of risk per-
ception . These features were also considered in the sturcturing of our analy-
these issues here, but acknowledge that they are clearly relevant to any
empirical efforts since, in the real world, individuals do face multiple risks.
This extension provides another potential explanation for the difficulties exper-
ienced in interpreting the results of field experiments within an expected utili-
ty framework. For the most part, these efforts have tended to ignore other
sources of risk. See Smith [1984bJ for discussion of some of these issues as
they relate to the measurement of risk aversion.
5-18
-------
sis. Based on a series of focus group discussion sessions (described in more
detail in Chapter 8), we found that individuals associated differing degrees of
control with respect to the siting of disposal site for hazardous wastes with
whether their community had a "say" in the decision.*
Equally important, and potentially related to the treatment of these two
features of risk, is a concept discussed by Hershey, Kunreuther, and Schoe-
maker [1982] that might be described as assignment of the risk. That is,
what are the assumed "property rights" of the individual for the level of risk
to which he is endowed? Our design reflects all three considerations. First,
in the structure of the contingent valuation experiment, we considered two
types of scenarios—payments for reductions in risk and payments to avoid in-
creases in risk. In the latter case, our survey design varied the sources of
the risk according to whether it was allowed by the Federal government or
voted for by the individual's town council. In addition, our design also
allowed evaluation of a situation in which risk increased but so did an individ-
ual's income.t Finally, comparison of valuations for reduction in hazardous
wastes with those for risks on the job will also reflect the effects of voluntary
selection because the latter were posed as being associated with new jobs and
the individual is asked a wage increment that would induce him to accept a
job with the new risk conditions.
AH of these factors fall within the general category of context effects.
They imply that how a risk is explained to an individual may well influence his
response to it (see Schoemaker [1982], PP- 547-48, for further discussion).
Rather than interpret them as potentially implying some form of irrational be-
havior, they can also be interpreted to indicate that analysts have done a poor
job at communicating their questions to survey respondents.
One of the important aspects of the design of the empirical component of
this research has been the use of focus groups in the development of the
*This is clearly consistent with findings observed in studies of the siting
of nuclear facilities. See, e.g. , Carnes et al. [1982], Carnes and Copenhaver
[1983], and Carnes et al. [1983]. It is also consistent with the program of
research recently described by Kunreuther and Kleindorfer [1984].
tThis was done using a contingent ranking format that is described in
Chapter 14. A general discussion on the use of the method in benefit estima-
tion is given in Desvousges, Smith, and McGivney [1983].
5-19
-------
wording of the questionnaire, the format of the vehicles used to explain risk,
and in the pretesting and revisions to the questionnaire. Since the specific
steps in this process are described in Chapter 8, it is sufficient to acknowl-
edge here their role in adjusting the structure of the empirical component of
this research to reflect what has been learned for the psychological research
on decisionmaking under uncertainty,
5.4 SUMMARY
This chapter has used the framework of the planned expenditure function
to describe how individuals would value risk reductions. It has illustrated
how these valuation concepts will be affected by the mechanisms that are
assumed to be available to the individual for adjustment. By using a simple
two-state example, it has been possible to relate the valuation concepts to both
the past literature on the valuation of risks of death and to the discussion of
option price as a valuation concept in environmental economics,
As acknowledged at the outset, our focus to this point has been on what
might be designated ex ante use values. It is reasonable to expect that indi-
viduals may hold a form of ex ante existence or intrinsic values for risk reduc-
tions because they serve to affect other aspects of the natural environment
whose existence yields utility even though they do not provide user services
in the conventional, consumptive sense. In the next chapter this general
framework is used to discuss how these values might be defined and integrated
into an ex ante perspective for benefit analysis. Following that, we introduce
the discussion of the design of our questionnaire and survey with a chapter
describing the relationship between the conceptual analysis and the constraints
within which it was implemented.
5-20
-------
CHAPTER 6
ECOLOGICAL AND INTRINSIC VALUES UNDER UNCERTAINTY
6,1 INTRODUCTION
As described in Chapter 2, hazardous wastes pose risks to ecological sys-
tems as well as to human health, and hazardous wastes regulations can reduce
'these risks. The purpose of this chapter is to extend the analysis of the val-
uation of risk reductions presented in Chapters 3 through 5 to consider the
problems posed by developing a consistent system for valuing reduced risks
to environmental and ecological resources. However, this extension first re-
quires a consideration of the nature of the economic values people derive from
ecological resources.
Ecological systems can yield benefits to people in a variety of forms.
For example, both managed and natural ecosystems can yield food or fiber for
market. In such instances, the ecological system is an input to a production
process that also involves capital and labor in the cultivation and harvest of
plant and animal species. We might call these production or market benefits
because the harvest activities are undertaken in response to market forces
and profit incentives. The benefits of changes to ecosystems used for market
purposes come in the form of changes in the prices of goods and factor inputs.
This is in contrast to those human actions involving uses of the ecological sys-
tem that yield utility directly to the individuals concerned. Examples of such
direct use benefits include the values attributable to recreation activities such
as hunting, fishing, wildlife observation, and nature photography.
It has also been argued that natural environments, including their ecologi-
cal components, can yield benefits that are not associated with their direct
use. This class of benefits has been variously named intrinsic, nonuser, and
nonuse benefits. Such benefits are said to arise from a variety of motives,
including the valuing of the knowledge of the existence of a particular environ-
mental or ecological attribute, a desire to bequeath certain environmental assets
6-1
-------
to one's heirs or to future generations, and a sense of stewardship or respon-
sibility for preserving certain features of natural environments.
One of the objectives of this chapter is to develop a logical and consistent
set of definitions and concepts that can guide further theoretical analysis and
empirical testing of propositions about intrinsic values. Toward this end, Sec-
tion 8.2 is devoted to a systematic examination of the several types of intrinsic
benefits associated with ecological resources that have been discussed in the
literature. This section also considers alternative ways of "specifying prefer-
ence functions to reflect the various forms of intrinsic benefits. One issue
here is the particular circumstances under which it is possible (or meaningful)
to partition a total benefit measure into components--e.g., use, bequest, pure
existence, and so forth. Another Issue concerns the relationship between in-
trinsic benefits arid the benefits associated with the direct use of the environ-
ment. Section 6.3 extends the discussion of existence values to the situation
in which a policy alters the probability that the resource will exist and consid-
ers further the implications of ex ante versus ex post perspectives for the
valuation of risk changes. Section 6.4 offers some conclusions and discusses
the implications of this analysis for approaches to measuring ecological values.
6.2 EXISTENCE AND USE VALUES UNDER CONDITIONS OF CERTAINTY
In this section we take up several questions concerning the relationship
between use and existence values and possible motivations for existence value.
In all cases this analysis maintains the assumption of certainty. Let us assume
that an individual derives utility from the consumption of a vector of private
goods, X, and some measure of the quality of the ecological system at the site,
0. In this general formulation, Q can be taken to be a scalar measure of some
critical characteristic—e.g., the population or biomass of an important species
or the number of different plant or animal species present in the ecosystem.
Alternatively, Q could be interpreted as a dichotomous variable taking the
value Qt = 0 in the absence of some critical ecological attribute and the value
t?2f -> Qi)) when that attribute is present. I n the latter case, the marginal
utility of Q is assumed to be positive in the interval Qr - Q2 and 0 otherwise.
To give the problem additional structure, let Xx be some market good
associated with use of the ecological resource. Examples could- include the use
6-2
-------
of the services of a site for purposes of recreation, rental of a boat for fish-
ing, or hiring the services of a guide to conduct a visit to an ecological re-
source. If Xj ¦= 0, we interpret this to mean, that the ecosystem has not been
used by the individual.
Assume that the individual maximizes utility subject to the budget con-
straint M - P'X = 0, where P is a vector of goods' prices. The solution to
this maximization problem yields a set of demand functions for X. In the ab-
sence of further restrictions on the form of this utility function f the demand
functions can be written as Equation (6.1):
Xi = Xi(P, M, Q) . (6.1)
The minimum expenditure necessary to attain any given level of utility is
E = E(P, Q, U) . (6.2)
If U* is the solution to the utility maximization problem given P, M, and Q»
then the compensating surplus measure of the benefit of an increase in Q from
Qx to Q2 is given in Equation (6.3):
S = E(P, Qt, U*> - E(P, Q2, U*)
(6.3)
q2
= - X 3E/3Q ¦ dQ .
Qt
In this general formulation, S could be a pure use value, a pure nonuse
or existence value, or some combination of the two. If the conditions defining
Maler's weak complementarity hold, then S is a pure use value (Maler [1974] I-,
Two conditions on the utility and demand functions must be satisfied in order
to fit Maler's definition of weak complementarity. First, there must be a value
for Plf designated as Pf(Q), such that the demand for Xx is zero:
Xx = XjCPfCQ), P2, ... PR, M, Qa) = 0 . (6.4)
6-3
-------
And, second, at that price, the marginal utility or marginal welfare of changes
in Q must be zero:
3E (Pf(Q, P2, Pn, Q2/ U* )/'9Q = 0 , (6.5a)
or, equivalently,
i
3U (O, X2, ... Xn, Q2)/9Q = 0 . (6,5b)
As is now welt known, the conditions defining weak complementarity also
allow this pure use value for changes in Q to be estimated by appropriate anal-
ysis of the demand function for X y. Specifically, S is equal to the area be-
tween the compensated demand curves for Xt when Q increases from Qx to
Q2.* That is, S is defined by Equation (6.6):
Pf
S = J Hi (Pw P2, P , U*, Q2)dP1
P<
(6.8)
Pf
- J H t (P j, P2, P , U*, Q1)dPl ,
p,
where
H i = the compensated demand function for Xt
Pj = the given market price.
The process of using ordinary demand functions to approximate the com-
pensating surplus measure of a use value for a quality change can be complex.
If we are willing to assume that the quality change affects the effective price
(or the quantity of the resource services, Xt in this case), then the analysis
of Willig [1376] or Randall and Stoli [1980] can be used to describe how S
can be approximated by the area between the ordinary (i.e., Marshatlian) de-
mand functions for Xx at the two levels of Q, In the case of Q acting through
the price, we are essentially maintaining that the effect of a change in Q is
the same as the effect of the corresponding changes in the price of XP If the
*For an elaboration, see Maler [1974], pp. 183-89 or Freeman [1979],
pp. 72-75.
6-4
-------
role of Q in the demand for X1 cannot be distinguished in this way, then the
relationship between the Marshailian and Hicksian demand functions depends
on properties of the expenditure function that are not considered in either
the W'illig or Randall and StolI analyses. Consequently, these approaches can-
not guide an evaluation of the relationship between Marshailian and Hicksian
measures of the welfare change associated with a change in ecological qua lity.
Pure existence value occurs when X1 = 0 at all Pi > 0 but 3U/3Q > Q--
i. e., when the second condition defining weak complementarity is violated.
Given this condition on use, pure existence value EX is given by Equation
(6,7):
EX = E(P, Qx, U*) - E(P, Q2, U*) , (6.7)
where
Qi - 0
Q2 > 0.
The necessary and sufficient condition for pure existence value is that the
utility function be strongly separable in Q. One consequence of this strong
separability is that changes in Q have no effect on market behavior. Thus,
there is no basis for estimating pure existence values from observations of
changes in market prices or quantities.
Some authors have questioned the plausibility of a pure existence value
that is truely independent of any use of the site. In justification for pure
existence value, Krutilla suggested that, "An option demand may exist, there-
fore, not only among persons currently and perspectively active in the market
for the object of the demand, but among others who place a value on the mere
existence of biological and/or geomorphological variety and its widespread dis-
tribution" [Krutilla, 1967, p. 781], In an accompanying footnote, he also sug-
gested that the "phenomenon discussed may have an exclusive sentimental
basis, but if we consider the bequest motivation in economic behavior, dis-
cussed below, it may be explained by an interest in preserving an option for
one's heirs to view or use the object in question" [Krutilla, 1967, p. 781, n].
Later, Krutilla and Fisher wrote,
Perhaps closely associated with option value is the value some indi-
viduals derive from the knowledge of the existence of unspoiled wil-
6-5
-------
derness, wild and scenic rivers, and related phenomena of peculiarly
remarkable quality. ... In the case of existence value, we con-
ceived of individuals valuing an environment regardless of the fact
1 that they feel certain they will never demand in situ the services it
provides . , , however, if we acknowledge that a bequest motivation
operates in individual utility-maximizing behavior . . », the existence
value may be simply the value of preserving a peculiarly remarkable
environment for benefit of heirs, (Krutilla and Fisher [1975], p. 124)
While Krutilla and Fisher offer a bequest motivation as one of several pos-
sible explanations for a pure existence value, McConnell takes a different point
cjf view:
The notion that a good is valued only for its existence, that it pro-
vides no in situ services, is far fetched. In most cases, resources
are valued for their use. Existence value occurs only insofar as
bequest or altruistic notions prevail. We want resources there be-
cause they are valued by others of our own generation or by our
heirs. Thus use value is the ultimate goal of preferences that yieid
existence demand, though the existence and use may be experienced
by different individuals. (McConnell [1983], p. 258)
In contrast to McConnelI's view, Randall and StolI recognize that people
might experience other than altruistically motivated benefits from the existence
of a site without visiting the site. However, they argue that all such non
in situ uses are associated with some aspect of market-related behavior and
that these values thus constitute a form of use they label "vicarious consump-
tion": "Thus, we consider the values generated by reading about Q in a book
or magazine, looking at it in photographic representations, for example, to be
use values. Clearly our definition of use includes vicarious consumption"
(Randall and Stotl [1983], p. 267). In terms of our model, they view Q as
enhancing the utility of perhaps severaf goods in the vector X,
Neither McConneK nor Randall and Stoll recognize concern for the exist-
ence of a species out of ethical considerations as a possible motive for pure
existence values. While ethical philosophers are not in agreement as to the
validity and proper form of such concern,* it is possible that some people hold
such values and are willing to commit resources on that basis.
This discussion of the possible motivations for pure existence value is
inconclusive. This is at least in part because some of the arguments of the
*For discussion of these issues, see, for example, Norton [1982], Sagoff
[1980], and Rescher [19801, PP- 79-92,
6-6
-------
authors cited are misdirected in at least two respects. The first concerns var-
ious definitions of existence value. Definitions can be considered in part a
matter of taste. A set of definitions can be considered useful if it furthers
the research objectives and leads to useful answers to meaningful questions
and if the definitions are based on operationally meaningful distinctions. If
use values are limited by definition to those associated with in situ uses, these
definitions have the virtue of distinguishing between cases where use of a site
generates observable data and cases where no meaningful data can be obtained
by observing market transactions.
One problem with so-called vicarious uses is that the observable market
transaction--e.g., the purchase of a nature magazine—often entails the simul-
taneous or joint use of many environmental resources so that allocation of the
market transactions to specific resources is not possible. Furthermore, vicari-
ous use has the odd feature that use can occui—e. g., through viewing of film
and photographs--even though the resource no longer exists. Finally, where
vicarious uses involve information conveyed by photographs and so forth, the
public good dimension of information seems likely to virtually destroy any mean-
ingful relationship between observed market behavior and underlying values.
The second respect in which the preceding arguments may be misdirected
has to do with the role of possible existence values in policy analysis. We
are concerned with the question of existence values because resource misalloca-
tions will result if they are of significant size, unmeasured, and therefore omit-
ted from benefit-cost calculations. The arguments about motivations for exist-
ence values seem to be offered for the primary purpose of persuading the
reader of the plausibility of the hypothesis that existence values are positive.
But the real test of this hypothesis will come from the data. Rather than fur-
ther debating definitions and possible motivations, the most useful step would
be to proceed with a test of the hypothesis that existence values (defined in
a way to make testing of the hypothesis feasible) are positive. If the evidence
supports this hypothesis, then further research efforts might be devoted to
testing hypotheses about the determinants (motivations) or the size of existence
values in different cases. Thus, consideration of the motivations for exist-
ence values is important (at this stage in our empirical research) only to the
extent that these motivations affect the discussion used to explain the concept
to individuals in a contingent valuation framework.
6-7
-------
So far we have considered two polar cases in which value accrues to indi-
viduals only through use (weak complementarity) and in which value is entirely
independent of use (pure existence value). Now we take up the intermediate
case where value accrues through use but the conditions defining weak comple-
mentarity do not hold. Using the model of preferences developed in this sec-
tion, we will show that there is a subtle distinction between existence value
and nonuse value when there is some level of Q (e.g. , Q = 0) at which no
use is possible at any price for Xj. Finally, we wilt consider the problems of
measuring the total benefit and its components by various techniques.
The use value of the site being visited is the increase in expenditure
necessary to compensate for an increase in the price of a visit sufficient to
reduce the number of visits to zero. Thus, this value provides a dollar meas-
ure of the welfare change associated with the use that takes place at the exist-
ing price, Pi. To measure the use value of a quality change, we are interest-
ed in how the welfare change associated with having these access conditions
(i.e., the price of Px) would itself change with a change in the quality of the
resource. Thus, the use value of an increase in quality from Qt to Q2 (where
Q? > Qi) is the increase in the use value of the site:
Sy = E[Pf(Q), P2, ..., Pn, Q2, U*] - E(P1; P2, . Pn, Q2/ U*)
(6.8)
" E [Pf (Q) / P2, Pn, Qi, U*] + E(P x, P2, . Pp, Qi, U*) ,
where
Pf(Q) - the price at which Xx = 0
Pi = the original price per visit.
Implicit in this formulation is the assumption that X, > 0 at PT and Ql. Notice
that Sy can be defined only if there is a price that chokes off demand. 5^.
can also be measured by the area between the compensated demand curves for
X J at the two levels of Q.
Now let us define nonuse benefits as that change in expenditure that
holds total utility constant given that the price of visits is so high as to elimi-
nate use of the site. In terms of the expenditure function, nonuse benefits,
S^, are defined as follows:
6-8
-------
SN = E[Pf(Q), P2, p , Qi, U*]
9)
- E(Pf(Q), P2, Pr, Q2, U*) ,
According to this definition, existence benefits can be positive for potential
users and even for those who do use the site when Pt is less than Pf.
Now define an individual's totaf benefit from a change in Q as the sum of
that individual's use benefit and nonuse benefit:
S = Sy + SN . (6.10)
Substituting Equations (6.8) and (6.3) into (6,10) gives the following expres-
sion :
S = E [Pf (Q), P2, Pn, Q2/ U*] - E (P t, P2, ... Pn, Q2, U*)
- EIPf(Q), P2, . P , Qlf U*] + E( P!, P2, P , Qx, U*)
n n (6.11)
+ E [ Pf (Q), P2, Pn, Qlf U * ] - E[Pf(Q), P2, .. ., Pn, Q2, U*]
= E(PL, P2, Pn, Q,, U*) - E(P!, P2< ... Pn, Q2, U*) ,
This expression gives the increase in the value of a resource as it increases
in size or quality. But it does not shed any light on the value of existence
versus nonexistence of the resource. Let Q represent the minimum level of Q
at which it can be said that the resource exists. Clearly, Q represents a
threshold or minimum viable level of the resource. At Q, use value is given
in Equation (6.12):
5U = E[Pf(Q), P2, Pp, Q, U*]
- E(P j, P2, ... Pn, Q, U*) .
(6.12)
Existence value is given in Equation (6.13):
SE = E [ Pf (Q), P2, . Pn, Qi, U*]
- EIPf(Q), P2, ... Pn, Q, U*] ,
(6.13)
where Ql < Q.
Defining total value S as gives
6-9
-------
5 = E[P*(Q), P2, Pn, Q i, U*J
" E(Plt P2, ... Pp/ Q, U*) .
(6,14)
Comparing the first terms of Equations (6.11) and (6.14) is instructive. If
Qj_ is less than Q, the first term in Equation (6.11) does not accurately convey
the implications of the connection between Xl and Q, especially the manner in
which that association constrains the decisions an individual can make. The
conventional expenditure function is the solution to the dual of the utility max-
imization problem subject to the usual constraints. When the level of Q exceeds
0> the constraint on Xt is not binding, and, consequently, the form of the
expenditure function will be different than when this constraint is applied.
More specifically, with Q < Q, must be zero, and greater expenditures are
required to realize the utility level U*. The first term in the reduced version
of Equation (6.11) does not make this apparent. It appears that the only con-
straint to the level of consumption of Xt is the price, Pj. The distinction
would be apparent if we solved analytically for the expenditure function using
some specific functional form for the utility function. When Qt is assumed to
exceed Q, the selections of all X/s can be assumed to be consistent with an
interior solution. However, when Qx is not greater, then the solution involves
a boundary value or corner in (i.e., with X1 = 0). There is, however,
some additional information we can use. This case must be equivalent to the
expenditures made at level Qx when Pj = Pf(Q). Thus, we can use this infor-
mation and substitute in Equation (6.11) for the first term to derive Equation
(6.14).
In conclusion, the total nonuse benefits of an increase in Q can stiil be
defined as in Equation (6.9). But if Qx < Q, nonuse benefits have two com-
ponents, one related to existence atone (S^) and one related to magnitude or
s ze of the resource.
What does this analysis imply about the measurement of existence and non-
use values? The first implication is that nonuse value and use value can only
be meaningfully distinguished in those cases in which there is some price (Pf)
above which use drops to zero. The definitions of both use and existence
values are predicated on the existence of some price at which use falls to zero.
And that can be assumed only if there is some nondivisibility in Xx such that
6-10
-------
quantity demanded must be zero at Pj > M. Otherwise, total value can be
defined as in Equation (6.11), but no allocation between use and nonuse value
is possible. Second, Maler's definition of weak complementarity is equivalent
to saying that nonuse values are zero. But if weak complementarity does not
hold, and if the present price of the visit is equal to or greater than Pf, then
use value is zero while existence value may be positive. And at any price,
even "nonusers" might become users if the price of a visit were to fall suffi-
ciently.
Third, as Maler has shown, even if a complete system of demand functions
for X has been estimated on the basis of market data, the expenditure function
cannot be recovered unless the conditions for weak complementarity hold
(Maler [ 1974J, pp. 121-25, 183-89). But positive existence value implies the
violation of the weak complementarity conditions. Thus, if existence value is
positive, the total value of a change in Q cannot be estimated from observa-
tions of market data. It appears that contingent valuation techniques must be
relied upon in this case.
The fourth implication concerns the measurement of use value. The
accurate measurement of use value requires knowledge of the compensated de-
mand function for visits. But this demand function cannot be recovered from
market data unless the conditions for weak complementarity hoid--i .e. unless
existence value is zero.- Of course, in some cases (as described above), use
value can be measured as a reasonable approximation through the use of the
ordinary demand functions for visits.
What can be said about measuring or 5^ for users? One approach
would be to use contingent valuation techniques to estimate total values for a
set of users and use market techniques such as the travel cost model to esti-
mate Sy for the same group. A comparison of the estimates of S and 5^ would
constitute a test of the hypothesis that nonuse values are positive for users.
Another approach is to ask people their willingness to pay for an improvement
in Q or to preserve an ecological site of given Q even if they knew they would
never be able to visit the site. This is the approach taken by Desvousges,
Smith, and McGivney [1983] to estimate existence values for water quality in
the Monongahela River. One problem with this approach is that it asks people
to place themselves in a counterfactual situation. It might be helpful to pro-
vide an explanation as to why they should imagine that they would not be able
6-11
-------
to visit the site.* for example, they might be told that the price of visits
had been increased to some very high number, effectively choking off demand
for visits. Or they might be told that all visits had been banned to prevent
damage to some fragile component of the ecosystem.
Finally, individuals might be asked to reveal their total value and then
asked to allocate this total between use and nonuse values. One problem with
this approach is that respondents typically are given no guidance as to what
conditions to assume when they perform the allocation. Since nothing is said
in this sort of question about the assumed price of visits, there is no reason
to believe that the respondent's mental processes will reproduce the conditions
defining existence and nonuse values stated above,
6.3 UNCERTAINTY OF EXISTENCE
In this section we extend the discussion of use, nonuse, and existence
values to the case where the individual is uncertain as to the existence or sup-
ply of the environmental resources. We assume the individual has assigned
probabilities to the two states of nature--the resource exists, Q = Q; and the
resource does not exist, Q = 0. We develop measures of value for regulation
that cause the individual to revise upward the probability that Q= Q. And
we consider the possible relationship between these measures of value and
observable ex post measures, namely changes in expected use values.
Our analysis will also restrict the general framework for describing indi-
vidual choice that was discussed in Chapters 3 and 4 by assuming there is a
specific source for the state dependency in individual utility functions. Recall
that, in Chapter 4, the planned expenditure function was defined by acknowl-
edging the existence of state-dependent utility functions but without describing
the factors that caused the marginal utility of income to vary with state. Here
*ln the Desvousges, Smith, and McGivney [19S3] effort this was done
through the use of the value card. This interviewer aid was used to explain
to respondents the different potential types of values for a water quality im-
provement including the use, option and existence values. After a few ques-
tions designed to provide respondents practice with the proposed taxonomy,
the framework was used to elicit the components of the total vaiue of the re-
source. It is, of course, an open question as to whether this approach facili-
tates the task that confronts survey respondents.
6-12
-------
we will assume that the state dependency arises exclusively as a result of the
existence of the environmental resource. Our framework is completely consist-
ent with a state-dependent specification with two states. We have simply given
a somewhat more specific form to the function by specifying that
Uitxa = U(Xlf 0)
and
U2(X2) = U(X2, Q).
Let qv and rt (rl < qa) be the probabilities that the resource will not
exist in the absence of and with the policy, respectively. So, q2( = 1 - qj )
and r2( = 1 - rx) are the probabilities of existence or supply — with r2 > q2 -
Expected utility in the absence of the policy is given in Equation (8.15);
E(U )* = q, U(X|, 0) + q2 U(X2, Q) , (8.15)
The subscripts on the goods vectors X. allow for the possibility that purchases
of market goods—visits to the resource in particular—will be affected by the
availability of the resource.
As discussed in Chapter 5, assuming that the individual minimizes the
planned expenditures required to realize a given expected utility subject to
the set of contingent prices, the planned or ex ante expenditure function can
be written as Equation (6.16):
l[P/ qi/ Q, E(U)*] , (8.16)
where
P = the vector of prices for contingent claims.
The ex ante benefit of the poiicy that raises the probability of supply is the
decrease in expenditure to attain E ( U ) * made possible by the higher probability
of supply as given in Equation (8.17):
S = E[P, q,, q2/ Q, E(U)*] - E[P, r,, r2, Q, E(U)*] . (6.17)
6-13
-------
This measure captures both ex ante use and existence values. Estimates of S
might be elicited by appropriately designed contingent valuation questions that
describe the ecological resource and the change in risk associated with the
pol icy.
Often analysts have estimates of ex post use values derived from observa-
tions of actual users based on, for example, the travel cost site demand model.
A natural question is whether ex post use values can be the basis of estimates
of ex ante values. Two kinds of problems are encountered in trying to calcu-
late ex ante values from observed ex post use values. The first problem, of
course, is that there is no logical relationship between use values and nonuse
and existence values, even for users. So to the extent that nonuse values
and existence values are significant, observations of use values will yield
underestimates of total values. Moreover, the error potentially could be quite
large.
The second kind of problem arises because of the difference in perspec-
tives between the desired ex ante value and the observed ex post value. The
remainder of this section expands upon the material developed in Chapter 4
by focusing on ecological values and the relationship between ex post and
ex ante use values within a framework that assumes a specific source of the
state dependency in utility and a specific institutional framework for individual
adjustment in response to a risk change. This focus permits an evaluation of
the nature and extent of possible errors involved in using ex post values as
estimators of ex ante use values.
To focus attention on use values, we assume that nonuse and existence
values are zero. We also assume that income and prices are constant across
states of nature. Finally, we assume that there are no contingent claims mar-
kets and that state-dependent payments for the resource are not feasible, so
that the maximum state independent payment or option price (OP) for the
reduction in risk is the relevant ex ante welfare measure. Option price is
that constant payment for the policy that makes the expected utility with the
policy equal to expected utility without the project. It is the solution to Equa-
tion (6.18):
6-14
-------
qtV(Y, 0) + q2V(Y, Q)
($.18)
= r!V(Y - OP, 0) + r2V(Y - OP, Q) f
where V( •) is the indirect utility function associated with U(*). It is assumed
constant in all states and Y is income.
Since the ex post use value, 5y , is the solution to Equation (6.19), I
V(Y - Su, Q) = V(Y, 0) , (6.19)
Equation (6.18) can be written as Equation (6.20):
QiV(Y - Su, Q) + q2V(Y, Q) = r,V(Y - OP, 0) + r2V(Y - OP, Q) , (6.20)
In the most general analysis, four possible patterns of supply uncertainty
and risk reduction can be distinguished on the basis of whether the policy
eliminates uncertainty (r2 = 1) or not (r2 < 1) and whether or not there is a
possibility of supply in the absence of the policy.* These cases can be sum-
marized as follows:
Case A: No policy, no supply.
With policy, sure supply--q2 =0, r2 = 1.
Case B: No policy, possible supply.
With policy, sure supply--q2 >0, r2 = 1.
Case C: No policy, no supply.
With policy, possible supply--q2 =0, r2 < 1.
Case D: No policy, possible supply.
With policy, possible supply-~G < q2 < r2 < 1.
The relationship between OP and the expectation of 5y can be analyzed
for each of these cases by imposing the appropriate probability conditions on
Equations (6.18) or (6.20) and solving for OP.
For Case A (q2 =0, r2 = 1), Equation (6,20) reduces to Equation (6,21a);
V(Y - Su, Q) = V(Y - OP, Q) , (6.21a)
~This analysts is based on Freeman [forthcoming b].
6-15
-------
Therefore,
OP = Su = 0, r2 = 1), Equation (6.20) becomes Equation (6. 22):
'
qiV(Y " V + ^2V(Y' = V(Y " OP' Q) ¦ (6• 22)
Btshop [1982] and Brookshire, Eubanks, and Randall {1983] present (respec-
tively) mathematical and graphical proofs that option price is greater than ex-
pected use value for risk averse individuals.* A graphical proof can be
presented with the aid of Figure 8-1, which shows utility as a function of in-
come, given that the resource is available. Assume that q2 = 1/2. The left
side of Equation (8.22) gives E(U )* as shown in the figure. Now suppose
that with the program the individual must make a state-independent payment
equal to (r2 - q2)S^ = 1/2 Sy The expected utility of this payment scheme
is E(U), > ECU)*. Thus, the maximum state-independent payment or option
price is greater than 1/2 5y. The intuition is straightforward. In the ab-
sence of the program', the individual, in effect, holds a lottery on Q. The
risk-averse individual would pay more than the expected monetary equivalent
of the lottery (expected 5y) to eliminate the uncertainty associated with the
lottery. The excess of option price over expected Su is a risk-aversion premi-
um or supply side option value in Bishop's terminology.
For Case C (q2 = 0, r2 < 1 ), Equation (6.20) becomes Equation (6.23):
V(Y - S , Q) = rJV(Y - OP, 0) + r2V(Y - OP, Q) , (6.23)
*Both papers were concerned with a different formulation of the question .
They defined supply side option value as the difference between OP and ex-
pected 5 . and asked whether supply side option value was positive or nega-
tive.
6-16
-------
Figure 6-1. Option price and expected uss value with risk aversion.
6-17
-------
in this case, the sign of option value is indeterminate. A mathematical proof
requires the introduction of two new terms.*
Let Y* = Y - OP and define S * by
V(Y* - S *, Q) = V(Y*, 0) . (6.24)
Strict concavity of V in income implies:
V[rj(Y* - S*) + r2Y*, Q] > TjV(Y* - S *, Q) + r2V(Y*, Q) . (8.25)
Using Equations (8.23), (6.25), and the definitions gives
Thus,
V(Y - OP* - S* + r2SJ, Q > V(Y - SLJ Q) . (6,26)
OP - S,* + r2S* >'SU,
or, after some rearrangement,
OP < r2S{J + (1 - r2) (5U - S *) . (6,2?)
If Sy is independent of income, then the second term on the right side
of Equation (6.2?) drops out and option price is less than expected Sy. But
til the more likely case that Q is a normal good or has a positive price flexi-
bility of income, then 5* > Sy. Although Equation (6.27) must still hold, OP
could exceed r2SLJ.
The behavior of as a function of income is the basis of Cook and
Graham's [1977] measure of the irreplaceabifity of a good:
dJu = , _ 3V(Y,0)A>Y _ (6 28)
av(Y - S(J,Q)9Y
According to Equation (6.22), if Q is replaceable in this sense (dS^/dY = 0),
then supply .side option value must be negative. Smith (1984) has also used
the index of irreplaceabiiity or uniqueness in establishing bounds on demand
side option value. It is important to draw attention to the role played by the
*We are indebted to John Fitzgerald for suggesting this proof.
6-18
-------
Cook-Graham index in these analyses. It provides a gauge of the element in
a state-dependent model of consumer behavior under uncertainty that is impor-
tant for the outcomes and valuations of policy changes implied by the model.
That is, it provides a simple description of the extent of the difference in the
marginal utility of income at points that would be regarded as equivalent in
terms of their respective levels of the total utility. That is, the point de-
scribed by Q = Q and income at Y - Sy has the same total utility as the point
at which Q = 0 and income is Y. They are on the utility certainty locus.
However, a marginal change in income means something quite different in the
two states. It is this point that Arrow [1974] identified as the key element
in a state-dependent specification. With our restriction in the source of the
state dependency for the analysis of this chapter, this result then describes
how the importance of the state dependency is realized through the change in
a component (S^) of the ex post measure of the change in Q.
In Case D, all of the probabilities are positive. We have not been able to
find a general proof regarding the relationship between OP and (r2 - q2)S^.
However we have done sample numerical calculations with alternative utility
functions, parameters, and probabilities and have found examples to show that
OP - (r2 - q2)Sy can be either positive or negative. Some of these calcula-
tions are shown in Appendix A. These calculations seem to suggest that the
difference between OP and (r2 - QjjSjj maY relatively small; but this is not
a firm conclusion. The question requires further research.*
To summarize the results of this analysis, expected use value is an
ex post valuation measure. But if the indirect or von Neumann-lVSorgenstern
utility function is known, then it may be possible to derive analytical expres-
sions for the calculation of option price as a function, of expected Sy and the
parameters of V(•). Thus, if option price is the desired ex ante welfare indi-
cator, it may be possible to compute option price from the available ex post
indicators and assumptions concerning degree of risk aversion and so forth.
As we discussed in Chapter 4, option price is a measure of welfare change
that assumes a specific set of opportunities for adjusting to risk are available
*See Freeman {1984a] for some results of an investigation of the likely
magnitude of demand side or "timeless" option value.
6-19
-------
to the individual such that state-dependent payments are not possible, If
there are actuarially fair opportunities for insuring against risk either through
contingent claim markets or alternative payment plans, then option price is
an underestimate of the ex ante welfare measure. And, in any case, none of
these measures based on use data reflects any form of nonuse value.
6.4 CONCLUSIONS
In this section we draw out some of the implications of the preceding anal-
ysis for efforts to estimate the ecological and intrinsic benefits stemming from
hazardous waste regulation, in a world of uncertainty, individuals can be
placed in one of three categories with respect to their possible use of the eco-
logical resource. First, there are those who are certain to use the resource
if it is available. Second, there are those who are uncertain of their use of
the. resource. They are potential or possible users. And third, there are
those whose probability of using the resource is effectively zero--i .e. , they
are nonusers. The first and second categories of individuals can have both
use and existence values for the resource. The third category of individuals
can have only existence values. Of course, the boundary between the second
and third categories may be indistinct in practice. If we ask individuals to
identify themselves as either potential users or nonusers, some people with
low but nonzero probabilities of future use may identify themselves as non-
users. And statistical models for predicting probability of use may generate
trivially small but nonzero probabilities for many individuals. As a practical
matter, they should be treated as nonusers.
For the moment let us assume that the probability of the supply of the
resource is one. Use values for actual users (drawn from both the first and
second categories) can be estimated by existing indirect methods such as the
travel cost model. But these methods are incapable of shedding any light on
possible existence values.
One approach to estimating the total value for users is to ask them a
contingent valuation question about their total willingness to pay for the re-
source. If respondents understand that this value is to encompass both use
and existence values, then their answers are all that is needed for policy pur-
poses. However, to test hypotheses about the magnitude of and determinants
of existence value, it would be useful to have the total value broken down
6-20
-------
into its two components. Some researchers have simply asked peopie to allocate
their total willingness to pay into use and existence categories. One difficulty
with this approach is that it asks people to place themselves in a hypothetical
situation, which may be difficult for them to imagine. That is, it asks them
to imagine that they are nonusers without specifying for them the reason that
they no longer use the resource. A recommended principle in the design of
contingent valuation instruments is that questions should correspond as closely
as possible to respondents' actual situations.* Another approach is to compare
the contingent value responses with estimates of use values derived from indi-
rect techniques. In principle, the difference between the two measures is
existence 'value. However, in practice, at least part of the difference may be
due to measurement errors in either or both measures.
For the second category of users, one approach is to estimate expected
consumer surplus from data on actual users and to use assumptions about the
structure of demand uncertainty and preferences to compute option price.
But this gives an estimate of the increase in expected utility associated only
with use. Again, the only way to get at existence values is to ask a contin-
gent valuation question about total willingness to pay. And, finally, for non-
users, contingent valuation questions are the only basis for drawing inferences
f
about existence values.
In the case of uncertainty in supply and programs to increase the proba-
bility of availability, consumer surpluses of actual users may provide a basis
for estimating increases in expected use values. But, as in the case of only
demand uncertainty, contingent valuation questions are required to obtain total
values that include existence values.
~For a more complete evaluation of the reference operating conditions
that include this requirement, see the Cummings, Brookshire, and Schuize
[1984] definition.
4*
See Brookshire, Eubanks, and Randall [1983] for an example in which
certain nonusers in the sample were identified. Their responses were Inter-
preted as pure existence values.
6-21
-------
PART II
RESEARCH DESIGN, QUESTIONNAIRE
DEVELOPMENT, AND THE SURVEY
Part If of this draft interim report describes how we implemented the con-
ceptual framework developed in Part I through designing and completing a con-
tingent valuation survey to measure the benefits expected to accompany haz-
ardous waste regulations. Part II comprises the following four chapters:
Chapter 7 - Research Design: The Transition from Theory to
Practice
Chapter 8 - Survey Questionnaire Development
Chapter 9 - Sampling Plan and Survey Procedures
Chapter 10 - Profile: The Survey Area and Its Population
As suggested by their titles, the first three chapters describe the evolution
and development of the survey questionnaire, the experimental design, and
our survey administration and sampling procedures. The fast chapter in this
part briefly describes the survey area, the information on hazardous wastes
available to survey respondents, and the attitude and character of survey
respondents.
In the process of conducting and reporting on a fairly long, complex re-
search effort, the specific details of the tasks involved in the research, both
important and tangential, can obscure the reader's overall perception of the
research objectives. For this project it is important to remember that the pri-
mary objective was to value changes in the risk of exposure to hazardous
wastes. In particular, in contrast to the strategy adopted by the Cummings,
Brookshire, et al. [1983} study that sought to value regulations, our premise
is that the hazardous waste regulations provide a reduction in the risk of ex-
posure to these wastes. In effect, the regulations deliver a risk change--and
a change in a very specific kind of risk at that: the risk of exposure to haz-
11-1
-------
ardous waste. Thus, to measure the benefits of a regulatory policy, it is
necessary to value this change in risk. This implies that it is necessary to
know how individuals value changes in risk and to obtain empirical estimates
of those values. We have argued in the conceptual analysis that not ail risks
are the same. Therefore, it is also important to know how the empirical esti-
mates of these values are influenced by the specific features of the risk (e.g.,
the attributes of hazardous waste risks are likely to differ from occupational
risks), the circumstances of what is at risk, and the characteristics of the
individuals who are asked to envision themselves as experiencing the changes
in risk. These observations are not new. Indeed, the literature on people's
ability to process risk information--both from experts and ordinary individu-
als—suggests that all of these elements will be important to interpreting the
results of any effort to value risk changes.
The experimental economics and psychology literature on individuals' be-
havior under uncertainty provide valuable insights that influenced several
dimensions of our research design for valuing changes in hazardous waste risk.
For example, work by Schoemaker [1982], Hershey, Kunreuther, and Schoe-
maker [1982], Tversky and Kahneman [1981], and Slavic and Lichtenstein
[1984] points out the need to consider various features of the risk itself.
That is, hazardous waste risks may have certain attributes or characteristics
that will affect people's values for reductions in these risks. The importance
of the context of the risk also clearly emerges from this literature. Context
implies that the circumstances through which the individual experiences the
risk (whether real or hypothetical) may affect his valuation of a risk change.
One of its central elements in any description of risk is the implicit property
rights surrounding the risk change.
Some elements of the research also stem from another closely related set
of research — the recent findings of the state-of-the-art assessment of the con-
tingent valuation method (see Cummings, Brookshire, and Schulze [1984]).
Chapter 7 begins this section by discussing how the conceptual framework
influenced the structure of the questionnaire and its implementation in the sur-
vey design. ,
Of course, it is also important to note that the nature of the research
design was significantly influenced by the focus groups conducted in the early
II 2
-------
stages of developing methods for discussing hazardous waste and risk with
individuals. These activities are described in some detail in Chapter 8 (and
in greater detail In Desvousges et al. {1984a, b) ).
The sampling and survey design, described in Chapter 3, highlight the
target population, the sampling procedures used to obtain a representative
sample of the target population, and the survey procedures that implemented
the sampling, and in fact, the research design. Chief among these are the
detailed quality controls for the monitoring of interviewing process.
Chapter 10 provides a brief overview of the survey area, the target pop-
ulation, and how our respondents compare with that population. In addition,
it also includes a brief description of several hazardous waste contamination
incidents that have occurred in the survey area and types, amounts, and
sources of the information concerning them. Finally, the chapter profiles cer-
tain key features of the survey respondents including their knowledge and
perception of hazardous wastes.
II-3
-------
CHAPTER 7
RESEARCH DESIGN: THE TRANSITION FROM
THEORY TO PRACTICE
7.1 INTRODUCTION
This chapter describes the research design that underpins our contingent
valuation survey for measuring an individual's values for reductions in the risk
of exposure to hazardous wastes, (t has a difficult but important task because
it translates theoretical concepts and findings into their empirical counterparts.
The research design links the conceptual analysis, developed in Part I of this
report with the questionnaire development effort and the survey sampling and
administration procedures described in this part. Equally important, it also
provides some of the rationale for the analyses of the survey data that are
described in Part III. In essence, then, the research design explains the
reasons behind the structure of the empirical research and outlines in general
terms hypotheses to be tested in the empirical analysis.
With valuing changes in hazardous waste risk as Its focal point, our de-
sign tries to determine the most salient features of risk as a commodity. In
performing this task, the design considers the sources of value (for both use
and intrinsic values), the attributes or characteristics of risk, the assignment
of property rights, and the basic elements of the risk change itself—initial
values, endpoints, and outcomes at risk. To organize these efforts, the chap-
ter examines how the risk-related concepts affect the main objectives of our
research. It also draws on our conceptual analysis from Part I for most of
the guideposts of our organization.
The scope and complexity of concepts relating to valuing changes in risk
suggest that developing an effective research design will be difficult. For
example, the concepts ignore neatly drawn disciplinary boundaries by involving
changing mixtures of economic, psychological, and sociological phenomena that
researchers from each of the disciplines have considered. With this diversity
of disciplines, the research design presented in the chapter follows from our
7-1
-------
primary objectives of estimating the use arid intrinsic benefits associated with
reductions in the risk of exposure to hazardous wastes. However, it also tries
to blend together those elements from various disciplines that seemed most im-
portant for how people perceive and process the information on, and ultimately
value the changes in, the risk of exposure to hazardous wastes. The final
blend follows from our review of the literature, our experiences in the focus
group discussion sessions, and suggestions received from many outside review-
ers. Consequently, this chapter describes how each of the research issues
affects our objectives, considers their importance for valuing reductions in
the risk of exposure to hazardous wastes and for the comparison of different
approaches for valuing risk changes, and pinpoints how they are reflected in
the overall design.
7.2 GUIDE TO THE CHAPTER
Section 7.3 of this chapter provides an overview of the project leading
up to the development of the research design. Section 7.4 describes the types
of values—use and intrinsic—that are addressed in the research design. Sec-
tion 7.5 addresses the importance of different initial levels of risk and sizes
of risk reductions on individuals' values of reductions in risk. Section 7.6
provides the rationale for and treatment of the assignments of the property
rights of risk changes in the design. Section 7.7 highlights the two types of
risk included in the research: risks of exposure to hazardous wastes and
occupational risks. It also discusses risk attributes and their inclusion in the
research design. Section 7.8 considers the context of hazardous waste risks
and how it affects the research design. Section 7.9 describes risk outcomes
and endpoints. Section 7.10 examines three issues from the contingent valua-
tion literature that were important to the research design: the role of the
question formats used to elicit risk values, the information provided to re-
spondents, and the perceptions of the contingent commodity. Section 7.11
discusses the features of the design that allow for a comparison of its values
with those measured using indirect approaches for benefits measurement. Sec-
tion 7.12 explains the interconnections in the research design. Finally, Section
7.13 considers the implications of the various issues discussed in this chapter
for the research design.
7-2
-------
Figure 7-1. Overview of the origins of the research design
for valuing reductions in hazardous waste risks.
7-4
-------
ceptuai analysis in Part I identifies two primary sources of values for reduc-
tions in hazardous waste risks; reduction in risks of exposure to members of
a household and reductions in risks of exposure to the ecological system. Risk
reductions to household members are analogous to the traditional category of
use benefits. There is an important difference in perspective. They affect
the household's ability to attain satisfaction i_n expected value terms.
Reductions in ecological risks do not accrue directly to the household.
Rather, they affect components of the ecosystem such as the flora and fauna.
To the extent the household does not use the services of the affected compo-
nents of the ecosystem, we can expect that the household realizes only the
knowledge that these risks to the ecosystem are reduced. We have identified
these values as existence or intrinsic benefits, but, our understanding of the
motives for these values is limited. For example, Randall and Stoll [1933] sug-
gest that a form of altruism is the primary motive for intrinsic values, but
this view is far from a consensus. In addition, as discussed in Chapter 6,
ethical concerns may provide motives for these values. Presently, there is
not a strong a priori basis for identifying which motive, or set of motives, is
most important for these values. Nonetheless, the focus groups that were used
to help develop the questionnaire give some suggestive but informal informa-
tion. For example, the focus group participants, especially in church groups,
frequently used the term stewardship when describing their motives for "critter
values." (See Chapter 8. )
Our research design considers both types of values but attaches greater
weight to eliciting households' values for reducing their own risks of exposure
to hazardous waste. More attention is paid to these use values because they
are more central to our primary objective. Nevertheless^ this emphasis does
not reflect a judgment that intrinsic values are less important. Rather, it re-
flects our need to focus primarily on what can be addressed in the present
research with the information available on the motives for intrinsic values under
risk.
7.4.1 Measurement Concerns
The issues surrounding the measurement of individuals' valuations for
risk reductions are especially complex. Consequently, estimates for them are
likely to be the most controversial. For example, Kahneman's [1984] comments
7-5
-------
on the use of contingent valuation methods to estimate the value of some amen-
ity resources do not offer much encouragement for trying to measure these
values. His comments imply that these benefits may be value-laden with ideo-
logical overtones.* Specifically, he argues that, where there has not been
experience in purchasing a commodity (and, presumably, this experience could
be direct, as with market goods, or indirect, as would be the case with com-
modities requiring the individual to incur costs to experience the service),
the expression of preferences may lead to nonsensical estimates of value and
be a "symbolic demand." Under this view, the use of contingent valuation
for measuring the values of goods or services having no indirect basis for val-
uation would be questionable. He summarized his concerns by noting that
In particular, I question the existence of a coherent preference or-
der at the individual level which is waiting to be revealed by market
behavior. I am not sure that I have a "true" dollar value for the
trees that I can see out of my window. . . (Kahneman , 1984, p. 233]
Given this view—which might be regarded as an indirect implication of Cum-
mings, Brookshire, and Schulze's [1984] reference operating conditions — how
does one proceed to try to measure the values of risk changes, especially when
contingent valuation offers the only approach presently available? Our research
design addresses the measurement of these values in several very specific
ways. First, it recognizes that individuals may have different capacities to
envision the proposed risk changes and to value them. It is difficult to dis-
agree with the Kahneman position or even the position suggested by Freeman
[1984b] that people are being asked to perform very difficult tasks for which
they have little prior experience in that particular range of their preference
structure. The research design reflects this position by eliciting values for
two different risk changes from each individual. This allows the empirical
analysis to address the effects of differences among individuals in their ability
*The relationship between a change in utility and dollar measures of that
change have been controversial since Alfred Marshall first introduced the con-
cept of consumer surplus. For the most part, the literature on developing
these valuation measures has accepted a Hicksian framework and focused on
valuing price and quantity changes (see Morey [1984]). The development of
dollar measures for the utility changes associated with quality changes is not
as clearcut. For an illustration of these issues, see Desvousges and Smith
[1984].
7-6
-------
to research their preferences. For example, the models discussed in Chapter
13 recognize that values will differ depending on differences in income, educa-
tion, and knowledge of hazardous wastes across individuals. Moreover, they
attempt to adjust for the differential performance of the model itself in explain-
ing valuations across individuals.
The focus group and videotape sessions also played an important role in
designing a set of questions that attempted to reflect the Kahneman concerns
about measuring values for commodities that are not routinely a part of pur-
chase and consumption decisions. These sessions asked people about how they
thought about these values, their motives, and the sources of their values,
In effect, these sessions explored ways that might make it easier for people
to search these new areas of their preferences. Based on the focus group
sessions, it appeared that the more specific the situation in which the risks to
individuals (and to the ecosystem) was framed in the hypothetical questions,
the easier it was for people to appraise their valuations for these risks. This
finding is consistent with Wallsten and Budescu's [1983] evaluation of approach-
es to encoding probabilistic information from experts on particular phenomena.
They suggest that the analyst has to carefully specify the class of events in
question, the sources of information to be considered, and the causes of unre-
liability in the information. Thus, presurvey attempts to understand how
people formed their preferences substantially affected the research design for
eliciting and measuring the valuation of risk reductions,
7,4,2 Sequence Effects and Intrinsic Values
The conceptual framework developed in Part I of this report described
the rationale for measuring individuals' values for risk reductions in an ex ante
framework for both use and intrinsic values for changes in risk. Yet there
may be differences in how risks affect these values. That is, in describing a
risk change to a household, to elicit what is described as an ex ante use value,
the risk change must be experienced by the household. In contrast, the in-
trinsic values are associated with changes in risks to the ecosystem (and not
the household). The description and character of each is distinct. However,
in some ways, this is an easier separation to explain than with the services
associated with many other environmental policies. For example, to elicit the
existence value of a water quality change in a specific lake or river, circum-
7-7
-------
stances must be described to an individual to preclude use of the improved
resource by him or his household members. As we acknowledged in Chapter 6,
this hypothetical situation may be so implausible that it becomes completely
unrealistic. By contrast, with a risk change, it is possible that disposal prac-
tices might reduce the risks experienced by one group (e.g., households) but
not those of another (i.e., nonhuman species that constitute the ecosystem)
or vice versa.
Past research on the process of eliciting these values- suggests that the
order or sequence of the valuation requests in a questionnaire may influence
the authenticity of the values provided. Because of this plausible separation
in the mechanism delivering the risk reduction, it appears that the sequencing
issue may be less important for estimating different types of values for risk
reductions. Nonetheless, it is important to discuss the sequence used in the
research design and the potential relationship it might have for the valuation
estimates. Mitchell and Carson [1984], Randall, Hoehn, and Tolley [1981j,
and Cummings, Brookshire, and Schulze [1984] all have expressed concern
over the potential importance of the sequence in which a value is elicited.
For example, Randall, Hoehn, and Tolley [1981 ] found that the question se-
quence eliciting the value people placed on visibility improvements in the
Grand Canyon affected the value. Our research does not explicitly provide a
test for the effect of question sequences. Instead, the intrinsic value question
was asked after the use values, always as an incremental amount.
All of the questionnaire types in our survey maintained this incremental
format--!.e., eliciting intrinsic values as an additional amount. This is in con-
trast to the Mitchell and Carson [ 1984J procedure that elicited a "total" value
for water quality that reflected both use and intrinsic values. One reason
for our use of a different procedure was to avoid mixing the very different
characters of the two risks when they were presented to people. For example,
the events at risk are fundamentally different. The use value is a reduction
in risk of exposure (and potentially death, depending on the conditional risk)
to the household, while the intrinsic value is a reduction in risks only for
species in the natural environment. Our explanations also implied that the
character of the risks would differ in terms of their respective endpoints.
The endpoints for the household risks were always at some specific level of
7-8
-------
risk (or zero for those in the ranking version). The endpoint for species in
the ecosystem was the unspecified risk level that these creatures face in their
natural habitat,
The design does provide for two different sets of questions to precede
the intrinsic value question, depending on the question format used to elicit
the use values. For example, respondents receiving the direct question ver-
sion (discussed in more detail later in this chapter) were asked to reveal
amounts that they were willing to pay to obtain two successively lower levels
of risk that varied within the design. However, the direct question version
of the questionnaire did inform people m advance that the valuation exercise
would elicit the two changes in the household's risk of exposure and an addi-
tional amount for critters. This advance notification was one element used in
the structure of the questionnaire that attempted to reduce the potential se-
quencing effect in the direct question version.
In addition, the design elicited an intrinsic value from survey respond-
ents, who gave zero bids for reducing their own household risks. Therefore,
all respondents had the opportunity to express a value for intrinsic benefits.
This approach contrasts with that used in Desvousges, Smith, and McGivney
[1383], where values were elicited only from respondents who had given a pos-
itive dollar value to earlier use value questions. In the present survey, the
individuals also differ in the initial levels of risk described for the intrinsic
value question. Zero bidders, who chose not to "purchase" household risk re-
ductions, had higher initial risk levels posed in the critter question than the
people who purchased one or more reductions in household risk. Thus, the
present design allows for a somewhat fuller treatment of intrinsic values. The
importance of this alteration is an empirical question that is addressed in
Chapter 11.
Although the design used the same question to elicit the intrinsic value
questions for respondents who received contingent ranking version (also dis-
cussed later in this chapter)., the procedure differed from the direct question
version. The individual receiving the ranking version was asked to rank-order
four pairs of exposure risks and payments. Following the ranking, a contin-
gent valuation question was posed to elicit the willingness to pay to reduce
household risks to zero was posed. This process provides another initial level
7-9
-------
of risk depending on whether or not the person purchased the household risk
reduction. (The zero risk question was designed to address another issue
hypothesized to have an important effect on behavior under uncertainty—the
so-called "certainty effect," which is discussed later in this chapter.} in ad-
dition, in the ranking questionnaires, the individual was not toid that an in-
trinsic value would be elicited after the value for reducing the household risk
to zero. While this notification was inadvertently omitted and not a planned
part of the design, it may provide a basis for evaluating some aspects of the
sequencing problem. Thus, the research design placed the intrinsic value
questions in very specific positions depending on the version of the question-
naire administered. The amount of information provided to people and the ini-
tial risk level was designed to permit differences in the starting risk level
across versions of the questionnaire.
7.5 THE EFFECT OF RISK LEVELS AND CHANGES
This section highlights three dimensions of the risk information used in
the research design that has been found in past research to be important to
individuals' behavior under uncertainty: the level of the risk, the size of
the risk change, and the specific set of probabilities (i.e., exposure and con-
ditional ) leading to these outcomes.
7.5.1 Risk Levels
Past research, both theoretical and empirical, has suggested that the level
of the risk that confronts the individual when an increment (reduction or in-
crease) is proposed can affect his marginal valuation. One rationale for in-
cluding the level of risk as a feature of the design is that it was found to be
important to the marginal valuation of risk (see, for example, Chapter 5).
A second reason may explain differences in the valuations for risk reduc-
tions from very low initial levels. This explanation works in the opposite di-
rection to that used in most of the economics literature discussed in Chapter 5.
That is, people may have higher values for risk reductions starting at very
low levels because they may perceive that there is less technical knowledge or
experience about these risk levels. For example, if a disease is known to be
fatal in one out of every five cases, the individual may value a reduction in
risk quite differently than if the fatality of the disease was one out of a million
7-10
-------
cases. In the second case, because the disease is rare, the individual may
well regard the information provided on low-probability events as less accurate.
That is, the individual is perceiving a different second-order probability dis-
tribution for the risks of death in the first case, where death is a more fre-
quent occurrence, than, in the latter case. Thus, the increased marginal val-
ues for reductions may reflect the greater perceived second-order uncertainty
in the information on the risk.
The importance of assessing the effect of different initial levels of risk
for valuing changes in hazardous waste risks is heightened by the uncertain
nature of the technical information about the initial levels of risk. Three fea-
tures complicate the technical estimation of the risks from hazardous wastes.
One, research on exposure pathways, waste toxicity, and even the volume of
the waste is in fairly early stages. For example, the Conservation Foundation
[1384] and Office of Technology Assessment [1983] both point, out the need
for more and better technical information. Two, it seems possible that, even
witt better technical estimates of the risks from hazardous wastes, there will
be a substantial range of these risks depending on the characteristics of the
specific site. Sharefkin, Schechter, and Kneese [1984] stress the importance
of site specific features such as geohydrology. Three, differences in response
among receptors of exposure—e.g., people or ecosystems—are not well under-
stood. Thus, having a research design that allows for different initial levels
of risk is important not only from the perspective of consistency with the con-
ceptual analysis used to define valuation measures for risk changes, but also
from a very practical point of view that the situations in which regulations
for hazardous wastes are proposed may involve a rather wide range of risk
levels.
Allowing for varying initial levels of risks in the research design is also
important because these levels may affect how individuals process information
about risks. In effect, the kinds of thought processes that individuals can
bring to bear on a question involving risk will be important to the valuation
task that is central to contingent valuation. These thought processes may
differ for different levels of risks. Wallsten and Budescu [1983] suggest that
in its most general and far-reaching terms, the psychological considerations in-
volve factors that influence memory and how people use information. Fischhoff,
Slovic, and Lichtenstein [1980], Tversky and Kahneman [1974] and Kahneman
7-11
-------
and Tversky [1979] have documented the heuristics or judgment devices--e.g.
anchoring,* representativeness, and availability--that may bias an individual's
ability to judge situations involving risk. For example, if the individual felt
that the high initial values of risks in the research design were not "repre-
sentative" of hazardous wastes, then it may affect their ability, or willingness,
to search their preferences and provide an estimate of value. One way the
research design attempted to deal with the representativenesss heuristic was
in the questionnaire design. The interviewers asked respondents to consider
the hypothetical levels as if they were the actual levels but acknowledged that
even experts did not know for sure the exact size of the initial values •" A
second way was to use different initial levels for different individuals to try
to assess the potential effect within the design itself.
The availability heurisitic suggests that people may assess the probability
of an event by its familiarity. That is, the more information available to the
individual (e.g., newspaper or television articles), the more likely he may be
to "overestimate" the probability of an event occurring. The research design
allows for an examination of the relationship between availability and the initial
level of risk by asking in the questionnaire about the amount of information
that respondents had available. In addition, it asked respondents whether
they had attended town meetings about hazardous wastes.
All of the discussion of responses to risk levels, whether reflecting indi-
viduals' perceptions of the quality of the information provided or based on
the heuristics suggested by some psychologists as the means used to process
*The anchoring heuristic in which people's values might be affected by
implied starting values or anchors especially is important for the format ef the
valuation question and is discussed in detail later in this chapter,
|The potential role of information on values is ubiquitous. For example,
Cummings, Brookshire, and Schulze [1984] suggest that it maybe hard to pin
down just exactly the effect of information. This seems to be the case with
the availability heuristic. If individuals hear or read about heart disease or
cancer or car accidents and then believe that they are more prevalent because
of this information, then this seems consistent with the heuristic bias. How-
ever, if these same individuals read and or hear--and retain — factual informa-
tion about the incidence of severity of causes of death, then it seems the bias
does not exist. Thus, considerable caution will be required in trying to ceter-
mine any relationships between the initial level of risk and the avahaoility
heuristic.
7-12
-------
risk information, are important because they are alternative descriptions of
the risk perception process, In short, when the questionnaire presents indi-
viduals with a risk level, do they believe it? How do they interpret it or
adjust, the value they are given when formulating a response to a proposed
risk change? Clearly, the valuation responses will be related to what each
individual perceives his risk level will be under the circumstances described
in the contingent valuation questions. Control of the magnitude of the risk
level and of the postulated change in risk that are posed tQ respondents does
not in itself ensure the analyst will have control over the respondents' per-
ceived risk level and the changes in it. This is the reason for attempting to
understand the risk perception process and how the character of our contin-
gent valuation questions would be interpreted within it.
7.5,2 The Size of the Risk Change and the Rote of the Conditional Risk
Our discussion up to this point has focused exclusively on the importance
of the initial levels of the exposure risk in our research, design. However,
there are two other closely related elements that are also addressed in the re-
search design: the size of the change in risk and the role of the conditional
probability of death given exposure to hazardous wastes. In the design, the
changes in the levels of exposure risks were held constant—in percentages
terms—across the varying initial levels of risk. For example, the percentage
change in the initial exposure risk level (e.g., A) to the intermediate risk
level (B) was the same in the four vectors of the design that relate to the
levels of risk. However, the percentage change from the intermediate level
(B) to the final level (C) was held constant at a different percentage change.
In effect, each individual values two distinct risk changes—from Level A to
Level B and from Level B to Level C. If we assume the values from the two
different levels can be grouped together for statistical analysis, then it is pos-
sible to evaluate differences in individuals' understanding of the contingent
valuation exercise in our marginal valuation models of risk changes (see Chap-
ter 13), Finally, because of the findings of Kahneman and Tversky [1379]
and others that individuals may respond more easily to percentage changes,
the increments were held constant in percentage terms rather than using con-
stant numerical increments.
7-13
-------
The role of the conditional probability of a health effect (usually death)
is the last numerical feature of risk addressed in the research design. Our
approach to presenting probabilistic information about hazardous waste risks
involved splitting the risk information into three risk circles that related to
exposure, dying if exposed, and the combined risk of exposure and death.
This format was derived as a direct result of the focus group sessions. Since
it is important to our presentation of information on risk for the valuation task,
the specific details behind its development are discussed in detail in Chapter 8.
To examine the potential importance of the size of the conditional probability,
which was assumed not to be affected by the hypothetical regulations in the
scenario, the design allows for a full factorial design for three groups of expo-
sure risks and two levels of conditional risks and an additional (1 x 2) design
using lower exposure risk probabilities and two conditional probabilities.
These lower probabilities were one-tenth the size of the other design points.
In summary, the specific dimensions of the risk information — the size of
the initial level of risk, the change in risk, and the conditional risks—are
treated tn an experimental design. The specific features of the experimental
design are explained in Section 7,9.
7.6 PROPERTY RIGHTS AND RISK VALUATION
An important dimension of the design is the examination of the influence
of property rights on individuals' values for changes in risk. Property rights
involve the set of legal entitlements, either implied or expressed, to a particu-
lar good or service. Mitchell and Carson [1984] stress the importance of prop-
erty rights In a contingent valuation survey. Even for a fairly welt under-
stood public good like water quality, they find that the property rights can
have an influence on valuation responses. In the case of hazardous wastes,
where we have assumed the property right applies to the household having
"the right" to some level of exposure to hazardous wastes, the issues are even
more complex. This research did not attempt to deal with all of the issues
that can be involved. Rather, we have offered a few reasons for their poten-
tial importance and then incorporated one simple means for considering their
implications in the research design.
The importance of property rights for valuing changes in hazardous waste
risks derives from three sources: their role in the economics literature, their
7-14
-------
role in the psychology literature, and their prominence in the focus group ses-
sions, In the economics literature, particularly in the contingent valuation
subset, property rights discussions have focused on the wilfingness-to-pay/
willingness-to-accept issues. The literature has numerous examples of the dif-
ficulties of asking willingness-to-accept questions, with Knetsch and Sinden
[1984] as the most recent example. In our research design for valuing haz-
ardous waste risk changes, we have followed the recommendation of the Cum-
mings, Brookshire, and Schulze [1984] reference operating conditions and used
the willingness-to-pay format. Nonetheless, a willingness-to-accept approach
is used in a different context within our research design: in eliciting the wage
increment necessary for accepting higher risks from a new job. The rationable
for using it in this context is that the acceptance structure was more plausible
than the payment structure for this problem.
Property rights issues also appear to have been important within the psy-
chology literature. For example, Kahneman and Tversky (1979] have argued
that, contrary to the expected utility hypothesis, people have very different
preferences for gains relative to losses. One interpretation of their arguments
is that the gain versus loss phenomena may be a reflection of differences in
the property rights that individuals perceive. Hershey, Kunreuther, and
Schoemaker [1982] also discuss property rights in their evaluation of the im-
plications of the assignment of risk for experimental evaluations of the expected
utility framework. If respondents feel that they have some existing low level
of exposure risk to hazardous wastes and are now faced with a possible in-
crease in the risk (e.g., due to the siting of a hazardous waste landfill or a
commercial waste processing facility), they might feet that a property right--
the lower risk level—would be taken away from them. Their value for the
risk change could be markedly affected by the implicit assignment of these
rights.
The focus groups provided another reason for considering property rights
within the design. Participants in these sessions frequently expressed views
that were equivalent to a suggestion that how the property rights were handled
in the hypothetical situations influenced their responses. For example, their
comments and reactions differed depending on types of government actions in-
volving risk. To organize these discussions, we deliberately chose actions
7-15
-------
that implied one assignment of property rights and then asked whether an
alternative assignment made a difference. As a consequence of our experiences
with these groups we developed ways to include property rights in the design.
In the design, one examination of "property rights" effects is accomm-
odated by comparing individuals' valuations of a given risk change for a reduc-
tion in a specified level of risk with an equal increase. That is, the proposed
and starting endpoints are simply reversed for the two changes in risk so that
it is only the assignment of rights (and with it the direction of change in risk)
that is different across the questions. Since payment for the risk reduction
yields the endpoint and avoids it for the increase, the actual endpoints are
the same.
The design also included a second feature in the property right issue.
The focus group research suggested that individuals responded differently to
the property rights issue depending on how they perceived the action wds tak-
ing place. In effect, was it imposed on them or was the case described as if
there had been an opportunity to affect the decision? This issue was -eflected
in our research design as a component of the hypothetical scenario for avoiding
the risk increase. It was also varied independently from the changes in the
risks across design points so that it would be possible to evaluate the implica-
tions of the degree of control available to individuals when changes in rights
were taking place. To accomplish this task, the sample was divided in half
with one group having the risk increase scenario that indicated the town coun-
cil had voted to approve the change, while the other was told that the Federal
government had decided to allow the change.
7.7 TYPES OF RISKS AND RISK ATTRIBUTES
This section addresses the influence of different types of risks on indiv-
iduals' values for reductions in risk. Recent research--e. g., see Schoemaker
[1982]; Hershey, Kunreuther, and Schoemaker [1982]; and Slovic [1984]--has
stressed the importance of the different types of risk in influencing individ-
uals' perceptions and their values of risk changes. One way of attempting to
formalize the modeling of reasons for differences in individuals' responses to
different types of risk is to assume that risks have attributes. Therefore, to
understand the differences in responses to these risks, we must model how
these attributes of risk affect individual utility and, in turn, their behavior
7-16
-------
in the presence of risk. The exact boundary between types of risk and risk
attributes is poorly marked. The main distinction that we draw is that the
types of risks may embody more than one attribute. This section considers
discussions of risk types and then the use of a framework that tries to identify
a set of attributes of risk which would describe the types separately. It then
describes how both sets of research have influenced the research design.
7.7.1 Types of Risks
Two of the most important of the types of risks influencing an individual's
willingness-to-pay response for changing exposure to hazardous waste risks are
the "dread risk" and the "known risk" (Slovic [1934]). The first involves the
notion that an event is dreaded because it is potentially catastrophic, involving
many people. The second type relates to both perceptions about the individ-
ual's knowledge of the risk and whether the events at risk are delayed in time,
The research of Slovic and his associates suggests that these factors influence
how individuals respond to uncertainty. Other recent work by Von Winterfeldt
and Edwards (1984] also stresses the importance of types of risks for assessing
policy conflicts over technologies. For our analysis, this would imply that
individuals may value incremental reductions in the risk of death from different
sources quite differently. That is, individuals may value changes, in hazardous
waste risks quite differently than an equivalent risk change for another type
of risk where it is known and not dreaded.
The character of the hazardous waste risks has other important implica-
tions for interpreting the values of the posed reductions in risk. For example,
the hypothetical situation states that the outcome (i.e., premature death) of
the exposure risk and the corresponding conditional risk will not be known to
the household for 30 years. This long time horizon, although probably consist-
ent with at least some hazardous wastes, may substantially affect how people
process the information about the risks. For example, Bjorkman [1984] sug-
gests that people make riskier decisions the further in future their conse-
quences are experienced. In addition, the time dimension may affect the im-
portance of the event itself. Lundberg et at. [1975] have found that events
10 years in the future are considered one-third as important as present events.
While their research did not relate explicitly to a risk of death, their general
implication seems relevant. Finally, Svensen [1984] has shown that the time
7-17
-------
character of the risks can affect people's perceptions of the risks. He found
that people over estimated short-term high risks in relation to long-term small
risks. However, he also notes that his exposure interval never exceeded 1
year which differs from our situation.
7.7.2 Risk Attributes
As noted earlier, in attempting to understand why individuals respond
differently to different risks, researchers have stressed the.importance o" par-
ticular characteristics or attributes that a risk embodies. The ability to con-
trol risks and the extent to which risks are voluntary are two of the attr.butes
most frequently identified as important. In addition, the focus group partici-
pants frequently mentioned these attributes as important to their perceptions
of hazardous waste risks. In particular, they suggested that the extent to
which they had a say in a decision involving risk significantly affected how
they felt about the risk.*
Raiffa, Schwartz, and Weinstein [1977] suggest identifiability as an im-
portant attribute of risk. Identifiability is the extent to which individual lives
are associated with decisions involving risk. They further differentiate between
ex ante identifiability--individuals' identities are known prior to the decision--
and ex post identifiability--individuals' deaths can be attributed only after
the decision. For example, they suggest that decisions involving risks faced
by trapped coal miners are identifiable both ex ante and ex post. On the
other hand, the individual workers who die from exposure to asbestos or vinyl
chloride can be identified only after the fact. Individuals, and collectively
society, have a higher willingness to pay for a change in risk the larger the
extent to which the risk is identifiable.
Our conceptual framework suggests that identifiability may not be a risk
attribute. Instead, it is a reflection of the difference between ex ante and
ex post analytical perspectives. That is, identifiability pertains to value; when
the outcome at risk is known. To draw from their example, it is no ionger
*This is clearly consistent with findings observed in studies of the siting
of nuclear facilties. See for example Carnes et al. [1982], Games and Copen-
haver [1983], and Carnes et al. [1983]. It is also consistent with the program
of research recently described by Kunreuther and Kleindorfer [1984].
7-18
-------
the risk of the coal miners being trapped underground that is valued but the
outcome of the risk. This ex post perspective is inconsistent for valuing wel-
fare changes from regulatory policies for reducing hazardous waste risks be-
cause the policy decisions are made prior to the outcome being known,
Weinstein and Quinn [1982] suggest that anxiety may be an important
attribute of risk that could Influence people's willingness to pay for reductions
in risk. Anxiety causes people to have disutility from experiencing the risk.
Weinstein and Quinn cite evidence to suggest that people* may be willing to
pay for risky diagnostic tests even when their overall prospects for survival
are poor. They suggest that the additional expenditures may enable people
to make better plans for either their death or survival. The focus group par-
ticipants indicated some consideration of anxiety as an attribute of hazardous
wastes in developing their valuation responses. Some suggested that the anxi-
ety stemmed from the highly uncertain state of information about the effects--
and extent—of exposure to hazardous wastes. Finally, some participants men-
tioned the possible anxiety from the potentially long latency periods that were
discussed earlier. Clearly, anxiety and the other attributes of hazardous
waste risks will be important for interpreting research findings.
7.7.3 The Rote of Differences in the Types of Risk
for the Research Design
To develop some information relevant to the potential effects of risk attri-
butes and types of risk, the research design elicits values from individuals
for changes in two types of risk—exposures to hazardous wastes and fatal acci-
dents on the job. However, we do not have complete information on the valu-
ation of both types of risk for all respondents. The job risk questions were
asked only of those respondents who were working for pay — either on a full-
time or part-time basis—at the time of the survey. Also noted earlier, the
job risk valuations are posed in terms of the wage premium needed to accept
the higher risks rather than willingness to pay. The job risks also were elic-
ited using a different vehicle to express the risk change. Employed individuals
were asked to place their perceived risk of dying from an accident on the job
this year on a risk ladder (see Chapter 8). The questions then posed 50 per-
cent and 100 percent increments in risk and elicit the wage change need to
accept new jobs with these higher levels of risk. The reason for using the
7-19
-------
risk ladder for the job valuations is that this vehicle provided the easiest
means of dealing with the differences expected in individuals' perceptions of
their actual risks on the job. Nonetheless, any direct comparison, of the values
from the two different types of risk will be difficult and is beyond the scope
of the research associated with Phase I of this project. Nevertheless, the
value for changes in occupational risks elicited in the questionnaire can be
compared with estimates from hedonic wage models (see Viscusi [1984] and
Smith [1983]) as a rough gauge of the plausibility of the sampled individuals'
responses,*
7.8 CONTEXT OF RISK
The context of a change in risk is another important element to consider
when interpreting elicited values. The exact definition of context is difficult
to pin down because different researchers, often from different disciplines,
have used the term differently. For example, Mitchell and Carson [1984] dis-
cuss context as a type of misspecification bias in contingent valuation. In
their terminology, context includes not only the setting of the contingent valu-
ation interview but also what might be termed the mental setting created by
the material in the questionnaire itself. On the other hand, Schoemaker [1980]
uses the term to refer to what happens when respondents evaluate exactly the
same information differently when it is in a different context. For example,
his research showed respondents evaluating the same gambles differently in
the context of a lottery rather than insurance.
Not only is context a difficult concept to define, but it is also difficult
to distinguish from some of what we and others have designated as the attri-
butes of the risk itself. For example, a context effect might occur because
the way a risk is presented may imply—at least implicitly—a different set of
attributes. Because several previous sections have described the general char-
*The "property rights" effects can also be examined in part through the
job risk questions. However, in this case, the questions use the individual's
existing job as the basis for describing the risk changes. Thus, the level of
this risk was not controlled as part of the experimental design. Moreover,
since the sample was designed to be a representative sample of households in
suburban Boston (with oversampling of Acton), there are good reasons to ex-
pect that it will not provide a representative sample of the occupation related
risks experienced by individuals.
7-20
-------
acter of hazardous waste risks, there will be no further discussion of this po-
tential dimension of context.
Context effects also may imply that how a risk is explained to an indi-
vidual may influence his response to it (see Schoemaker [1982], pp. 547-48,
for further discussion). Instead of suggesting some form of irrational behav-
ior, context effects also might suggest that analysts have done a poor job of
communicating their questions to survey respondents. Consequently, one of
the most important aspects of this research design has been the use of focus
groups in the development of the wording of the questionnaire, the format of
the vehicles used to explain risk, and in the pretesting and revision of the
questionnaire. Since the specific steps in this process are described in Chap-
ter 8, here we simply acknowledge their role in adjusting the structure of the
empirical component of this research to reflect what has been learned from the
varied sources of research on decisionmaking under uncertainty. The relevant
sections of the questionnaire that describe the mental setting view of context
are highlighted in Chapter 11.
7.9 RISK OUTCOMES AND ENDPOINTS
Two other considerations of risk are important in our research design:
the events or outcomes at risk and the use of certainty as an endpoint. Our
discussion of hazardous waste risks has focused almost exclusively on mortality
as a potential consequence of exposure to hazardous wastes. This limitation
was due primarily to deciding what was feasible to consider in one research
effort. However, although it does not imply that morbidity effects are unim-
portant, the almost exclusive use of death as the health outcome has important
implications. In their discussion of behavior under uncertainty, Weinstein and
Quinn [1982] suggest that a risk situation (or gamble in their terms) that in-
cludes death can affect how people consider the situation. Not only might
death be important, but how one dies—the quality of the death--may also be
important.
The research design addresses death as an outcome in several ways.
One, in the process of eliciting individuals' values for reductions in the risk
of exposure to hazardous wastes, no specific cause of death is mentioned.
People are then asked if they had a cause of death in mind when giving their
values to provide some information on whether the perceived cause of death
7-21
-------
influenced their bids. In addition, after the individuals' values for reductions
in the risk of exposure to hazardous wastes are elicited, two variations were
used to change the information about possible health consequences. Specific-
ally, a change in the health end state (e.g., whether the cause of death was
due to damage to the body's immune system, or whether the risk was associated
with birth defects severe enough to cause lifetime mental or physical handicaps)
was posed to the respondents to see if these would alter their value of the
risk change.
The second aspect considered in the research design is the effect of cer-
tainty as a risk endpoint on individuals' values for reduction in risk. Tversky
and Kahneman [1981] suggest that people will value a protective action that
reduces the probability of a harm from 1 percent to zero more highly than an
action that reduces the probability of the same harm from 2 percent to 1 pei—
cent. They attribute this phenomenon to the shape of their value function.
Our research design addresses the certainty effect by eliciting values
for reducing the risk of exposure to hazardous wastes to zero. It is important
to note that these values were elicited only from a subset of our sample--those
respondents who received the contingent ranking version. In effect, they
had completed a task in which they ranked different combinations of monthly
payments and exposure risk levels prior to answering the certainty question
based on the same hypothetical situation used in the direct question format.
However, the certainty question posed a different hypothetical situation and
then used a direct question to elicit their value for reducing hazardous waste
exposure risks to zero.
In summary, the research plan addresses two dimensions of risk context--
health outcomes of the risk and the certainty effect. In the former case, indi-
viduals are asked if they want to revise their previous bid in response to dif-
ferent outcomes. In the latter case, values are elicited from a subset of the
sample for reducing the risk of exposure to zero.
7.10 CONTINGENT VALUATION AND ELICITING VALUES OF RISKS
An important set of issues considered in our research design stems from
the literature on contingent valuation. Rather than exhaustively evaluating
these issues, this section considers the three that are most relevant to our
7-2.2
-------
design: the question format used to elicit the values for reduced risk (i.e.,
our contingent commodity), the treatment of perceptions of the contingent com-
modity, and the role of information. Other aspects of the contingent valuation
fiterature--e. g. , Mitchell and Carson's [1984] context bias and related con-
cerns—are discussed prior to the empirical results in Chapter 11.
7.10.1 Question Format
One of the key features of the research design is the use of two different
formats--contingent ranking and direct question—to elicit values for reductions
in hazardous waste risks. In the direct question format, the interviewer di-
rectly asks the respondents to give his maximum valuations of the risk change.
These are option prices. By contrast, the contingent ranking format requires
that the respondent rank a set of cards showing alternative combinations of
payment amounts and risk levels. The alternatives are structured to prevent
one choice from dominating and to require tradeoffs between increased payments
and lower risks.
The importance of the influence of question format on valuations of risk
reductions stems from several sources. Desvousges, Smith, and Fisher [1984]
found that willingness-to-pay amounts are influenced by the format used to
elicit values in contingent valuation. While this research focused on bidding
games and payment card alternatives compared to the direct question format,
it does suggest the possible influence of question format. In related research,
Desvousges, Smith, and McGivney [ 1983J and Rae [1381a,b] found the contin-
gent ranking format to be a promising alternative, but their findings were not
conclusive. For example, in none of the evaluations had contingent ranking
been composed on a completely independent basis. In previous applications,
both the direct question, or some other alternative question format, has been
administered to the same respondent along with contingent ranking. By allow-
ing the contingent ran king format to be independently administered, our design
is capable of addressing this issue. It is important to note that the independ-
ence of the ranking format refers only to the format being used to elicit wil-
lingness to pay. The ranking versions were completely consistent with respect
to risk levels used in developing the alternatives to be ranked and the other
key elements of the research design that are discussed later in this chapter.
7-23
-------
There is a final rationale for using the contingent ranking format that
draws from the psychological literature. Fischhoff and Cox [1984] have noted
that the ordinal information processing task, like the one required in contin-
gent ranking, is an easier one for respondents to perform. He has noted this
advantage as especially important for tasks involving probabilistic information
or the type of value information required in contingent valuation. In the case
of our research, both of these elements are present, making a strong case for
including the ranking format. In effect, contingent ranking' requires respond-
ents to perform only an ordinal task but, in the analysis stage, with explicit
assumptions concerning the natui e of individuals' preferences, can yield esti-
mates of individuals' valuations for reductions in the risks of exposure to haz-
ardous wastes.
Some explanation is also necessary for our decision to use the direct ques-
tion format. One primary consideration in using this format is that it mini-
mizes the chance of the respondent's "anchoring" on some artificial reference
point in the interview, a possibility noted by Tversky and Kahneman [1981]
in their analysis of individuals' decisions under uncertainty. For example,
the starting point used in the bidding game format provides people with e> actly
such an anchor. It suggests to people a frame of reference for making their
decision. For example, is the interviewer expecting a value of $20 or S20G?
Recent evaluations by Desvousges, Smith, and Fisher [1984], • Mitchel I and
Carson [1984], and Boyle and Bishop (1984] all point to this troubling aspect
of the iterative bidding format for contingent valuation questions.
Table 7-1 provides a summary of the available results on the existence
and extent of starting point bias. Despite the promise of bidding games in
the earlier work by Thayer [1981 ] and Brookshire, Randall, and Stoll [1980],
recent studies by MitchelI and Carson [1984] and Boyle and Bishop [ 1984 ] have
provided strong evidence of starting point bias. Boyle and Bishop's results,
based on a sample of 176 recreationists, are probably the most telling evidence
to date. Indeed, they are led to conclude that bidding games may not be
worth the increased complexity. This conclusion also is supported bv Cum-
mings, Brookshire, and Schulze [1984].
In our view, iterative bidding does result in substantially higher
bids. . . . Mitchell and Carson as well as Bishop and Heberlein
are obviously correct in pointing to the lack of evidence that would
7-24
-------
TABLE 7-1. STARTING POINT BIAS: THE RESULTS
Study
Contingenl
commodity
Starling
points
used
Type
Of test
Sampling
procedure
Sample *i2t
Conclusion
RtmirHs
Rows, d'Arge, and
Brookshire [1380]
Visibility in Four
Corners
$1, $5, $10 Regression
Random sampling
of household in
f armingtori, NM,
ana Navajo Recre-
ation Area
a) 93 respondents
bidding in 3
scenarios
b) 3' respondents
bidding in 3
scenarios
Starting point bias
evident in regres-
sion of equivalent
surplus bids. Not
evident in smaller
sample of compen-
sating surplus (CS)
bids
Sampling and survey procedures
were not standard; ambitious
questionnaire also Hied to
address several other bases;
authors noted order of magni-
tude different;; between C3
bids and starting points. Small
sample sue «tl>o limited efft-. -
tiweness in CS case.
Brookshire, d'Arge, Visibility in $1,110,
Schulii, and Th»y»r Los A q»t . )S0
[1979]
Paired census
tracts in Los
Angeles area
12 communities with
sample Mfn*., rang-
ing from 2 to 16
Reject noil hypothe-
sis of no starting
point bias in 6 of
36 means tests;
tail to reject in 30
of 96
Small samplt sizes limit power
of statistical tests. No adjust-
ments made for other issues
tested in survey design.
KJ
en
Thayer 119811
Mitchell and Carson
11983),. interpreta-
tion of Greenley,
Walsh, and Young
(19831
Substitution of a
recreation site
Option price of
water quality in
Platte River Basin
*1,*10
Means and
regression
Regression
Random interviews
with recreationists
in Jemez Mountains
106
Random sampling of 161 (water bill)
households in 177 (sewer tax)
Denver and fort
Collins
No difference be-
tween aver age bids
at level of s*g-
nificance; nonsig-
nificant coefficient
for suiting bid
Mitchell and Carson
show different
implied starling
values by the
alternative pay-
ment vehicles
Well-defined commodity familiar
to respondents--somewhat iirtvl-
led range of starting values--
larger sample sues than in
many previous Mudi^s
Some disagreement about exact
commodity measured -see Chap-
ter 5 of Bfsvousges, Smith, and
MtGivney 119831 and Mitchell
and Carson [1344)
Boyle fix# Bishop
[1384)*
Brookshire et at.
119001
Scenic beauty on
lower Wisconsin
River
The right to hunt
eik for one annua!
season at various
levels of hunting
amenities, e.g.,
terrain and fre-
quency of encoun-
ter with elk
$10 to
$120
Randomly
chosen
$25,$75,
$200
Regression
Regression
and test of
bids
Rjmdom sampling o?
recreationists onsite
176
Unspecified
108 licensed elk
hunters
Found statistically
significant and
positive relationship
between starting
bid and willingness
to pay
Authors reject hy-
pothesis that start-
ing points Influenced
final values at Ihe *
0.05 level of sig-
nificance
Commodity Is somewhat abstract;
detailed examination of starting
points with ample sample size
and wide range of starting
values
Utility bill and hunting license
fee used $5 payment vehicles;
hypothesis that final bids af-
fected by payment vehicle re-
jected at .0! level.
Desvousges, Smith,
and McGivney
11983!
Option price for
water quality and
improvement
$25,$125 Regression
Stratified random
sample of house-
holds in 5 county
area of Monongahela
River basin
150
Some evidence of
starling paint bias
especially in com-
parison results;
high starting point
corresponded with
19 of 30 ouilymg bids
making statistical
results suggestive
but not com iusi ve
Most detailed sampling and sur-
vey plan; trained professional
interviewers; ample sample size
and wide range of starting bias
Boyle And Bishop cited results of other related research (Boyte, Bishop, and Welsh | forthcoming j that also suggested starling point problems,
unavailable to the authors at the time of this report
TIns study was
-------
support the attribution of such effects to the preference research
process,. . , . Moreover, we must acknowledge. . .that the par-
allel between the iterative bidding process and the iterative valua-
tion trials used in laboratory experiments. . . is without obvious sub-
stance. . . . Thus, all that can be said at this point in time Is
that iterative bidding rather consistently results in higher CVM val-
uations, but we are unable to explain such differences, [pp. 267-
268].
However, this conclusion may be too pessimistic with respect to our under-
standing of the processes involved with iterative bidding questions in contin-
gent valuation. One study that provides the basis for their conclusion that
iterative bidding leads to higher willingness-to-pay responses is their experi-
ence reported in Cummings, Brookshire, et al. [1983]. In these experiments,
the respondent first provided a value using a payment card and then was
"iterated" toward a maximum valuing by informing him that the commodity would
not be provided based on their first bid. This process is not an iteration
toward a maximum value but, instead, is a value provided under different con-
ditions of provision. That is, they have changed the terms of exchange in
the market. Rather than obtaining a maximum bid, it is hard to interpret the
exact nature of their final value. Mitchell and Carson [1984] used an analogous
procedure in their survey but are reluctant to interpret this bid as a maximum
bid because of the circumstances under which it was elicited.
In addition, the influence of starting points need not always be in a posi-
tive direction. For example, Figure 7-2 shows the distribution of bids from
two bidding games conducted in Desvousges, Smith, and McGivney [1983]. I n
the case of the $125, the iterations primarily are downward but the $25 starting
points have a substantial number of upward iterations. In this case it is un-
clear that bidding games lead to an upward bias. As noted by Mitchell and
Carson [1984], there are a substantial number of bids that are "anchored" at
the starting value, which is consistent with the Tversky and Kahneman [ 1981 ]
position. Also noteworthy are the large number of zeros with the $125 starting
point. Mitchell and Carson (1984] suggest that these respondents also may
have been affected by the "too-high" starting bid. As we indicated in Des-
vousges , Smith, and McGivney [1983], this bidding game also had 19 out of
32 respondents that we determined as outlying bids based on our regression
diagnostics.
7-26
-------
Iterative Biddinq
Framework—S25
Starting Point |
20
10
Ji
:l
' i i . » > i ¦ i
Iterative Bidding
Framework—$125
Starting Point
30
20
£ 10
25
3 4 3 ,,3
1 a >| ¦£»£.
JLl, Li -Li at, ill..
1 1 1
' * « ¦ I «-
_i_lj
Direct Question
Framework
Direct Question
Framework—
Paymctit Card
»
£ 10
14
li. i
u.
29
10
ii
i- > t X..I
-i—j—j—j-i i_i i_L
0 10 20 30 40 50 60 70 80 90 100 110 120130 140 150160
S Bid*
Figure 7-2. Effects of instrument—distribution of option price for a change
in water quality from boatable to fishabie, protest bids excluded.
7-27
-------
in summary, the iterative bidding format has little to recommend it at
this stage. Not only does this format seem to experience problems with start-
ing point bias, it may also increase the likelihood of a rejection of the terms
of the contingent market. As Mitchell and Carson [1984] suggest, the process
of iteration upwards and downwards in these games may also be even more
complicated than that assumed by the empirical models.
The other two formats considered were the anchored and unachored pay-
ment card. The anchored payment card, developed by Mitchell and Carson
[1981,1984], gives respondents a card with dollar amounts and anchors at vari-
ous amounts for other public goods like national defense and fire protection.
Despite Mitchell and Carson's [1984] experience, this format was not used be-
cause of the concern over a respondent relying exclusively on the anchors in
determining their valuation responses.
To illustrate what appears to be potential "anchoring" with the use of
payment cards, Table 7-2 provides summary statistics from Mitchell and Car-
son's [1984] contingent valuation survey to measure the benefits of national
water quality improvements that used a payment card with reference amounts.*
Some individuals seem to have relied exclusively on the amounts provided on
the card in forming their valuation responses. This seems especially the case
for the lower income groups with the large majority of the individuals select-
ing amounts from the card and relatively few giving a response not shown on
the card. However, it is not possible to conclude that this information clearly
implies anchoring has been a problem with their approach, since the amounts
on the card are also commonly used bids, such as $50 or $100 a year.
Another potential problem with the Mitchell-Carson payment card is indi-
viduals keying on the reference or anchor amount on their card. The third
column in Table 7-2 shows the number and percentage of bidders who gave a
bid within plus or minus one increment from the reference amounts on the
card. While the data do not suggest that this is a serious problem in Mitchell
and Carson's study, it does seem to indicate that it may be occurring to some
degree. For example, 160 out of 452 nonzero bids, or about 35 percent, were
*These comparisons are only possible because of the detailed information
provided in all of the Mitchell-Carson survey reports.
7-28
-------
T/5 BLE 7-2. WILLINGNESS TO PAY FOR BOATABLE WATER QUALITY3
Income
Number of respondents
choosing value on card
Total number
of respondents
Number of nonzero
respondents with ± 1
increment of anchor
Total nonzero
respondents
Less than 10,000
123
(98)
125
37
(41)
90
10,000 - 19,999
144
(94)
154
54
(42)
130
20,000 - 29,999
113
(87)
130
39
(35)
111
30,000 - 49,999
43
(44)
97
21
(25)
84
50,000 +
30
(73)
41
9
(24)
37
Total
453
(83)
547
160
(35)
452
Source: Mitchell and Carson (1984 J, Table 4.
CD
3
Numbers in parentheses are percentages.
-------
within pius or minus one increment of the anchors on their payment card.
Again the occurrence is more frequent at the two lower income levels, which
had 91 out of 220 nonzero bids or 41 percent.
Based on two pretests of their questionnaire, Mitchell and Carson [1384]
did not find any systematic bias from the anchored payment card. However,
they acknowledge that the sample was relatively small in one pretest and that
the range of the test--anchors that differed by 25 percent--may have been
two narrow. By contrast, Boyle and Bishop [1984] experienced mixed per-
formance when using it. Consequently, it seems prudent to conclude that we
need to know more about it before any definitive conclusions can be reached.
The unanchored payment card consists of a card with dollar amounts
arrayed from smalt to large. In this card, no anchors are used. The format
proved reasonably effective in several previous studies--Brookshire, Cummings,
et al. [1983] and Desvousges, Smith, and McGivney [ 1983]--and was used in
the focus groups as an alternative to the direct question. After using the
card with several focus groups, the participants suggested that the card was
of little value in helping them determine their willingness-to-pay amounts.
Based on these comments, and, perhaps more importantly, on the number of
issues that needed to be addressed in this research design, the research de-
sign did not attempt to compare the direct question and payment card formats.
7.10.2 Perceptions
People's perceptions of the contingent commodity is the second contingent
valuation issue that is relevant to our research design. Cummings, Brookshire,
and Schulze [1984] consider people's perceptions of the commodity--!.e. the
mental picture they envision--as one of the basic issues that affect the "accu-
racy" of contingent valuation as an approach for measuring the benefits of
changes in environmental quality. They suggest that four aspects of percep-
tions will affect the "accuracy" of contingent valuation;
Perceptions of hypothetical environment changes are consonant
with real effects.
All subjects are valuing the same commodity.
Perceptions of the commodity are invariant over time.
Perceptions of the commodity are independent of the quality
and quantity of information provided.
7-30
-------
Cummings, Brookshire, and Schulze [19841 concluded, based on the earlier
Burness et al. [1383] study, that it would not be possible to specify risk
effects of alternative policies related to the regulation of hazardous waste dis-
posal because contingent valuation has no real world or "practical anchor" for
accuracy. Clearly, their conclusion suggests that perception issues will be
crucial to our research design,*
Consequently, several aspects of our research design address the Cum-
mings, Brookshire, and Schulze [1984] concerns about perceptions. In fact,
one purpose of the focus groups and videotaped inter views--which allowed the
use of different vehicles in presenting risk changes and evaluating their per-
formance—was to aid in understanding people's perceptions. Based on this
experience, the questionnaire was structured to introduce risk and elicit indi-
viduals' perceptions with respect to a variety of different types of risks before
asking the valuation questions. For example, we used a risk ladder to elicit
people's perceived risk of dying from hazardous waste, (The exact details of
this development are reported in Chapter S.) This occurred prior to the fram-
ing of the contingent commodity to provide an independent evaluation of peo-
ple's perceived risk of dying from hazardous waste. We also questioned people
about the relative importance of specific pathways that they might perceive as
being important for exposure to hazardous wastes. Thus, information on per-
ceptions was elicited separately to provide some insights into the potential role
of perceptions on individuals' values of changes in hazardous waste risks.
We framed the contingent commodity as a change in the risk of being ex-
posed to hazardous wastes. This risk change was presented using an entirely
different vehicle than the risk ladder. The risk circles were used for each
of two components of the risk facing an individual. The first circle identified
*Some of the Cummings, Brookshire, and Schulze [1984] accuracy condi-
tions for perceptions are somewhat puzzling. It is unclear that this would
not also be the case for market revealed values. For example, the perceptions
of a person with perfect pitch of the quality of sound from a stereo speaker
might account for his having a higher willingness to pay for that speaker than
someone who is deaf to the full range of sounds from the speaker. Conven-
tional demand theory allows that differences in characteristics of individuals
may affect their demand for a commodity. Thus, a person's perception of a
commodity-contingent or otherwise--would seem important in influencing wil-
lingness to pay.
7-31
-------
a risk of exposure and the second the conditional risk of dying if exposed.
The decision to use the two circles resulted from the focus group participants'
comments that it was easier to understand the commodity--the risk change—that
the regulation was supposed to provide. Detailed explanations and visual aids
also were used in explaining risk to the respondent. These explanations were
followed by a specific hypothetical situation--expressed in concrete terms—that
finished the framing of the commodity prior to the elicitation of the values.
Finally, the respondent was given information about the baseline level risk
and asked to consider these as if they were the actual risks from the hypo-
thetical situation.
In summary, our efforts to recognize risk perceptions play an important
role in our final research design. Chief among these were the focus group
and videotape sessions that led to the separate treatment--indeed separate eli-
citation vehicles--perceived risks and the contingent commodity.
7.10.3 The Role of Information
The final contingent valuation issue is the role of information and its
effect on the design. The job risk part of the design included our attempt to
address the effect of information on individuals' values for reducing hazardous
waste risks. Specifically, after eliciting the values for the two changes in
risk, the interviewer then provided the individuals with information about their
actual risks of a fatal accident on the job. They were then allowed to revise
their amounts based on the new information if they wanted to do so. The im-
portance of this procedure is that it provides a gauge of how the individual
responds to new information, an issue highlighted in the Cummings, Brook-
shire, and Schulze [1984] overview of the contingent valuation literature. If
we assume that these individuals act the same way in response to new infor-
mation about hazardous waste risks, then it may be possible to use their re-
sponses from the job risk section in the analysis of the values for reductions
in hazardous waste risks. However, this must be treated as a maintained
assumption. Specific analysis of these responses was not undertaken as part
of the Phase I research but will play an important part in further research
with the survey results.
7-32
-------
7.11 THE DESIGN FOR COMPARISON WITH INDIRECT METHODS
Since one objective of this research was to undertake a comparative analy-
sis of the valuation estimates for a risk change implied by a hedonic property
value model with contingent valuation estimates, another factor influencing the
research design was related to the approach for developing consistent informa-
tion between the two methods. In this case, our approach accepted the dis-
tance of the housing unit from the hazardous waste site (see Harrison [1983])
as a proxy for the risk of exposure to these wastes. We attempted, first, to
determine whether distance served as a good proxy for individual's perceived
risk and, then, to develop information that would permit the estimation of the
demand for distance (as a mechanism for reducing risk). The specific details
of these steps are outlined in Chapter 15. What is important for our present
purposes is the independence in the design of this component of the question-
naire from the other features and the assumptions implicit in our structure.
Our approach poses a constant marginal price for distance to respondents and
then asks for their desired distance (for locating their homes) from a specified
hazardous waste site. The experimental design varied the marginal price
across individuais.
7.12 RESEARCH DESIGN; ITS INTEGRATION
The previous sections have highlighted some of the influences to our re-
search design for the contingent valuation survey. The design reflects both
one of the primary objectives of the research — to value changes in hazardous
waste risks and recognize important issues identified in past studies of behav-
ior under uncertainty--e.g., attributes of risks, context effects, the assign-
ment of property rights, and question formats. It also incorporates several
of the conclusions implied by our conceptual analysis — the importance of the
initial levels of risk for valuation and the role of intrinsic values. Finally, it
addresses the second objective of the research — to compare our survey results
with those from a hedonic property value study. This section explains how
these various goals are tied together to form our final research plan. To meet
these objectives, the research plan was designed, recognizing that
Different questions can be asked of each respondent
Different questions can be asked of different respondents--!. e.,
as part of an experimental design.
7-33
-------
To illustrate these two paths, Figure 7-3 provides a block diagram of the
major issues addressed in the research design. The first level shows our first
objective--to measure the benefits of reducing hazardous waste risks. The
next level shows the two types of values--!.e. use or Household, values and
intrinsic values—elicited from each individual in the sample. The design for
the research ends with the intrinsic values elicited as an increment to the
household values. Moreover, only the direct question format is used to elicit
the intrinsic values.
The remainder of the research design shown in Figure 7-3 pertains ex-
clusively to issues related to measuring the household or user values for re-
ducing hazardous waste risks. The third level of Figure 7-3 shows that each
respondent provided three different values for reducing hazardous waste risks:
two values for two distinct reductions in risk and a value to avoid an increase
in risk. The reduction in risk pertains directly to our first objective and the
risk avoidance value examines the effect of property rights on values. In
addition, to meet the needs of the comparative evaluation of different approach-
es to benefit estimation, the design uses a direct question format to eiicit indi-
viduals' desired distances from hazardous waste disposal sites, given that in-
creased distance will increase the price of the home. These responses provide
the information needed for one of the approaches for comparing contingent
valuation with hedonic models. The final level shows that two different ques-
tion formats were used to elicit the value for reduction in risks and that two
different levels of government were specified as actions for the risk avoidance
questions.
In the last level of Figure 7-3, the second path of our research plan,
the experimental design becomes more prominent. For example, values for risk
reductions are elicited using the direct questions from approximately 60 percent
of our sample, while 40 percent are elicited using the contingent ranking for-
mat. To illustrate the full design, Figure 7-4 shows the 24 separate versions
of the questionnaire allocated across the sample households. As shown in this
figure, key features of the design include the following:
Dividing the sample between the direct question (D) and the
contingent ranking (R) question formats to elicit the willingness
to pay for reductions in risk (Part A of Figure 7-4).
7-34
-------
Figure 7-3, Block diagram of expert menial design,
-------
Part A. Questionnaire Versions tor Valuing Reductions In.Risk
DUecl Qifeilwn fiSHltal (D)
Re Aiet(o«» In it*
ole*posnft
Mtk of itMlh, i •*poc«t
Vector
From To
vm
VM
WOO
mm
1
IS 1/1®
V» IBS
D3
D4
—
—
1!
Vtt M20
im
D1
D2
__
—
Ill
too UBO
VEO 1/150
D5
D8
—
_
IV
V300 WOO
mm vtsoo
_
—
D7
D8
Contingent Ranking Question Form* (R)
ImH ot risk
Amount of monthSf paymanl
Risks!
RbfcofdMlli.
Knpoud
Vector A
Vkctor B
Wtctor
$0 ttf »SS *15Q
I-2« IS $40 MO
WO
i
KB
ISO
mo
R1
R2
w»
({
ISO
wo
M60
WW
R3
R4
IflSO
Pari B. Questionnaire Versions tor Valuing Avoided Increases In Risk—Direct Question Formal Only (D)*
Part C. Questionnaire Versions for Valuing Moving Away
from a Hazardous Waste Site—Direct Question Only
"fowis Council-Approved f-Utk UKtftfts**
R«fcK«tMwb>iMi
olwpoaw
Rtak ol *M««, H npoMd
InciaMM In iM
fc
Pmm
VM
te#
vtao
mm
lb
f«o«
Mi
1M
VMM
1/200
ICS
V5
DM
(Ml
-
-
K5
115
032
042
--
~
vm
1110
on niib
«»*>
021
-
V50
V10
pejtt#
022
-
-
M60
ICO
FOI R41
-
-
—;—__
VEO
too
m? M2
-
-
—
11150
IOO
Dsi
D6t
Uft
W50
too
052
tw
i/ifioa
WOO
-
~
D8I
I«W
V300
-
-
OS
D82
Although Bit* part p f Bit design uwxt 01 >ly tti# tl« Part A dust^n,
Wwalon
((/MB*)
I
?S0
2
€00
3
1,000
4
1.300
Figure 7-4, Overview of questionnaire versions by experimental design component and question formal.
-------
Dividing the direct question versions to evaluate the influence
of different levels of exposure risk and conditional risk (direct
question format versions D1 through D8).
Dividing the contingent ranking versions to evaluate the influ-
ence of different combinations of exposure risk and payment
amounts (versions R1 through R4).
Matching the exposure and conditional risks used for the value
of avoiding risk increases (Part B of Figure 7-4) with the risk
levels used in Section F for valuing risk decreases.
Dividing the versions for avoiding risk increases to reflect dif-
ferences in the hypothetical scenario if the town council had
approved the risk increases (versions D11 through D81; R11
through R41) or if the Federal government had decided to allow
the increase (versions D12 through D82; R12 through R42).
Keeping this part of the design independent of the probability
levels resulted in 24 separate versions of the questionnaire.
As shown in Part A of Figure 7-4, the direct question versions focused
on the potential importance of exposure risks and the conditional risk of dying
(prematurely) from hazardous wastes if exposed. The groupings of the risk
levels in this part of the design into four vectors implies that each household-
er sample point—will provide values for two risk reductions. For example,
households receiving Version D3 in Vector I provided values for risk reduc-
tions from 1/5 to 1/10 and 1/10 to 1/25 with the conditional probability held
constant at 1/10. The values for the same two exposure risk changes were
elicited from households receiving Version D4 except that they were given a
conditional probability of 1/20. Overall, Vectors lf II, and III comprise a
3x2 factorial design for the initial levels of exposure risk and conditional
risks; and Vector IV is a 1 * 2 design for the lower probability cases.
Finally, the risk increments were developed in a very specific way to
account for how people respond to risk changes. Specifically, the percentage
change from the initial risk level to the intermediate level was held constant
across the exposure risk vectors as was the percentage change from the inter-
mediate level to the final level. However, the percentage changes were not
the same. As noted earlier, with the two different sets of percentage changes,
it may be possible to pool the responses to examine the effect of different per-
centage changes. In addition, the size of the risk change in attaining the
lower level was held constant in all elements of this part of the design at 1.67
7-37
-------
times the size of the initial risk change. For example, in questionnaire version
D1, the decrease in exposure risk went from 1 over 5 to 1 over 10 for the
first level, while in the second level it decreased from 1 over 10 to 1 over 25.
Part A in Figure 7-4 also shows the survey sample divided between the
direct question and contingent ranking versions to evaluate the effect of Ques-
tion format. Within the contingent ranking portion of the design, a 2 x 2 fac-
torial design was developed to evaluate the Influence of different paired com-
binations of exposure risks and payment amounts (Vectors I and II within the
R design). The specific combinations of exposure risks and payment amounts
used in the factorial design are also shown in Figure 7-4. The payments
amounts are structured to provide respondents with the central tradeoffs be-
tween lower exposure risks and larger monthly payments in higher prices and
taxes. For example, one set of payments, used in the R1 and R3 versions of
the questionnaire, provided one choice of zero payments for a baseline level
of risk, while the other set, used in the R2 and R4 versions of the question-
naire, provided one choice that would allow the respondent to reduce his pres-
ent payments by $20 but only with an increased risk of exposure. It is also
important to note that the payments and risk levels were given to the respond-
ent as ordered pairs.
The related aspects of the contingent ranking design focused exclusively
on different levels of. exposure risks. As shown in Figure 7-4, the levels of
exposure risks in the contingent ranking portion of the design overlap those
used in the direct question portion. However, there are some important dif-
ferences between the direct question and contingent ranking designs. For
example, the exposure risks used in versions R1 and R2 include an exposure
risk level of 1 over 100 that is not employed in the direct question design.
In addition, the ranking design holds the conditional risk of dying if exposed
constant throughout the design. The rationale, for this decision stemmed from
the difficult tradeoff involving the cost of additional design points and the
potential information to be gained. Although the conditional risks are poten-
tially important for this part of the design, we considered examining the influ-
ence of different combinations of exposure risk levels and payment amounts
more important for the ranking design. Earlier research (Desvousges, Smith.
7-38
-------
and McGivney [1383]) suggested the need to have a test for the influence of
different payment amounts, as well as the independent comparison of question
formats provided in the present design.
Part B of Figure 7-4 provides the details of the design for eliciting indi-
viduals' values of avoiding increases in risk. There are several distinguishing
features of this part of the design. First, only the direct question format was
used to elicit these values. Second, these values were elicited from all sample
households but the endpoints of the risk change were Hnked to the risk
changes used in the risk reduction part of the design. For example, the risk
endpoints for Version D31 in Part B, the risk increase portion, were 1/25 and
1/5 the same endpoints for Version D3 in Part A for risk reductions. Not only
v\ere the endpoints the same but the conditional risks were also the same to
avoid mixing the effects of the risk avoidance and the conditionat probability.
The third important feature of Part B accounts for the need to have the
24 separate versions of the questionnaire instead of 12. This feature is the
role of government that was specified in the hypothetical situation for the value
to avoid a risk increase. To allow the town council approved vs. Federal gov-
ernment allowed revisions to be independent of the risk changes, it was neces-
sary to have a separate version for each risk change. In continuing the above
example, there is a Version D32 that differs from D31 only in the type of gov-
ernment actions specified in Part B.
The last feature of this part of the design is that the overlap in the risk
levels in contingent ranking and direct question versions in Part A results in
three sets of observations in four cells of Part 8 of the design. For example,
this accounts for both D11 and R11 in value to avoid a risk increase to 1/50
from 1/10. This feature enables us to evaluate, at least for a subsample,
whether the question format in the prior design for risk reductions affected
values in the risk increase section that followed. However, there were also
several intervening questions (e.g. critter values and certainty effect) between
these major sections. This would be expected to reduce the effect of the for-
mat of the prior risk questions.
Finally, Part C of Figure .7-4 shows the design elements that related risk
reduction to distance. In this part of the design, the individual was offered
the hypothetical choice between purchasing two homes that were identical ex-
7-39
-------
cept for their respective distance from a manufacturing plant with landfill on
site that contained hazardous wastes. Using the average price of a home in
the neighborhood, the individual was asked how many miles he would want to
have the house away from the plant if it cost "$x/mile" in higher housing
prices. In effect, the individual was given a constant marginal price for re-
ducing risk by moving but the price varied across individuals. The design
specified four different prices per mile--$250, $600, $1,000, and $1,300. This
part of the design was considered independently of the other two parts of the
design, which implied that the versions in Part C would be assigned without
considering the other features in the overall design. Had this not been the
case, 36 separate versions of the questionnaire would have been necessary.
7,13 IMPLICATIONS
Using the main objectives of our research as guideposts, this chapter
has described how we have integrated some of the many facets of risk into a
research design for eliciting individuals' valuation for reductions in the risk
of exposure to hazardous wastes. Additionally, the chapter has highlighted
the underlying reasons for the different parts of the design. In this process
we have explained our reasons including or excluding certain facets of risk,
or some methodological concerns about contingent valuation, as part of the
design. Ultimately, our final design is somewhat eclectic but this chapter sug-
gests that its composite nature can be viewed as consistent with our main
objectives.
The final design suggests an important, and perhaps sometimes unappreci-
ated attribute of contingent valuation as an approach for benefits measurement.
That is, contingent valuation provides a very flexible framework for developing
tests for basic economic hypotheses. For example, in our final design, we
have used this flexibility to examine the importance of question format, initial
levels of risk, and different assignments of property rights by formulating
different versions and assigning them to different parts of our sample.
Although not controlled to the extent that is possible in a laboratory, it does
allow for some degree of control. Moreover, the subjects respond in the same
environment in which they make many economic decisions.
Finally, our design reflects the importance of the focus group sessions
and other questionnaire development experiences to our final design. They
7-40
-------
proved essential in evaluating both different ways of approaching a question
and the format for asking that question. Additionally, they suggested unanti-
cipated hypothesis that are included in the design. A last consideration is
that these activities were very compatible with the format and structure of
contingent valuation survey that ultimately, would be implemented.
7-41
-------
CHAPTER 8
SURVEY QUESTIONNAIRE DEVELOPMENT
8.1 INTRODUCTION
Developing a contingent valuation survey questionnaire that could effec-
tively measure the benefits of hazardous waste management regulations by eli-
citing individuals* valuations of risk reductions was a difficult problem. Exper-
imental psychological and economics research suggested that individuals' re-
sponses to questions involving decisions under uncertainty could be influenced
by a number of factors, including the respondent's previous experience, the
explanation of the situation, and the characterization of the uncertainty. In
addition, earlier research suggested several compelling reasons to expect par-
ticular difficulties with situations involving the risks associated witn hazard-
ous wastes (Cummings, Brookshire, and Schulze [1984]),
Thus, the questionnaire development effort for this research faced two
basic problems. First, we had to develop a set of procedures for the ques-
tionnaire that could effectively explain both the choices to be made under the
uncertainty associated with hazardous wastes and, equally important, "the
changes that could occur in the uncertainty itself. Past psychological and
economics research offered some valuable insights here, of course, but much
of it is based on laboratory experiments whose results did not seem clearly
transferable to a general population survey at the outset of this research
effort. Second, and especially important to the objectives of this research,
we had to develop an accurate description of the features of the risks associ-
ated with hazardous wastes and the ways in which regulatory actions might
affect those features.
In view of these problems, therefore, we spent a great deal of time and
effort during the questionnaire development and testing effort to try and un-
derstand accurately how people feel, think, and talk about risks, hazardous
wastes, and other related topics. Appendix B contains two versions of our
8-1
-------
-------
survey questionnaire--one for the direct question format and one for the con-
tingent ranking format. Concentrating primarily on the use of the focus group
discussion technique and the efforts to pretest the questionnaire and to video-
tape its administration, this chapter briefly outlines the evolution of the ques-
tionnaire and its basic logic, in particular, after offering a short chronological
overview of the entire questionnaire development process, the following sections
describe the focus groups, how and why they were used, and what we learned
from them' and the other questionnaire development and testing activities. Spe-
cifically, Section 8.1 highlights the questionnaire development process, includ-
ing the focus group, pretest, and videotaping activities; Section 8,2 describes
what focus groups are and how they work; and Section 8.3 explains their role
in contingent valuation. Section 8.4 describes how the focus groups were
organized. Section 8.5 profiles the participants, concentrating on their knowl-
edge and awareness of the hazardous waste problem; and Section 8.6 offers a
brief summary and overview of what the project team learned from the focus
groups. Section 8.7 describes the significant pretest activities conducted dur-
ing the post-focus-group effort to further refine and test the survey question-
naire; Section 8.8 provides the same information for the videotaped interviews;
and Section 8.9 concludes the chapter by highlighting some suggestions for
enhancing the overall questionnaire development process,
8.2 OVERVIEW: A BRIEF CHRONOLOGY
As a first step in the survey questionnaire development process, a series
of informal discussions--focus group sessions — were conducted with small
groups of citizens in North Carolina and Massachusetts during April, May,
June, and September 1983. The purpose of these sessions was to learn how
best to communicate risk information to individuals and to understand how they
think about hazardous wastes. Together, these sessions yielded substantial
information--primarily on what individuals feel, think, and say about the risks
associated with hazardous wastes—that was invaluable in the questionnaire
development process, fn particular, because the contingent valuation survey
approach requires a questionnaire that creates a hypothetical--or simulated- -
market for goods not usually bought and sold (in this case, reductions in the
levels of risk associated with hazardous wastes), the focus groups proved in-
8-2
-------
valuable in collecting information on attitudes, perceptions, and language that
helped frame the questionnaire's hypothetical market in terms that were both
credible and understandable to survey respondents.
Following the last series of focus groups, the survey questionnaire was
judged to have the appropriate structure. The sequence of questions, the
amount and types of information they contained, arid their general structure
and format seemed to be "working" reasonably well. Despite having an appro-
priate structure, however, the questionnaire was clearly not ready for actual
data collection because it had not been fully tested one-on-one under actual
field conditions with a respondent. For example, the questionnaire had always
been administered by a member of the project team--a situation that could not
be duplicated in field work conducted with professional interviewers. In addi-
tion, the questionnaire had not been tried in the residence of a respondent,
whose participation is always subject to varying interview conditions—televi-
sions, children, telephones, etc. To minimize the chances of encountering
unexpected problems in the field, therefore, the project team decided both to
field test the survey questionnaire and to videotape ten one-on-one interviews
with selected respondents. These activities resulted in changes to the ques-
tionnaire that substantially improved its ability to frame —i.e. , explain—the
hypothetical market for risk reductions in such a way that respondents could
understand it and make willingness-to-pay decisions based on it.
8.3 FOCUS GROUPS: THE BASIC INGREDIENTS
Focus groups are informal discussion sessions in which a skilled moderator
leads a group of individuals through an in-depth discussion of specific topics
to discover their attitudes and opinions. Neither the participants nor the mod-
erator is necessarily an expert on the topics. A concept that grew out of the
psychiatric techniques of group therapy, the focus group assumes that individ-
uals are more apt to talk about a problem in the security of a group environ-
ment than they are in a one-on-one interview. In the 1950s some researchers
extended focus groups beyond their intial therapeutic purpose and used them
to obtain qualitative information from consumers about product advertising and
promotional efforts [Bellenger, Bernhardt, and Goldstucker, 1979].
8-3
-------
Traditionally, focus groups have served as a toot in marketing research
to acquire qualitative data on markets, prices, and the advantages of new
products. In addition, focus groups have been used to
Generate new hypotheses
Provide background information on new product concepts, pac-
kaging, and advertising effectiveness
Understand the consumer language associated with specific
product categories or brands
Stimulate new ideas about older products
Structure and test questionnaires
Interpret previously obtained quantitative results.
This project used focus groups in yet another way--to obtain and evaluate
the information necessary to develop a contingent valuation survey question-
naire. Specifically, the focus groups provided an opportunity to listen as
individuals discussed various aspects of hazardous wastes; to observe their
responses to several tasks that would be used in the contingent valuation sur-
vey; and to try alternative methods for presenting information about the risks
of hazardous waste contamination and other low-probability events.
8.4 FOCUS GROUPS: THEIR ROLE IN CONTINGENT VALUATION
I n general, the focus groups were used in this research because ihev
offered a cost-effective way of discovering how best to ask economic ques-
tions—especial ly those concerned with risk--of noneconomists. In particular,
however, they were used to gather the kinds of information essential to the
effective use of the contingent valuation survey approach to estimate the Den-
efits of hazardous waste management regulations--information that could help
explain the survey questionnaire's hypothetical market for risk reductions in
terms the respondent's could easily understand. For example, contingent valu-
ation requires the resolution of issues related to framing--!. e. , the definition
of the commodity in its hypothetical market and how the transaction would
occur. Resolving these issues requires assessing whether responses are af-
fected, for example, by the information given, by the way in which the valu-
ation question is asked, or by the actual sequence of the questions on the
8-4
-------
questionnaire. Because they demonstrated in specific terms how respondents
may react to varying types of information, varying types of questions, and
varying question sequences, the focus groups helped assess these framing
issues as the questionnaire was developed.
In addition, using contingent valuation to estimate the benefits of hazard-
ous waste management regulations also requires detailed information on how
and the extent to which respondents understand risk (or probability) and how
government regulatory actions might change it. In particu-lar, it is essential
to determine what respondents are likely to know about these concepts before
they are given information necessary to help them form notions of willingness
to pay. Focus groups helped resolve this issue, particularly in discovering
whether respondents think of risk in two separate stages — risk of exposure to
hazardous wastes and risk of some resulting detrimental effect—and they helped
identify language that would effectively communicate hazardous waste concepts.
Finally, the focus groups also proved an excellent way to test alternative
methods of elicitng individuals' willingness to pay; to compare the workability
of direct questions to elicit willingness to pay values with that for contingent
ranking, which requires respondents to rank outcomes stated in terms of prob-
abilities and wlIingness-to-pay amounts; and to ensure the development of a
clearly worded, comprehensible survey instrument. The focus groups were
particularly helpful in the latter effort since the participants were able to point
out fuzzy language and muddy or inadequately described concepts before the
instrument was administered to the general target population.
8.5 FOCUS GROUPS: THEIR ORGANIZATION
The contingent valuation survey questionnaire evolved during a series of
activities that spanned six rounds of focus groups, involved conducting 19
sessions in a variety of geographic areas, and required the participation of
138 men and women from a variety of economic, social, and educational back-
grounds. Table 8-1 summarizes focus group session attendance. Round 1
consisted of general discussions centered around five major topics: risks in
general, environmental attitudes, hazardous waste knowledge, hazardous waste
risks, and attitudes toward paying for hazardous waste management. Figure
8-1 shows a sample of the questions used as guidelines for these discussions.
How and the extent to which the focus group participants responded to these
8-5
-------
TABLE 8-1. FOCUS GROUP SUMMARY
Number of
participants
Round
Session
Participating organization
Location
Date
Per
session
Per
round
1
1
Duke institute for Learning
Durham, NC
April
6, 1983
14
in Retirement
2
White Rock Baptist Church
Durham, NC
April
11, 1983
7
50
3
Vance County Heart Association
Henderson, NC
April
12, 1983
9
4
Triangle Presbyterian Church
Durham, NC
April
13, 1983
20
2
1
iNCO Sheltered Workchop
Henderson, NC
April
27, 1983
8
2
YWCA/Hobby Time Group
Durham, NC
April
28, 1983
11
27
3
Methodist Retirement Home
Durham, NC
April
29, 1983
8
3
1
Salem United Methodist Church, 1
Haw River, NC
May 5, 1983
12
2
Salem United Methodist Church, 11
1 Haw River, NC
May 24, 1983
7
35
3
Ridgeroad Home Extension Club
Durham, NC
May 25, 1983
16
4
1
Union Presbyterian Church
Carthage, NC
June
1, 1983
13
1Q
2
Saint Catherine Catholic Church
Wake Forest, NC
June
2, 1983
6
I j
5
1
Presidents Crime Watch Council
Wadesboro, NC
June 21, 1983
22
2
Morven Presbyterian Church
Morven, NC
June
22, 1983
5
Women of the Church
41
3
Morven Presbyterian Church,
Morven, NC
June
22, 1983
14
Evening Group
6
1
Acton Congregational Church
Acton, MA
Sept.
13, 1983
7
2
Concord Council on Aging
Concord, MA
Sept.
14, 1983
7
nc
3
Acton League of Women Voters
Acton, MA
Sept.
14, 1983
6
C.O
4
Needham American Red Cross
Needham, MA
Sept.
15, 1983
6
-------
HAZARDOUS wastes focus GROUP QUESTIONS/TOPICS
1. In what ways do you think that you individually pay (mortetiril y) as a result
of the "hazardous waste proOIem."
2. To whom do you pay? Wh^re decs the money go?
3. How it that money Spent by the recipient; s > on h«art)ous wast® management?
4. Have you personally or memBeri of your family actually experienced
bodily harm or loss or injury to property due to hazardous wastes*
5. Do you believe in the possibility of personal lots or injury to yourselves as a
result of hai»rdous wastes?
6. What do you think about the chances (orobabili ty) that you will actually experi-
ence persona! loss or injury due to hazardous wastes?
7. What do you think about the chances that the environment will actually be din-
aged by hazardous wastes.
8. if you think thai the chances art good thic you will personally experience lass
or injury from hazardous wastes, would you 0* willing to pay mors than you
now do to change the probabilities of toss or injury?
9. If you think that tn® chances are good that the environment will suffer damage,
would you be willing to pay mora than you now do to change she proeaoiiities
of loss or injury?
10. If you think chat there is no Chance that you or your immediate family will suf-
fer lost or injury as a result of hazardous wastes, would you Be willing to pay
mora than you now do to change the probabilities that others will Suffer loss
or injury?
11. If you think that thera is no chance that you or your immediate family will suf-
fer less or injury as a result of hazardous wastes, would you D# willing to pay-
more than you now da to change the probabilities that trie environment will Oe
damaged?
12. Whom do you hold responsible for proper hazardous wast® management'
13. whom do you hold resoonsibie For the "hazardous waste problem.?"
14. Ta whit degree do you hold each of the following responsible for proper hag-
araous waste management;
{1} yourselves (S) Federal Government
(2) society (S) hazardous waste producers
{3} local government (7) companies that dispose af
(4) State government hazardous wastes
15. To what degree to you hold each of the following responsible for hazard-
ous wast® cleanup;
(1) yourselves (.5) federal Government
(Z) society (6) hazardous waste producers
(3) local government (7) comoam«s that: dispose of
(4) State government hazardous wastes
Figure 8-1. Sample questions used in focus group discussions.
8-7
-------
and other questions provided the information necessary to judge what kinds
and amounts of information should be provided in the survey questionnaire so
the respondent could form his notion of willingness to pay for risk reductions
resulting from hazardous waste management regulations. As the focus group
sessions were conducted during Rounds 2, 3, 4, and 5, the types and amounts
of information given to the respondent--!. e., both the questions on the survey
questionnaire and the supplemental materials used in the interviewer's presen-
tation to the respondent--were substantially refined until, in Round 6, a first
draft of the questionnaire was administered.
8.6 FOCUS GROUP PARTICIPANTS: THEIR AWARENESS OF
THE HAZARDOUS WASTE PROBLEM
_ While the character of almost all the discussion sessions was largely the
product of one or a mix of such important demographic variables as economic,
social, and educational background, the factor with the greatest impact on the
participant feelings and attitudes about hazardous wastes and the risks associ-
ated with them was personal awareness or experience--! .e., whether or not
hazardous wastes and their risks had become a local issue for some reason.
Table 3-2 lists the location of each of the focus group sessions and briefly
indicates whether, to what extent, and how the participants in them became
aware of the hazardous waste problem.
As shown in Table 8-2, participant awareness of hazardous wastes and
their risks is particularly high in areas whose residents had experienced a
hazardous-waste-related accident, as had the participants in the sessions held
in Acton, Massachusetts, where the local water supply had recently been con-
taminated by chemicals leaking from a hazardous waste landfill site. Residents
of areas that had recently faced a landfill siting decision were also highly
aware of the hazardous waste problem and its potential risks, as illustrated
by the participants in the Warren County and Anson County, North Carolina,
sessions, whose communities, respectively, had unsuccessfully and successfully-
fought landfill siting decisions. In contrast, awareness of hazardous wastes
and their associated risks was very low in areas whose residents had not ex-
perienced a local incident or fought a landfill siting. The responses of the
participants in the Haw River, North Carolina, sessions, for example, show
little awareness--indeed, little understanding—for what hazardous wastes are
or the number and types of risks they might pose.
8-8
-------
\
TABLE 8-2.
FOCUS GROUP PROHIL: PAR1ICIPANT AWARUNE 1>S OF THE
HAZARDOUS WASTE PROBLEM
Location of ussion
Participating organisation
City
State
Description arid source of participant awareness
Ouke Institute for Learning in
Retirement
White Rock Baptist Church
Vance County Heart Association
Durham North Carolina
Durham
North Carolina
Henderson North Carolina
CO
cb
Triangle Presbyterian Church
Durham
North Carolina
inco Sheltered Workshop (Warren
County)
YMC A Hobby Time Group
Methodist Retirement Home
Henderson
North Carolina
Durham
Durham
North Carolina
North Carolina
The participants In this group had a heightened awareness
and understanding of hazardous wastes due to several local
incidents--*.a-. PC8 dumpings on North Carolina highways,
the Warren County PCS landfill siting controversy, and a fire
•t a chemical waste recycling company in Durham.
Most participants had a poor understanding of hazardous wastes,
although they were able to site local incidents they had heard
about in the ., the Warren County PCS landfill
controversy.
Most participants were aware of hazardous wastes and their risks
due to the eontroversey surrounding the siting of a PCB landfill
in adjacent Warren County against the strongly expressed pro-
tests of Warren County residents. Because of the proximity of
their community to the Warren County landfill site, these partici-
pants had well-developed ideas on hazardous waste, particularly
concerning possible compensation and its use in landfill siting
decisions.
Although this group had little personal experience with or aware-
ness of the hazardous waste problem, some participants' were
aware of the Warren County landfill siting controversy, and a
few people had detailed technical knowledge o< the hazardous
waste problem. Nevertheless, this group's understanding of
hazardous wastes was not precise, and at leaM some participants
expressed reservations about paying the costs ef control.
Perhaps because their community is in such close proximity to the
Warren County landfill site, these participant* fell the hazardous
waste problem was huge and perceived their probability of
exposure as nearly 100 percent. They used the term hopeless-
ness to describe the hazardous waste problem arid were eager lo
express their opinions.
A few of these participants cited local incidents — the Warren
County PC6 landfill and a chemical recycling plant fire in
Ourham--as sow res cf their awareness of the Iwardous waste
problem, but their understanding of what constitutes hazardous
wastes was incomplete.
These participants were very sensitive about how they were per-
ceived by others «ncfr consequently, were cryptic and defensive
about their awareness of the hazardous waste problem. They
seemed to understand that some substances are hazardous, but
not how they are related to manufacturing processes for con-
sumer goods.
i rrttl 1 lit) i§sH V
-------
TABLE 8-2 {continued)
Location of session
Participating organization City Slate
Salem United Methodist Church, 1 Haw River North Carolina
Salem United Methodist Church, II Haw River North Carolina
Ridgeroad Home Extension Club Durham North Carolina
Union Presbyterian Church Carthage North Carolina
St. Catherine Catholic Church Wake Forest North Carolina
Anson County Crime Watch Wadesboro North Carolina
President's Council
Women of Morven Presbyterian Morven North Carolina
Church . ... ... .
Description and source of participant awareness8
These participants were poorly informed about hazardous wastes:
One person asked what PCBs were, and another wondered if acid
rain came from Agent Orange. Perhaps due to their lack of
knowledge, these people were less afraid than most Of the
effects of hazardous waste exposure.
Though somewhat more Informed than the participants In the
previous discussion group, these individuals also had limited
knowledge of the hazardous waste problem, particularly of
effects or exposure, f-or example, they did not understand how
leaving PCB-lated oil on the shoulders of North Carolina's high-
ways could create an exposure problem.
These participants indicated they knew about hazardous wastes
through the media coverage of local events--e. g., the Warren
County PCB landfill siting controversy - - but they did not have •
clear understanding of what constituted hazardous wastes and
had difficulty giving specific examples: "Might have fumes
associated with it. "
These participants knew a great deal about the hazardous waste
problem. They were aware of various exposure paths {particu-
larly ingestion} and of the various products and manufacturing
processes that produce hazardous waste byproducts, in addi-
tion, they followed not just iocai incidents (such as the Warren
County landf
-------
TABLE (S-i? (continued)
Location of session
Participating organization City State
Morven Presbyterian Church Morven North Carolina
Members
Acton Congregational Church Acton Massachusetts
Concord Council on Aging Concord Massachusetts
Acton League of Women Voters Acton Massachusetts
Needham American Red Cross Needham Massachusetts
Description and source of participant awareness3
Like the previous group, these participants lived in a community
that had successfully fought a proposed hazardous waste landfill.
They were very well informed about hazardous wastes, their
risks, and the alternatives for waste cleanup,
Probably because hazardous wastes from a leaking chemical land-
fill site had contaminated their water supply, these participants
were well aware of the potential risks of hazardous waste expo-
sure and effects. In general they felt they were very likely to
be exposed to hazardous wastes, and, in particular, they felt
exposure would most likely occur through their drinking water
supply.
The participants in this group were also very aware of the nature
of the hazardous waste problem, probably because of the close
proximity of their community to Acton, whose water supply had
recently been contaminated. These participants were less sure
about the levels of risk associated with exposure, however, and
they had difficulty estimating cleanup costs.
Like the previous group held in Acton, this group was knowl-
edgable about hazardous wastes due to a recent local incident in
which their drinking water supply became contaminated by haz-
ardous wastes- However, the large extent to which the
participants identified with their own local incident prevented
them from thinking about hazardous wastes in the hypothetical—
i.e., they had difficulty describing what they would be willing
to pay to reduce their risks in a hypothetical situation involving
risks from hazardous wastes.
Because Needham is further than Concord from Acton, whose
drinking water recently became contaminated, these participants
were somewhat less aware of the hazardous waste problem than
were Concord participants. Unlike the Concord and Acton
participants, for example, they perceived their own risks as
zero, and they indicated they were less environmentally con-
cerned than the other Boston-area participants.
3For a more precise account of focus group participant awareness of and experience with hazardous wastes and their risks, see
Desvousges el al. [1984a).
-------
8.7 FOCUS GROUPS AND QUESTIONNAIRE DEVELOPMENT:
OVERVIEW AND SUMMARY FINDINGS :
Experience with the focus group sessions suggested that they were a
valuable tool in constructing the survey questionnaire, both, in terms of learn-
ing how people think, feel, and talk about different issues and in terms of
the mechanical aspects of organizing and writing individual questions and visual
aids for the final survey instrument. The following discussion briefly summar-
izes these judgments, concentrating on the significant mechanical and perceptu-
al issues of effectively presenting risk information to survey respondents. For
further details, the interested reader can consult Desvousges et al. (1984a, b ],
which this section summarizes, for more detailed discussions of how the focus
group sessions were organized, conducted, and analyzed.
8.7.1 Overview: Findings and Issues in Questionnaire Development
In almost every instance, the focus group participants provided important
information for the survey questionnaire development process, including both
substantive and editorial comments that resulted in substantial revisions to
the survey instrument. Many of the suggestions could not have been antici-
pated a priori. For example, participants found simple examples of everyday
risks useless for thinking about hazardous waste risks. In addition, while
circles (or probability wheels) were the easiest vehicle for communicating haz-
ardous waste risks, a risk ladder was more successful in eliciting responses
about perceived risks. Also, the participants found the visual aid used to
link the risk ladder and the probability wheels more confusing than helpful.
Fortunately, the participants were willing to provide explicit, detailed criticisms
of the visual aids and other survey materials.
The findings summarized below underscore the key element in the ques-
tionnaire development effort — the difficulty of presenting information about
risk to the general population. This task was a central objective of the focus
group research effort because it was the necessary first step to defining an
adequate way to "frame" (i.e., discuss and put in context) the hypothetical
commodity that ultimately would be valued in the contingent valuation survey.
The commodity to be framed in the survey is a change in the risk of exposure
to hazardous wastes and, corresponding to it, a change in the risk of a result-
8-12'
-------
ant effect, or death,* In effect, therefore, the questionnaire had to convey
information about a commodity or event that might or might not happen. t
Communicating the commodity itself is only one element in framing the
hypothetical commodity for a contingent valuation survey. It is also neces-
sary to provide a specific context for the commodity—in this case, a context
to explain how the exposure risk would arise, how it would be affected by
government regulations, and how people would "pay" for reducing the risk of
exposure (the "payment vehicle" in technical jargon). Once* the respondent is
g ven this information (i.e., the hypothetical commodity, the hypothetical con-
text, and the hypothetical market), he is asked to complete the valuation task,
during which he is asked to reveal his willingness to pay for the hypothetical
commodity.
Researchers have used many different formats to elicit willingness to pay-
in the valuation task. They have tried asking the respondent directly (Des-
vousges, Smith, and McGivney 11983]) and have used iterative bidding games
(Randall, Ives, and Eastman [1974]; Rowe, d'Arge, and Brookshire [1980];
Schulze, d'Arge, and Brookshire [1981]; and Desvousges, Smith, and McGivney
[1983]), They have used cards with payment amounts and anchors based on
average expenditures for other kinds of public goods (Mitchell and Carson
[1981 ] )--e. g. fire protection—and have tried rankings of specified payment
levels matched with levels of the hypothetical commodity (Rae [1981a,b) and
Desvousges, Smith, and McGivney [1983]).
Based on past experience, the direct question and the ranking formats
were selected for evaluation in the focus groups because they represented
two extremes in terms of the amount of information provided for the respond-
ent: no information in the direct question format and a great deal of informa-
tion (including specified payments) in the ranking format. Finally, these two
formats also avoid the problems caused by choosing the various starting points
*Other nonlethal health effects are possible from hazardous waste expo-
sure. For simplicity, the single effect of death was chosen because it is easier
to define than a particular severity of a specific illness.
tBrookshire, Cummings, et al. [1982] found that their willingness-to-pay
bids were quite sensitive to the changes in the framing of the hypothetical
commodity.
8-13
-------
.
necessary in the iterative bidding game format (see Mitchell and Carson [1981]
and Desvousges, Smith, and McGivney f 1983]).
8.7.2 Presentation of Probability
Introduction
The potential for difficulties in explaining the probabilistic nature of what
can be expected from hazardous waste regulations was evident from the outset
of the research. Previous research has identified many" potential problem
areas. Hershey, Kunreuther, and Schoemaker [1982] have found considerable
variation in individual preferences for uncertain outcomes depending on how
probability is presented. These findings are echoed by Tversky and Kahneman
[1981] and Fischhoff, Slovic, and Lichtenstein [1982], Unfortunately, the
available research has not provided an unambiguous judgment on how best to
present probabilities. Acton [1373] used bar charts to show alternative risk
levels but did not evaluate the effectiveness of this vehicle. Jones-Lee [1978]
and Frankef [1979] used fairly complex representations of probability distribu-
tions; and Loomes [1982] expressed probabilities in terms of deaths per 100,000
members of the population. He found significant differences in preferences
with this measure depending on the equity implications implied in the presenta-
tion. Slovic, Fischhoff, and Lichtenstein [1978] used specific probabilities in
numerical terms (percent measure of risk in some time period) in their research
on accident probabilities and seat belt usage.
Selvidge's [1975] work suggests a number of areas for caution and offers
some new insights. She cautions that "asking someone .who has not worked a
great deal with very small probabilities to make such distinctions is analogous
to asking a member of a stone-age tribe to make judgments about lengths of
time" [p. 200], Her insights are that individuals can be acclimated to the task
by working them through specific hypothetical situations, then, asking for prob-
ability information or an evaluation in relative terms. She also suggests the
use of visual aids to highlight probabilities. Specifically, she recommends an
urn filled with balls of one color and one ball of a different color. (This is
analogous to the visual aid used by Schoemaker [1982] in his research.) How-
ever, two important factors limit the applicability of Selvidge's research to the
task of the present research. Selvidge was working with experts, requesting
8-14
-------
that they encode probabilities, and she was not conducting her experiments in
a person's home (as is the case in the contingent valuation survey). * There-
fore, the project team adapted the idea of using circles, or probability wheels,
from risk assessment research, during which experts were asked to encode
the probabilities for different risky situations. Wallsten [1983] was instrumen-
tal in explaining the workings of the vehicle and how it has been used in the
past.
Overview
It was apparent from the focus groups in Round 1 that participants would
have difficulty thinking of hazardous wastes as numerical risks or probabilities
even though they frequently showed a good intuitive understanding of risks
and hazardous wastes. It was also apparent there would be a wide range of
understanding of the probability concept among participants. Some people
appear to naturally think of risk in terms of probability while others do not.
These different levels of understanding caused difficulty both in presenting
probability to the focus groups and in explaining it within the questionnaire.
To increase the understanding of probability among the focus group partici-
pants, examples of risky events that participants might face in their everyday
lives were cited. Moreover, circles with shaded slices along with these exam-
pies were used to indicate the chance outcome for these risky events. Later,
when participants were asked to perform the contingent ranking, circles were
again used to convey the chances of exposure to and effects from hazardous
wastes. It was hoped that participants would link what they learned in the
general probability presentation to the contingent ranking task, where they
were asked to make payment decisions based on the probabilities of reducing
exposure risks.
*The present experience with risk is an interesting contrast with their
experiences with water quality (see Desvousges, Smith, and McGivney [1983]).
When the water quality questionnaire was developed, Mitchell and Carson [1981 ]
already had conducted a large-scale survey using a ladder to represent differ-
ent water quality levels tied to recreational uses of water. Thus, the framing
of the hypothetical commodity was a much easier task. The present research
could not be based on the structure of the earlier contingent valuation study
involving hazardous wastes because Brookshire, Cummings, et al. [1982] spec-
ified the commodity as a regulation and not as a risk.
8-15
-------
In the early focus groups that included a presentation, probability was
explained using two circles,* The first circle represented the risk of expo-
sure, and the second, the combined risk, of exposure and effect. Simple
examples of risky events such as "rain," " IRS audit," "fishing," and "car
accident" were listed beside the exposure circle, and the effects--"get wet,"
"pay more money/' "catch a fish," and "get hurt/' respectively--were listed
beside the combined risk circles. Each circle had a different portion shaded
to indicate the probability of the events' occurring.
In the ranking exercise, four cards (Cards A, B, C, and D) were used
at first. The possible probabilities of exposure were 8/360, 6/360, 4/360, and
2/380. In the last two sessions of the first round, two additional cards (Cards
E and F) with exposure probabilities of 1/380 and 25/380 were added. The
risk of effect was always 4/360, In this round, a circle showing combined
probability —the risk of exposure times the risk of effect — was not included.
A sample of the cards used in the early focus groups have been included as
Figure 8-2.
There were many problems with the presentation described above. First,
participant comments indicated that the shaded circles did not do a good job
of relaying the idea of chance. Adding spinners to the circles was suggested
by many participants as a way to improve them as vehicles for relating chance.
Second,, participants indicated they did not understand how the combined
probability was formed. They were not perceiving either that the chance of
exposure and the chance of effect were separate, or that the combined proba-
bility was the result of multiplying the exposure by the effect probability. t
This was true in both the simple probability explanation and in the contingent
ranking task, with different levels of understanding frequently appearing with-
in all groups. After this round, it was hypothesized that participants would
have an easier time determining willingness to pay for hazardous waste man -
*focus groups conducted in Round 1 comprised only a general, spontan-
eous discussion of general topics related to risk and hazardous waste and,
therefore, did not include a presentation using visual aids,
tThis is consistent with some experimental research in psychology indicat-
ing that Individuals have difficulty with multistage lotteries. See Sehum
[1980J.
8-18
-------
RISK OF * ETOR
CA® A
WWRXS MASK SiSKS
RISK CF M EFT€CT
(WW REUIRHh IS PER HM IS KIWI P8ICB MS IMS
OUCl
KwnK ware was
MS. CP MEW*
RISK. OF m EFFECT
FWB(T SEOJIIHK fXW$M\n HIQtS: PR1EXS «t IA«5
owe
PZMHE«£8i98
SIX CF m OHSW
aia of m effect
HSKNT HOJIIG): SUB FH m W HIS0 PS!IB M WES
Figure 8-2. Probability circles with various combinations for risk of exposure and effect.
8-77
-------
agement regulations if they were given very explicit information about probabil-
ity. Participant comments in this round supported this hypothesis.
Third, with the exception of fishing, the simple examples (such as given
in Figure 8-3) were easy for the participants to understand:
The fishing didn't fit. Everything was a negative effect except for
fishing. That was positive. The other examples all seemed like
things you had control over.
However, participants did not find the simple examples of risk helpful in under-
standing the chances of exposure and effects from, hazardous wastes. They
indicated that the attributes of everyday chances were so different from those
of hazardous wastes that one did not help explain the other:
There were too many examples preceding the hazardous waste exam-
ple.
t understand the examples of the chance of rain, etc., but I don't
understand the great relationship between your chance here and our
deciding which is the best order to rank the cards in.
In ranking the cards you go through a process of reasoning which
is different from that of the simpler examples, like the chance of
rain.
Finally, participants had trouble believing that the hazardous waste expo-
sure probabilities were real. In genera) they felt they were too small:
I wondered if what you were presenting was unbiased because of
the extremely small chance of being exposed to hazardous -vastes.
1 wondered if you were trying to program the results.
For later focus groups, the probability presentation was expanded to in-
clude three circles: an exposure circle, a conditional risk of an effect circle,
and a combined risk of exposure and effect circle. This change was made to
address the participant's need in the previous round for a better explanation
of how the combined probability was formed. In addition, it was hoped this
more explicit probability presentation would help participants understand both
that the risk of exposure and the risk of effect were separate events and that
the probability of an effect is conditional on a given level of exposure.
In this round, the research team added more descriptive titles to each of
the three risk circle cards. Instead of just displaying the words chance,
probability, and risk, the exposure card now included the title "What Will Hap-
8-18
-------
Card 4
Examples of Risk
Event Outside
Your Control
Your Circumstances
When Event Happens
What St Means
To You
How it Might
Have Been Anticipated
It might rain
You might
have a
flat tire
You might be
exposed to
hazardous
wastes
Walking from car
to work
(store, school, etc.)
On the interstate
(versus in driveway)
Physical makeup
(hereditary
background,
resistance, diet,
smoking)
Get wet
Stranded on
road (late)
Reduced life
expectancy
Bring an umbrella
or raincoat
Have a spare,
change tires more
frequently
Manage wastes
properly, recycle
wastes
Figure 8-3, Card in tabular form to present probability and explain simple risks.
8-19
-------
pen?" and the effect card included the title "What it Means to You." Also,
the example of fishing was excluded, but the examples of "rain," "IRS audit,"
and "car accident" were still used to illustrate effects. The card entitled
"What it means to you" included the results "be outside," "make a mistake on
return," and "glass breaks," respectively. The third circle included the com-
bined risks--"that it will rain and you get wet," "IRS audit and pay -more
money," and "car accident and get hurt," It was hoped these changes would
make it easier for each participant to relate to each circle." In addition, due
to the suggestions of the first groups, spinners were added to the circles.
Finally, there were five cards (Cards A through E) in the contingent
ranking exercise with exposure risks of 4/380, 6/360, 2/360, 1/360, and 25/360.
The risk of effect was still 4/360 and the risks were not combined explicitly.
These cards are included as Figure 8-4.
Participants stilt had difficulty understanding probability even after these
changes. The spinners seemed to do little in helping them to understand
chance:
He was telling you that there's a certain amount of the stuff you're
going to get irregardless.
Without a dumpsite you are still going to get. your share.
In addition, adding the third circle in the explanation section did not seem to
help participants understand how the combined risk circle was derived; instead,
they focused on the fact that the effect probability did not change in the rank-
ing cards:
No matter how much money you spend, the effect's the same.
The effect is the same on all of them, so why should I pay $400 a
year for something my risk of getting an effect from it is the same :
as if I pay nothing?
Moreover, participants' comments also indicated they still did not understand
exposure and effect as separate events or effect as being contingent upon first
being exposed:
Question: Why do you think the risk of effect stays the same and
the risk of exposure changes?
I didn't notice.
'
8-20
-------
Card A
Hazardous Waste Risks
Risk of in Exposure RtsK of an Effect
Figure 8-4. Circles used for probability presentation.
8-21
-------
Obviously everyone exposed won't be harmed, some will, some won't,
but here's one, one out of 360 exposures and 4 out of 360 the risk
of effect, can't understand that, three of them got it that wasn't
even exposed.
Finally, participants infrequently felt that the probabilities were too small.
Rather, they indicated that they didn't perceive enough of a difference between
them to affect their payment decisions:
Obviously we're going to look at how much it costs since there's not
so much difference between the chances of exposure.
In the fourth and fifth rounds the circle cards and accompanying expla-
nations were made even more explicit. The spinners were removed; the title
on the exposure card was expanded to read "What Will Happen; Events Out-
side Your Control"; the effect card was changed to read "What It Means to
You: Your Circumstance When ft Happens"; and the combined risk card was
changed to read "What It Means to You." The examples corresponding to these
cards were changed to read, respectively, "rain tomorrow," "flat tire," "expo-
sure to hazardous wastes"; "walking from the car," "on the interstate," "your
hereditary background"; and "get wet," "flat tire," and "get cancer."
For the exposure to hazardous wastes example, the text on the cards
described exactly the association participants were supposed to make--"expo-
sure to hazardous wastes," "your heredity background," and "get cancer,"
Additionally, each circle card included the ratio of the part of the circle that
was shaded and some explanation to help participants understand what was
being conveyed on each card. The exposure card included the statement
"probability - chance spinner will fall in the shaded part," and the combined
probability card included the statement that "both of the earlier outcomes must
occur."
Besides the circles and examples, an additional card was added to help
participants make the association between the simple risk examples and the
hazardous waste risks. This card, entitled "Hazardous Wastes as a Risk,"
included the same information displayed on the circle but in tabular form.
Added to each example was a column entitiled "How it Might Have Been Antic-
ipated." For "rain" this included "bring an umbrella"; for "flat tire" this in-
cluded "have a spare"; and for "hazardous wastes" this included a question
8-22
-------
mark, it was hoped that using the hazardous waste example along with the
simple risk examples would help participants link the two.
The ranking cards (Cards A through E) were also expanded in this
round. Now, instead of each having two circles (risk of effect and the risk
of exposure), they also included a third circle, combined risk. The risk of
effect circle was also changed to read the "risk of effect if exposed." The
risks of exposure were 1/90, 2/90, 5/90, 10/90, and 20/90, The risk of effect
if exposed was 90/540. Combined risks were 1/540, 2/540, 5/540, 10/540, and
20/540. These cards are included as Figure 8-5. Round 5 cards were slightly
different. Instead of being asked to rank cards, participants were asked to
determine a willingness-to-pay amount. Therefore, only three cards were used,
with risks of exposure of 1/90, 5/90, and 10/90.
Participant comments in these rounds indicated much greater understand-
ing of probability. First, they appeared finally to have understood that the
risk of effect is merely a multiplier;
Question; What about that middle circle? Anybody have some feel-
ings on the meaning or the use of that middle circle?
At that point there's nothing you can do about it.
It's just a multiplier.
It's an arbitrary fact at that point.
They also seemed to be looking at exposure and effects from hazardous wastes
as only being a chance occurrence:
The thing that came across to me was that you were using the cir-
cles to point out that it could be controlled by just chance in the
control of hazardous wastes and the effects on the people would just
be a chance.
tt is important to note, however, that the groups in Rounds 4 and 5 were well
educated arid/or very knowlegable about hazardous wastes.
In the final round of focus groups, where the first draft of the survey
was administered, circles were no longer used in the probability explanation
to explain simple risk. Instead, the card explaining risks in tabular form was
made more explicit. It still included three examples, but each one was ex-
plained more clearly. Circles were still used on the ranking cards and were
exactly the same as in Rounds 4 and 5.
8-23
-------
Card A
Hazardous Waste Risks
Risk. Of Exposure
Risk of Effect
it Exposed
Combined Risk:
Exposure and Elftct
Payment required: J400 per year m hifte prices and taxes
Card 0
Hazardous Waste Risks
Payment required: $225 per year in hiffNf prices and taxes
Card C
Hazardous Waste Risks
Risk of Effect Combined Bisk:
Risk of Exposure if Exposed Exposure and Effeci
Payment .reetuiretl: $ 125 »r yea# m Ngh«r prices and taxes
Figure 8-5, Sample cards {A through C) used for probabii y presentation.
8-24
-------
Participants in this round indicated that they found the simple examples
of risk unnecessary and confusing;
Question: Card 4--examples of risk. What did you think of that?
You are confusing the problem of hazardous waste by introducing
irrelevent examples of risk--like if it rains, are you going to have
a flat tire. That is so remote from what hazardous waste Involves,
it seems like you're trying to put some of these risks in the same
classification [as hazardous wastes].
Thus, the everyday risk examples were eliminated from the probability explan-
ation. This decision seemed counterintuitive to what was expected a priori.
However, participants in each session indicated that the context in which they
think about hazardous waste risks is too different from that in which they
think about simple risks. In addition, the attributes they associate with each
type of risk differ. Simple risks were veiwed as voluntary or controllable
events such as wearing seatbelts to reduce risk of death in a car accident.
Hazardous wastes risks, on the other hand, were seen as involuntary and un-
controllable. Instead of everyday examples, very explicit explanations using
local or well-known hazardous waste incidents were used to illustrate probabili-
ties of exposure and effect.
The main criticism surrounding the probability explanation was its length.
Some participants indicated that their minds were wandering by the time the
probability of effect was explained. In fact, those who did not understand
the concept seemed to stop listening right after the first circle was described.
However, those who had some knowledge of probability seemed to listen more
intently. This, is evident in the following example, in which one participant
i| able to explain what is being said to another;
| I still can't in my mind figure out how this is the combined risk,
|
j Two percentages. You have half of a quarter times a half is what
| is an eighth and this is a six times a 3 percent times 16 percent is
i the combination that comes out so you take 10 percent of 16 percent
is 1.6 percent or something like that, so that you are getting it
down . . .
: But don't most people react to this because none of us know our
heredity and how we personally are going to be impacted. But this
is an external thing that we can sort of take in.
8-25
-------
You have been told that the middle is the average of all the popula-
tion in that you are generally going to fit into that category, ;
Participants still had difficulty believing that the probabilities used on the
cards were realistic:
Again I wondered where you came up with these. It looked as if it
could be almost arbitrary.
Many helpful suggestions were made by the participants in clarifying
the cards, Most of these surrounded the mathematical representation of prob-
ability, Using percentages was advocated by participants in all groups:
One of the -things is the math that gets you down. Use a percent-
age figure or one out of thousand or hundred thousand, 10 over 30
and 10 over 540,
I would have used ratios. If you went from 1 in 54 to 1 in 10, I
wouldn't use any circles.
They could be converted into percentage relationship. That I could
read.
I kept wondering why you didn't put percentages here. 10/90
doesn't mean anything to me but 11 percent does.
Scientific notation, that we are going to lose most people. Put in
terms of a one-over kind of number (i.e., 1/100,000) as opposed to
ten-to-the-minus number.
Two out of 100,000 or something like that. , . .
One participant also suggested putting more description on the hazardous waste
exposure risk cards:
Why not describe what it is (on the card], i.e., heredity, back-
ground, pathways.
These suggestions were all taken into account when the circle cards were
designed for subsequent survey drafts. The final version of the circle cards
includes three circles entitled "Risk of Exposure," "Risk of Death if Exposed,"
and "Combined Risk: Exposure and Death," Each circle's significance is fur-
ther explained by a caption underneath. The exposure circle is captioned
"Possible Pathways"; the effect circle,, "Heredity and Health"; and the com-
bined risk circle, "Personal Risk." Each circle has a portion that is shaded
to signify chance or probability of risk. Both the percent and ratio of the
8-26
-------
shaded portion of the circle are on the circle card. The actual probabilities
vary since there are several survey versions that will be administered to
respondents. One version of these circle cards is included as Figure 8-8.
In addition, instead of giving payment amounts, some respondents will be
asked to rank payment cards. These cards are identical except that each will
have a title giving the payment amount. This title is also more explicit than
in previous rounds. It includes both a monthly and yearly amount and states
directly that this is in higher prices and taxes.
8.7,3 Perception of Exposure Risk
Requesting that individuals shade portions of empty circles was the first
means used to elicit participants' perceived risks of exposure to hazardous
wastes. However, participants indicated that the circle was not really the best
way of doing this and that they often very arbitrarily selected the portion of
the circle to shade. It became apparent that some kind of benchmark or
anchor was needed to guide their responses.
Risk Ladder
A risk ladder was then used as a visual aid in determining participants'
perceived risks of dying from hazardous waste exposure. In the early rounds,
the risk of dying from exposure to three different kinds of hazardous wastes
was placed on a ladder among the risks of death from other kinds of events.
A copy of this risk ladder is included as Figure 8-7. In this first draft ver-
sion of the ladder, three estimates of hazardous wastes risks from a risk
assessment study were used in an attempt to determine how respondents would
react to this {and other) information. The ladder was based roughly on the
number of people who die annual ly from various causes or activities. Partici-
pants in general seemed comfortable with the ladder as a graphical representa-
tion :
I think we're all used to seeing things represented in graphs like
these and that it's easier than to start comparing circles.
They were, however, very sensitive to the other events on the ladder. For
example, "eating peanut butter," one of these other events, was brought up
for discussion in each group. Participants were also disturbed by the proba-
bilities used in association with each event and in most cases were reluctant
8-27
-------
Card A 1
Risk of Death Combined Risk:
Risk Of Exposure it Exposed Exposure and Death
Pathways and Health Risk
Card C-l
Risk, of Exposure
Risk of Death
if Exposed
Possible
Pathways
Hereditv
and Health
Combined Risk:
Exposure and Death
Figure 8-6. Two cards {A-l and C-1) with final format.
8 28
-------
Risk Ladder:
Comparing Risks of Death
Smoking1 orw pack
of cigaftRM a day
(500)
MoWfCVChrv?
(200)
Esthng
14)
Ooving • <
(171
(smtu buRar
X-ray* for
(11
Using nccfwm
1.2)
Hazardous wiati *2
'80, tricMOFMCtTytana)
#1
(SO. banzww)
LmAarru
(81
Bun ow iff c*
ISI
H«a«jew wast# #3
(2. BwMecwKhitnat
Tornado
!.2)
Ughtning
(.11
Figure 8-7. Initial risk ladder including exposure to three kinds
of hazardous wast® among risks from other events.
8-?9
-------
to believe they were accurate numbers. They indicated that if the team were
going to try to use these numbers as true probability occurrences they ought
to include at the least a source and some explanation of what they were based
on:
I never took it as an accurate measure of what the probabilities
were. If I were to take it as an accurate number, I'd have to know
what you meant by hazardous wastes # 2, where #1 set that expo-
sure and what does that mean. I just took it as a general idea that
we are exposed to a hazardous waste generates these possibilities
rather than to graphically represent possibilities of it occurring.
Additionally, some participants felt the ladder was misleading because it wasn't
drawn to scale:
For true representation, don't you need to put a broken scale on it?
In the next round the same ladder was used, but this time the exposure
risks to hazardous wastes were removed. Participants were asked to place
their perceived risk of dying from hazardous waste exposure on the ladder.
By and large, participants were able to perform this task, but their comments
indicted they had the same misgivings with the ladder as in the previous
rounds:
This is a realty misleading risk ladder. Your rates are not accur-
ate . They're not age specific. The data is just not accurate.
You're asking an individual for a certain age and this is just not
accurate .for an individual of that age. . . . The way you're trying
to ask your questions, you can't extrapolate from death data for
the whole population very accurately and then ask individuals where
you put yourself on here.
In the final round, when a draft of the questionnaire was administered,
the ladder was changed substantially. This ladder included occupational risks
on one side and risks of dying from various events on the other. The prob-
abilities were removed from each event, and each portion of the ladder was
shaded differently. There was a break between each of these shaded portions
on the ladder to give it the appearance of being more to scale, A copy of
this ladder is included as Figure 8-8, tn addition, a second card was included
that attempted to tie the ladder to the risk circles that had been previously
used to explain probability. This card had both a ladder and circles on it.
The ladder had just three events on it of high, low, and medium death risks.
8-30
-------
Card 5
Risk laddar: Comparing Risks of Death
Stuntman
Truck driver
Sttilwofksr
-~r
Poitea Ofln.fr
Smoking Ons Pack
of Cigar»rtej a Cay
Skydivir>9
Horn* Accident , '<:•**
Car Accident
Horn# Fir*
Poisoning
Flood
Figure 8-8. Revised risk ladder separating occupational risks
from other events and introducing breaks in ladder.
8-31
-------
Next to each event was a circle partially shaded to indicate probability of death
from that event. This card (Card 6) is included as Figure 8-9. Participants
in the first session in this round indicated this card was not helpful in mak-
ing a transition between circles and the ladder. In fact, the card confused
them:
Question; Did card No. 6 help make a transition between the ladder
and the circles?
Pointless.
If you can't keep it maintained to a 100 times for all three, it's
meaningless.
This card was eliminated from subsequent sessions in this round.
Participants in this round had both graphical and conceptual suggestions
for making the ladder and task clearer. The graphical comments revolved
around shading and putting the events more to scale:
Question: What did you think of the risk ladder? Was it helpful?
Not helpful? What kind of impression did you get out of
it?
If you did the graph in a different format it might become a little
clearer to more people. The gradation and shading are a little trou-
blesome at first. There is not a great distinction between the grada-
tion that one notices the distinction until you go back and study it.
The arrows going in two directions rather than one.
The breaks are not clear. If you're working with hard numbers,
it's easier to see and to integrate it . . , to try to figure out how
much space there is between steelworker and car accident, you're
just left to your imagination. It could be a little or a lot; the per-
son just has no idea.
I had a question when you explained the ladder. The breaks in the
ladder appear to indicate that this is a long ladder. Is there a big
gap between smoking one pack a day and a stuntman or are they
right on top of each other? That is something that isn't clear. I
think it would help if you couid somehow or other indicate that--
maybe on a numerical scale—because then you wouldn't be con-
strained by the size of the page or whatever else.
One of the difficulties is the way the break comes across. Cigarette
smoking is at the top of the break and if there had been a wider
break you would see it's not in the same class as stuntman.
8-32
-------
Card 6
Risk Ladder: Comparing Risks of Death Combined Risk Circle:
Comparing Bisks of Death
Smoking One Pick
of Cigarettu a Day
Car Accidant
Flood
Magnified Viaw
{100 times largar than
actual si tea)
Magnified View
(100 timas largar than
actual sliettl
Magnified V>ew
(10,000 tim«i larger than
actual siicai
Figure 8-9, Card attempting to tie risk ladder
to probability circles.
8-33
-------
In addition, participant comments indicated that the examples made it dif-
ficult for them to assess where on the ladder they should place their perceived
risks of death. For one participant, all the risks were accidental except smok-
ing . Because participants did not see death from hazardous wastes as acci-
dental , they tended to put their perceived death risks closer to smoking and
thus higher on the ladder than they really felt was accurate:
Do you mean the risk of premature death? Because all of these are
by accident except for smoking. The rest are premature death due
to some kind of accident.
it's hard to relate the risk [of death] from hazardous waste (with
the other examples of risks of death] because it's more like the cig-
arettes than all of these other things.
Comparing [hazardous wastes] to all these accidental deaths made me
keep pulling it up the ladder.
You could compare it to smoking a pack of cigarettes a day. The
problem is there is nothing else like that on here.
Other participants didn't feel there were enough examples on the bottom of
the ladder:
These seem to be all very high risk ... at least from home fire
up. I would have liked to have had something at the other end of
the scale. In between flood and poisoning because everything else
seemed too high up.
Many participants had difficulty in relating to the types of occupations used
as examples:
But the skydiving and stuntman are so remote from the average per-
son's experiences, maybe you ought to have death of a heart attack
at age 60, something that people relate to.
The women in particular thought there were too many male dominated occupa-
tions:
The occupations are not ones I related to very easily. They tend
to be more male occupations.
Most participants wanted to see some indication of the probability of dying from
the events listed on the ladder:
When I saw cigarettes way up there, I didn't think it very believ-
able. I didn't bel ieve it— it looked like someone just did it.
Shouldn't it say based on insurance statistics or something?
8-34
-------
Everything else in Che thing is done with numbers. You might very
well, since all these are different levels, just put numbers along
side of them, it might be easier.
In the group where the participants were asked to place their perceived occu-
pational risks on the ladder, it became apparent that more examples that pro-
fessionals could relate to were needed;
I can't relate to your probabilities, I work in an office and the
worst thing that is going to happen to me is hypertension and I have
a heart attack.
If they doubled the exposure from those CRT terminals. If it radi-
ated more stuff, that's in an office,
:
I couldn't even get on the first rung of the ladder. It's zero.
1 might have a problem getting to and from work; that's a problem.
I did have a little difficulty identifying, say, with the sky diving,
for example, or with drunk driving.
The older group of participants had the most difficulty understanding the
exercise. They indicated more text around the ladder would clarify the task:
Question: Does anyone have any reaction to the risk-ladder card?
Did yo« find it helpful, confusing?
Confusing.
t just didn't understand it. A graph like that says nothing to me.
You have to put it in words in a paragraph.
Finally, some participants suggested ways to reword the question to make it
*
clearer. The comments indicated our question had to provide more specific
details on the situation they were evaluating:
It might have helped us if you said "premature" before "death,"
What about age. Some people might not care if it means they are
going to live to 70-75.
Whether it's an actual exposure to hazardous waste or what is your
potential of being exposed to hazardous waste. If you have an
actual chemical spill in your town, that's different than what you
think your chances are of being exposed.
Participants' suggestions were taken into account to construct a ladder for
the final version of the survey that is quite different from that used in the
8-35
-------
focus groups. Each segment of the ladder is a different color to show more
clearly the breaks that signify changes from one probability level to another.
The events are no longer listed on two sides of the ladder but down the mid-
dle. Risks of death from more common professional occupations are included,
such as insurance agent, engineer, or banker. Probabilities of each event
have been included, not in fraction form but as the number of persons out of
100,000 who will die every year. An uncolored copy (reduced in size) of the
ladder is included as Figure 8-10.
I n addition, the survey script explaining the ladder is much more expi ic-
it. It points out the breaks in the ladder and what they signify, documents
the probabilities, and explains them--e.g., 11 out of every 100,000 people will
die from home accidents each year. The development of the risk ladder clearly
demonstrates the effectiveness of using focus groups to develop a contingent
valuation questionnaire. Specific, immediate feedback enabled the research
team to alter the ladder to resolve confusions.
8.7.4 Summary
Although some of the information gathered during the focus group ses-
sions could have been obtained as easily in a one-on-one pretest situation,
not all of it could. For example, in many cases the group envi ronment stimu-
lated participants to think of and verbalize ideas they probably would not have
expressed in a one-on-one interview. In addition, the focus groups conducted
in Boston allowed the questionnaire materials to be evaluated using households
comparable to those in the survey population and thus provided access to spe-
cific local details that might have affected the survey results. Furthermore,
the focus groups allowed the targeting of a specific group composed of people
from a variety of educational backgrounds and income levels that had experi-
enced a hazardous waste incident. This was particularly crucial with sucn a
complex topic.
The focus groups did prove less successful in one area. The transition
from the oral to a written instrument was not smooth. This was apparent in
the difficulty participants had answering the valuation question when the first
draft of the survey was administered in the last round of focus groups. This
difficulty occurred even though participants had little difficulty with the same
question in the previous round of focus groups, where a less formal presenta-
8-36
-------
Card 5
Risk Ladder; Comparing Annual Risks of Death
IS
Sturuman
2.000 of 100.000
14
13
12
11
10
09
CM
87'
08
OS
04
03
02
Snwlcsf*
Skvdiwr
Shipbuildw/TVuckdnvSr
Mo»n«butid«r
PoliCT Officer
.. tfwtM
Horn# Acud.nl
—J
01
Sartntf/EngiMtf
insurants Amm
Jlan&EoL
jAirpianji
Poiaonmq
Boatf
300 of 100,000
200 of 100,000
99 of 100,000
77 of 100.000
47 of 100,000
22 at 100.000
1S.1 of 100,000
11 ot 100.000
S Of T00.000
4 of 100,000
2.8 of 100.000
0.8 of 100,000
Q.8 of 100,000
.08 of 100.000
"At taMt una pick pv 4w.
Figure 8-10. Final version of risk ladder incorporating suggestions
from participants.
8-37
-------
tion was used. Therefore, whereas focus groups are extremely valuable in
the testing of ideas and techniques and in constructing a first draft of a sur-
vey questionnaire, they will not serve as a substitute for a pretest.
Finally, although the advantage of hindsight now suggests, perhaps, that
some of the 19 sessions conducted during this research could have been elimi-
nated by additional planning, the experimental nature of using focus groups
in a major contingent valuation survey questionnaire development effort and
the desire to learn as much as possible about how people feel, think, and talk
about risks from hazardous wastes were compelling reasons to conduct a large--
rather than an optimal (i.e., smaller) — number of sessions.
8.3 PRETEST OF CONTINGENT VALUATION SURVEY QUESTIONNAIRE
After a draft version of the survey questionnaire was administered during
the final round of focus groups, the comments of the focus group partici-
pants—both on content and on presentation of in forma tion--we re analyzed and
then incorporated into a second draft. However, although it was judged to
have the appropriate structure, sequencing, content, and presentation, this
draft was not considered ready for data collection because it had not been
administered under actual field conditions. To minimize the occurrence of un-
expected problems during data collection, therefore, we elected to conduct a
pretest of the questionnaire using trained interviewers and a number of pretest
interviews.
To prepare for the fieldtest, or pretest, two interviewers were trained
in a day-long session at the Research Triangle Institute (RTI) in Research
Triangle Park, North Carolina. Subsequently, one of these interviewers, who
later supervised the data collection on a day-to-day basis in the field, trained
two professional interviewers in the Boston area to help collect the fieldtest
data. For the pretest, a total of four interviewers completed 45 interviews in
two locations: suburban Boston, Massachusetts, and the Research Triangle
area of North Carolina. The latter area was chosen to take advantage of the
services of an interviewer who had prior contingent valuation survey experi-
ence and who had demonstrated an uncanny knack for not only identifying
trouble spots but also suggesting solutions. Nine of the interviews were com-
pleted in the Research Triangle area and 38 in suburban Boston. The inter-
views were divided about equally between the direct question and ranking ver-
8-38
-------
sions. The interviewers used no specific criteria to select respondents,
although the project team did request that they interview respondents from
several socioeconomic groups.
To evaluate the effectiveness of the questionnaire, the project team con-
ducted two half-day debriefing sessions with the interviewers at each location.
The completed questionnaires also were analyzed for general consistency in
responses. The outcome of these efforts was that the questionnaire generally
was on the right track but that several trouble spots needed improvement.
Generally, the interviewers were able to identify these areas and to indicate
the kinds of problems either they or the respondents had experienced. Thus,
the insights obtained from the pretest dealt almost exclusively with the work-
ability of the questionnaire. The pretest samples were too small and nonrandom
to yield any insights into the potential variances in willingness to pay amounts
in the actual survey. In contrast, Mitchell and Carson [1984] found that their
willingness-to-pay bids from a 100-interview pretest had variances almost iden-
tical to these in their full survey of 800. Information about variances is criti-
cal for judging the adequacy of the statistical power for the planned sample
size but was beyond the capability of our pretest.
The pretest suggested that the main trouble spots in the questionnaire
involved the overall language and the explanations at certain points. Specific-
ally, the pretest questionnaire sounded too much like an interviewer reading
and not enough like an interviewer talking. It simply was too formal and not
conversational. To illustrate the value of the pretest in making this point,
the following excerpts compare the pretest version with the final questionnaire.
However, it should be noted that the final version reflects the efforts from
other revisions, including those from the videotaped interviews and suggestions
from outside reviewers:
Pretest version
; Throughout life there are chances that people may die from many
different causes. Every day of our lives there is a chance that we
may die from some accident on the job, at home, or somewhere else.
There is also the chance that we may die from some long-term illness
or disease or we may die suddenly from some health problem. On
the other hand, there is a chance that we may fully live out our
¦ lives and die of natural causes. Some common risks of death are
shown on this risk ladder (see Figure 8-10).
8-39
-------
Final version
Throughout our lives there are many different risks of dying.
There is a risk or chance we may die from an accident or some long-
term illness, or we may die suddenly from some health problem.
The pretest experiences also indicated several problems in the introduction
to the risk ladder. Specifically, the example used to illustrate how the
respondent was to use the (adder was misleading and the importance of the
different sections was not emphasized:
Pretest version
The ladder will help you compare different risks of death. Notice
that the ladder is divided into six sections to show that the differ-
ences in risk levels are quite large between sections. Each section
shows the relative sizes of the risks of dying during any year of a
person's lifetime based on national averages. Beside each cause of
death there are figures that show the number of people who die each
year from that cause. For example, the risks to stuntmen show that
in any year 2,000 out of every 100,000 stuntmen will die from an
accident on the job.
Final version
This ladder shows the different risks of dying associated with a
variety of common activities, including accidents, habits, hobbies,
illnesses, natural disasters, and job accidents. The numbers on
the right show the risks for each of the activities listed. The lad-
der displays these risks from low to high so you can easily compare
them. The two types of risks shown and those based on some of
the people and those based on all of the people in the United States.
For example, numbers shown for occupations, skydiver, and smoker-
are based only on people in these activities. This means, for in-
stance, that during the next year 47 of every 100,000 homebuilders
in the United States will die from an on-the-job accident. However,
the numbers shown for the remaining risks are based on averages
for all people in the United States. This means, for instance, that
during the next year, 77 out of 100,000 people in this country wilI
die from a stroke. Notice also that there are breaks between the
five parts of the ladder to show that the difference in risk levels is
quite large between each part.*
The explanation of the risk circles was the area most frequently recom-
mended for major revisions. Interviewers found the explanation in the pretest
version both redundant and confusing:
*Another important change was also made in response to suggestions ' rom
A. L. Nichols and several other reviewers from the U.S. Environmental Protec-
tion Agency (EPA). They suggested that all risks, except for the occupational
risks, be put on a consistent basis. The pretest version had some risks that
appl ied only to people who presently experienced the health condition.
8-40
-------
Pretest, version
Another way of thinking about hazardous wastes as involving risk
is with this card (HAND RESPONDENT HAZARDOUS WASTE RISKS
CARD A, WITHOUT DOLLAR AMOUNT). It uses circles to stand
for two types of hazardous waste risk that we want you to think
about: the first circle, which shows the risk or chance that you
(or a member of your household) would be exposed to hazardous
wastes. By exposed, 1 mean touching, breathing, eating, or drink-
ing a large enough amount of a hazardous wast# over a period of
time so that it could harm the health of whoever is exposed. Expo-
sure through the pathways we have discussed could be a brief, one-
time exposure, or it could be over months or years. The importance
of the second circle is that even if a person is exposed, there is
another and different risk or chance that he would develop a health
problem and die. With many of the kinds of health problems that
could be caused by hazardous wastes, it might be 10 to 30 years
before a person would know that he was seriously ill and die. The
third circle combines the two types of risks into risks to a person.
Final version
Another way to think about hazardous wastes and risk is with this
card. It uses circles to stand for two different kinds of risks we
face from hazardous waste.
Pretest version
The middle circle on Card A stands for the second type of hazardous
I waste risk--the chance of a harmful health effect after being ex-
posed . This risk means that even if you are exposed, there is a
: chance, not a certainty, that you will be harmed. For example, if
one person catches a cold at home or at work, everyone around will
not get sick. Some people are healthier or have better resistance.
The same idea is true for hazardous wastes. Whether or not you
are actually harmed is based on your physical makeup--your heredity
and your overall health. Looking at both of these circles, you can't
be harmed by hazardous wastes if you are never exposed to them,
j You would never have to spin the pointer in the middle circle as
long as the pointer on the first circle (POINT TO FIRST CIRCLE)
never landed in the darkened area.
Final version
The importance of the middle circle is that it stands for the second,
and different, type of hazardous waste risk--the chance of dying
after being exposed. This means that even if you're exposed,
there's a separate chance~~not a certainty--that you would die. For
example, some people are healthier or have better resistance.
Whether or not you're actually harmed is based upon your physical
makeup, heredity, and overall health. An important thing to remem-
ber about the first two circles is that you would never have to spin
the pointer on the second circle as long as the pointer on the first
circle never landed in the blackened area. In other words, there's
no chance you would die from the effects of hazardous wastes if
you're never exposed to them.
8-41
-------
The interviewers also pointed out that respondents had trouble with the tran-
sition between hypothetical scenarios. It was necessary to repeat entire sec-
tions because the respondent was unclear about the ground rules. The tran-
sition at Section G of the questionnaire (willingness to pay to avoid an increase
in risk) was especially troublesome because respondents frequently thought
their bids in the previous question also applied to this one:
Pretest version
Now let's consider a completely different situation.
Final version
Now let's consider a completely different situation. That is, your
dollar amounts and answers to previous questions are not carried
over to this one.
The pretest also confirmed the effectiveness of the focus groups in eval-
uating the visual aids used in the interview. With one exception, the payment
vehicle card, the interviewers felt like these visual aids worked well. The
payment vehicle card subsequently was revised and the interviewers (in the
final field survey) confirmed that the changes had remedied the problems with
the payment vehicle card.
In summary, the pretest and the subsequent discussions with the inter-
viewers provided valuable information on the workability of the questionnaire.
These steps led to major revisions that clarified the exposition. They also
clearly demonstrated the importance of how a questionnaire "sounds." To be
effective, good exposition is not enough; the questionnaire also must sound
appropriate when spoken .
I n addition, the project team felt that there was little difference in the
information obtained in the suburban Boston and Research Triangle area pre-
tests. That was encouraging for three reasons: First, the local pretest was
less expensive than the onsite pretest because there were no travel costs for
training or debriefing. Second, with the interviewer working only in the local
area, it was easier for the project team to communicate on a more frequent
basis. Third, the lack of any significant differences also implied that the vid-
eotape interviews could be done in the local area at considerable cost savings
with probably only minor losses in information.
Finally, caution is required in drawing a general conclusion from our ex-
perience that a local pretest can substitute for one conducted at the actual
8-42
-------
survey " location. One difficulty is that although the context of our hazardous
waste valuation scenario was for a specific site, the actual location could have
applied to any town. The critical question to be answered is whether there
are any reasons to expect that respondents in different areas would react dif-
ferently to the framing of the questionnaire. This does not suggest that they
would necessarily have the same willingness to pay. Indeed, we would expect
differences based on income and other relevant explanatory variables. How-
ever, it does imply that the same behavioral model applied -to two populations
would fit each equally well. Even with hindsight, It would seem desirable to
perform the on site pretest because it provided relatively low cost insurance for
avoiding major problems in the actual survey.
3.9 VIDEOTAPED INTERVIEWS
To supplement the field pretest, ten one-on-one videotaped interviews
were also conducted with members of the RTI staff. As the final stage of the
questionnaire development process, these videotaped interviews provided infor-
mation necessary to evaluate additional aspects of the final questionnaire's
workability. They were especially helpful in identifying the various verbal
and visual cues that respondents used to develop their answers to specific
questions.
In evaluating whether or not the questionnaire "worked," the videotaped
interview sessions focused on five key elements:
The respondent's perceptions of the questionnaire's framing--
e.g., the hypothetical commodity and the payment vehicle.
The usefulness of the visual materials as aids in the framing
process,
The effectiveness of the risk circles in communicating very small
probabilities.
The logical progression of the questionnaire.
The sound of the questionnaire's language.
Ten separate interviews were videotaped with RTI employees in a conference
room at the Institute. The employees included two maintenance workers, two
data entry workers, a mid-level statistician, an electrician, a painter, a car-
penter, and two secretaries. The interviews were divided equally between
8-43
-------
men and women. Respondents also were chosen to represent a wide range of
ages and educational levels.
The videotape camera was placed in one location and operated automatically
eliminating the need for a camera operator. One project team member observed
the session while another conducted both the interview and the subsequent
discussion. It was explained that the purpose of the session was to evaluate
the questionnaire, that there were no right or wrong answers, and that parti-
cipants were to respond the same as if they were in their own living rooms.
No one-way mirrors were used to conceal the observer. However, the partici-
pants seemed unaffected by the presence of the observer or the camera after
the initial explanation of the purposes of the session.
Although it is difficult to isolate the specific changes that resulted exclu-
sively from the videotape sessions, several general conclusions can be high-
lighted from the videotapings based both on the observations of the interviews
and on the discussions with respondents. For example, in their explanations
of how they formed their willingness to pay bids, almost all respondents men-
tioned one key feature: their monthly income and their present expenses.
The respondents cfearly used this as their common anchoring point. Although
the bids varied quite substantially, the first thing each person mentioned in
describing his thought process was his budget constraint. It seemed that tne
use of monthly amounts rather than annual amounts made it easier for h'm to
consider his budget constraint. If the budget constraint as the primary
anchor were common to contingent valuation surveys, it may help to explain,
at least in part, why respondents have shown considerable difficulty in devel-
oping their willingness to accept bids (see Knetsch and Sinden [1984], Meyer
f 19791 , and Rowe, d'Arge, and Brookshire [1980]). In the willingness-to-
accept case, they lose the common anchor on which they rely in the willing-
ness-to-pay case. Of course, the difficulty may also in part be due to an
unwillingness to be moral ly responsible for accepting a payment for degradation
of the environment ( see Kahneman [1984]).
The discussions in the videotape sessions also focused on the adequacy of
the framing for the hypotheticat commodity, reductions in the risk of exposure
to hazardous wastes. In particular, the respondents were asked about how
they used the circle cards in relation to the various hypothetical scenarios.
8-44
-------
Some described using the visual relationships between circles, while others
said that they felt more comfortable with the numerical expressions—a finding
consistent with our focus group experience. They understood the link between
the changes in the risks and the proposed regulations in the hypothetical sce-
nario. Some focused on the exposure circles while others used changes in
the combined circle in forming their bid. The majority indicated that the three
separate circles communicated the relationships between exposure, their own
heredity, and the risk of death. The videotape sessions reinforced the earlier
judgment that how the respondents responded to the probability information
will be one of the central questions to be evaluated in the empirical analysis.
Another important use of the videotape sessions was to evaluate the feas-
ibility of using the risk circles to communicate the low probability parts of
the experimental design. In response to suggestions from reviewers, the ex-
perimental design was expanded to include two additional direct question ver-
sions of the questionnaire. One new version had combined risks of exposure
and death ranging from 1/30,000 to 1/150,000 arid the other, risks ranged from
1/80,000 to 1/300,000. These probabilities were 100 times smaller than the
risk levels that previously had been evaluated with the risk circles. About
half of the total videotape sessions consisted of the lower probability cases.
The general conclusion was that the respondents seemed to be able to use the
risk circles equally well to see the reductions due to the regulations. In ef-
fect, the videotape sessions provided low-cost insurance that the additional
design points were workable before more resources were committed to collect
data from these additional designs.
The videotape sessions also indicated that the improved introduction to
the risk ladder (noted in Section 8.8) made it easier for respondents to use
the ladder in expressing their perceived risk of dying from various causes,
including exposure to hazardous wastes. The respondent descriptions of how
they used the ladder reinforced the focus group finding that some preferred
the numerical expressions while others used the various anchors of other types
of risk. Each of the different kinds of risks--job risks, health risks, risks
from different activities, and risks from natural hazards — was mentioned by
respondents in their descriptions of how they used the ladder.
The videotape sessions helped to evaluate another important aspect of a
workable questionnaire—its logical progression. In the followup discussions,
8-45
-------
respondents indicated that they felt comfortable with the order of both informa-
tion and questions. They pointed out the importance of the order of informa-
tion on Card 1 that related hazardous wastes and common products. Almost
every person cited some part of this information in their explanation of how
the questionnaire oriented them in thinking about hazardous wastes. They
also felt that the sequence of the risk discussion using the circle card, fol-
lowed by the payment vehicle and then the hypothetical situation seemed
straightforward. Several noted that the explanations were longer than they
needed (e.g., the circle cards) but others felt that the additional information
helped them.
Finally, the videotape sessions afforded the opportunity to listen to the
questionnaire to evaluate its sound. After the pretest, the interviewers had
stressed the importance of having the questionnaire sound like an interviewer
talking and not simply reading. By observing and listening to the session it
was easy to evaluate the sound of different questionnaire sections as they were
administered. The videotape also enabled the team member conducting :he
interview to replay these same sections and elicit the respondent comments on
what caused a puzzled expression or some other kind of response. In listening
to the interview, some words or vagueness had a jarring effect and prompted
the search for simple and/or more concrete words to replace technical or vague
language. The repetition of interviews by a team member also led to improved
interviewer instructions on how to use the visual aids to make the question-
naire more interactive.
8.10 THE QUESTIONNAIRE DEVELOPMENT PROCESS: REFLECTIONS
AND SUGGESTIONS FOR IMPROVEMENT
While the actual process of developing the questionnaire evolved over a
period of about 1 year and had to respond to other objectives besides the pri-
mary one, the passage of time, the advantages of hindsight, and some missteps
have all yielded some useful impressions about the overall process. Genera I v.
focus groups, field pretests, and the videotaped interviews should be viewed
as complements rather than substitutes. Each seemed to offer some advantages
relative to the other but there were also some disadvantages. The focus
groups were especially effective ;n getting a general sense of people's knowl-
edge and perceptions of hazardous wastes. This was especially useful fc- - his
8-46
-------
application, since very little information was available in the literature on how
to meaningfully present hazardous waste risks in a household survey.* On
the other hand, the pretest was a better indicator of trouble spots in the
questionnaire due to either logic or language. The pretest also focused atten-
tion on the administration of the questionnaire and the importance of the ver-
balized form or "sound" of each question. The videotape sessions proved very
effective in evaluating whether or not revisions aided either "sound" or work-
ability. Both the focus groups and videotape sessions were- excellent for get-
ting people to explain their thought processes and for determining the effec-
tiveness of the visual materials in aiding the information processing. In addi-
tion, caution is'required in using the pretest for the purpose of knowing what
the respondent was thinking. This information came from experienced observ-
ers (the interviewers) rather than the respondent. This shortcoming can be
minimized by encouraging the interviewers to seek out the respondent's reac-
tions rather than relying exclusively on their impressions, but the possibility
of inaccurate filtering still remains.
The complementary nature of focus groups, pretests, and videotaping
implies that a blend of the three can be every effective tools in dealing with
complex environmental commodities. However, better integration iikely would
enhance their complementarity. After the first two rounds of focus groups,
additional time to prepare a written draft of the questionnaire likely would have
permitted the more rapid development of a final questionnaire. Using an early
draft questionnaire in several videotape sessions perhaps could have replaced
at least one round of focus groups. This change would have shortened the
time involved in planning and the logistics of focus group sessions and allowed
more time for the team to work on the questionnaire itself. The videotape ses-
sions, supplemented by simply reading the questionnaire into a tape recorder
as revisions are attempted, likely would have enhanced the way the pretest
version sounded.
Following the videotaping and subsequent revisions, a round of focus
groups to administer the draft questionnaire to participants from the survey
; *Recall the earlier study (Burness et al. [1983]) treated the problem as
one of valuing a regulation with general uncertainty as to the exact nature of
hazardous waste risks.
8-47
-------
area would provide valuable feedback on the respondents thought processes
as well as the effectiveness of the questionnaire and visual aids. However,
the cost differential between local and on site pretesting could be kept relatively
small by foregoing sn-person training and debriefing. Both activities could
be done by telephone supplemented with programmed training. These two sub-
stitutions could enable pretests both onsite and locally for about the same cost
as one full-scale onsite effort with expensive personal training. However, the
in-person training supplemented with practice interviews and intensive discus-
sions proved critical to the success of our actual field survey, since the cost
of mistakes could have been much higher.
In summary, the process of questionnaire development could have been
enhanced by better integration of focus groups, pretests, and videotape inter-
views. Focus groups seem to diminish in effectiveness after two or three ses-
sions. They are most useful with longer periods of time between sessions and
a corresponding larger amount of time for better formalizing ideas. The sooner
a written draft can be prepared the better. Speaking rather than read ng
even early versions makes a major difference in the way they sound. Video-
taping is a fast, relatively inexpensive way to explore how the respondents
are using different parts of the questionnaire. Finally, field pretests are still
useful in simulating actual field conditions.
8-48
-------
CHAPTER 9
SAMPLING PLAN AND SURVEY PROCEDURES
3.1 INTRODUCTION
| This chapter summarizes the sampling plan and the survey procedures
used to gather the information required by the experimental design. Specific-
ally, Section 9.2 defines the target population, Section 9.3 gives a brief over-
view of the sampling plan and its relationship to the experimental design, and
Section 3.4 describes how the survey questionnaire was administered to the
target population, including discussions of interviewer training, quality control,
data collection, and interviewer debriefing procedures. Section 9,5 concludes
the chapter with a brief summary of its main points.
9.2 THE TARGET POPULATION
As noted in Chapter 8, the experimental design for the survey called for
approximately 800 completed interviews with economic decisionmakers in house-
holds in suburban Boston — specifically, the Boston, Massachusetts, standard
metropolitan statistical area (SMSA), exclusive of the City of Boston. Figure 3-
1 shows this target geographic area and indicates, in the shaded portions, the
location of the areas eventually selected for the survey interviews that composed
the sample.
The experimental design required that all survey respondents be economic
decisionmakers, not just a randomly selected member of the household. There-
fore, the target population actually consisted only of persons who made primary
economic decisions for groups of household members residing in the target geo-
graphic area. These groups of household members, called economic reporting
units, consisted of the members of a household who act as a single economic
entity to make expenditures in three categories--food, housing, and other ex-
penses, Representing these groups of household members (or economic re-
porting units), the economic decisionmaker was the single individual most re-
3-1
-------
S ====l^ Mil*!
ue«t»o
Note: Shaded areas indicate locations selected for survey interviews.
Figure 9-1. Map of survey area.
9-2
-------
¦sponsible for deciding how, when, and from whom to buy goods rn these three
categories, This target population was selected as the group most relevant to
the overall project objective of estimating the benefits of the risk reductions
that accompany hazardous waste management regulations.
Whenever field interviewers could not identify a single individual decision-
maker for an economic reporting unit, they used a random number list to select
one at random. In addition, white the household itself comprised the economic
reporting unit for most of the sample, this was not the case for all households in
the sample. For example, the project team considered all related members of a
household to comprise a single economic reporting unit, classifying household
members not related to anyone else in the household as separate economic
reporting units.
9.3 THE SAMPLING PLAN
This section briefly describes this sampling plan and how and why it
evolved as it did. It considers, first, the overall two-stage sample design and,
second, the role of the experimental design.
9.3.1 Overview
Drawing on the interview completion rates from in-person surveys of simi-
lar size and scope, a stratified, two-stage sampling plan was designed to select
enough eligible and willing respondents to achieve the goal of approximately 600
in-person interviews required by the experimental design. The first stage, or
primary sample, was composed of 100 U.S. Census blocks or block clusters
selected from two geographic strata—the town of Acton, Massachusetts, and the
balance of the suburban Boston area, exclusive of the city of Boston. To
accommodate the experimental design and the population distribution in the
target geographic area, 20 of these Census blocks/block clusters were selected
from Acton, and 80 were selected from the remaining portion of the suburban
Boston area.
We had two interrelated reasons for selecting so many Census block/block
clusters in the Town of Acton. First, the town had recently experienced a
number of incidents involving hazardous wastes, including a contamination of
two municipal drinking water wells, that resulted in a substantial amount of
information about hazardous wastes being disseminated in the community. (See
9-3
-------
Chapter 10 for a more detailed discussion of hazardous wastes and Acton.)
Because this information could have an important impact on people's valuations
of risk reductions, we oversampled the population in Acton so we could compare
the valuations of Acton residents with those of the rest of the target population.
Second, oversamplmg in Acton also helped us meet the objective of comparing
the results of our study and those of Harrison [1984], Specifically, because
Harrison [1984] used a hedonic property value model (discussed in detail in
Chapter 15) and two other methods (i.e., a risk assessment, and an analysis of
averting costs) to develop policy analyses of alternative regulations of the
disposal of hazardous wastes, including estimates of the benefits for avoiding
exposure to hazardous wastes for homeowners in the Town of Acton, our
oversampling of Acton residents will allow us to compare our survey estimates
with Harrison's.
The second stage, or secondary sample, was derived from the U.S. Census
biocks/block clusters in the first-stage sample by first listing and then
selecting specific housing units in the two target geographic strata. A totai of
915 housing units were listed and selected for the second-stage sample--189
from the Acton stratum and 756 from the remaining portion of the surou-'^an
Boston area. Sample weighting procedures—equal weights within strata, differ-
ent weights between strata — were also developed to help ensure accurate compi-
lation of data from the surveyed population. Appendix C provides more infor-
mation on the listing of housing units within the two strata, and Appendix D
contains a more detailed discussion of the first- and second-stage sampling
procedures.
9.3.2 Experimental Design Considerations
The experimental design raised three important questions for the sampling
procedures used to sample the survey's target population:
How would the des gn be allocated across the sample without
confounding it?
How many sample housing units would be required to achieve
the planned number of observations for each cell in each part
of the experimental design?
How many sample housing units would be required to yield the
desired number of completed interviews?
9-4
-------
In answering the first question, the 24 versions of the questionnaire were
randomly ordered across the entire sample, and this random ordering was repli-
cated in units of 24 across the entire sample. This procedure assured that
each interviewer and each sample housing unit had an equal probability of re-
ceiving any one of the 24 versions of the questionnaire. This randomization
was selected in an attempt to reduce the potential confounding of the design
with either the sampling procedures or the procedures used to assign sampling
housing units to specific interviewers.
The answer to the second question involves a tradeoff between the ex-
pected cost of obtaining a completed interview and the number of sample hous-
ing units required to permit reasonably powerful tests of the hypotheses that
were implied by our conceptual analysis, given the elements of the experimental
design. This process also has implications for the precision of our estimates
of option price functions. Trying to anticipate the necessary sample size for
estimating the values associated with changes in hazardous waste risks is com-
plicated by the lack of previous studies and the potential for nonlinearities in
these tradeoffs. Given our uncertainty over the precise forms of some of the
functional forms and final tests for the models we estimated, and given our
desire to test a variety of hypotheses, to estimate payment (option price) func-
tions, and to realize other estimation objectives simultaneously, we did not at-
tempt to derive the sample design allocations through a constrained optimization
problem {e.g., see Conlisk arid Watts [1979] and Aigner [1979]. A flexible
full-factorial design was selected for part of the direct-question design, with
separate blocks to consider the effects of low probability scenarios, and an
independent full factorial design was selected for the contingent ranking com-
ponent of the design. We allocated a somewhat larger number of observations
to the contingent ranking design points in an effort to permit (within the limits
of the budget for the survey) separate indirect utility functions to be estimated
for each design point. In all cases, however, the sample sizes exceed conven-
t onal rules of thumb for testing of hypothesis concerning means and are more
than adequate (given the experimental design) for multivariate analysis.
Figure 9-2 is a matrix showing both the planned sample sizes and the
number of observations obtained for each element of the experimental design,
including direct question and contingent ranking question formats for valuing
.
risk reductions and two versions of the direct question format for valuing the
9-5
-------
Value of Reductions in Risk
Direct Question Formal (0)
Contingent Ranking Question Formal (R)
Reduction* in fish
otexposure
Risk of death, if exposed
Levels oi r|$k
Amount of monthly payment
Vector
m&k oi
exposure
Risk of death,
if exposed
Vector A
Vector B
Vector
From To
mo
1/20
1/100
1/200
P A
P A
P A
P A
SO $20 $55 $150
$-20 $5 $40 $80
I
1/5 1/10
1/10 1/25
45 42
45 34
45 46
45 35
—
I
1/10
1/20
1/50
1/100
1/10
P A
P A
60 57
60 59
II
1/10 1/20
1/20 1/50
45 47
45 31
45 46
45 33
111
1 1
1 1
45 48
45 36
45 35
45 29
—
_
II
1/20
1/30
1/GO
1/150
1/10
60 56
60 55
IV
1/300 1/600
1/600 1/1500
—
45 53
45 32
45 4?
45 28
Town Council-Approved Risk increases
Federal Governmenl-A Mowed Risk Increases
Reductions In risk
of exposure
Risk of death, If exposed
Increases In risk
of exposure
Risk of death, 11 exposed
To
From
1/10
1/20
1/100
1/200
To
From
t/10
1/20
1/100
1/200
P
A
P
A
P A
P A
t/25
1/5
22
IS
22
20
_
1/25
1/5
23
20
23
21
-
-
1/50
1/10
83
m
22
22
-
1/50
1/10
62
73
23
16
1/60
1/20
60
47
-
-
-
1/60
1/20
60
44
-
-
-
1/150
1/30
22
21
22
tfl
-
-
1/150
1/30
23
22
23
13
1/1500
1/300
-
-
22 21
22 10
1/1500
1/300
-
-
23 26
23 ie
3 Although thrs par! oF the design used only the direct question format, bolh the ranking and effect question versions that correspond to the Part A design are
identified lo reflect the interrelationship between bolh pans of the design.
There are I wo observations for thss design pan* because oi overlaps in the probability levels in the Part A ctesign,
P = Planned A = Actual
Figure 9-2. Matrix of planned and actual observations for each cell of the experimental design.
-------
avoidance of risk increases. As shown in the figure, the experimental design
generally called for 45 observations in each of the cells for the direct question
versions arid 80 observations in each of the cells for the contingent ranking
versions. The final sample sizes exceeded the planned sizes for all but the
direct question part of the design, which asked each respondent to provide
values for two separate risk changes. Because some individuals declined to
pay for the second increment, some of the direct question cells had fewer ob-
servations than were planned.
To answer the third question, i.e., to determine the number of sample
housing units required to yield the 600 completed interviews, the results of
previous surveys of similar size and scope were used to develop target inter-
view completion rates. Specifically, a sample size larger than the desired num-
ber of completed interviews was selected because past experience indicated
that interviews would probably not be obtained from every economic reporting
unit included in the sample, For example, some units would be ineligible be-
cause they were vacant; in others, the respondent would refuse to be inter-
viewed, In anticipation of not obtaining interviews for alt units," therefore,
the following anticipated completion rates were used to develop the sample size
required to yield at least 600 completed interviews:
0.32 eligible occupied housing units per prelisted unit
0.92 enumerated housing units per eligible occupied housing
unit
0.75 interviewed economic reporting units per selected economic
reporting unit.
"able 9-1 shows the sample sizes developed for the two target strata using
these interview completion rates
TABLE 9-1. SAMPLE SIZES
Strata
Completed
interviews required
Sample a
housing units
Acton
Balance of suburban Boston
Total
120
480
600
945
189
756
'Computed as completed interviews/(0.92)(0.92)(0. 75) .
9-7
'
-------
9.4 SURVEY ADMINISTRATION
Once the target population had been identified and an appropriate sample
had been' scientifically selected, a set of survey procedures was designed to
fulfill the sampling protocol and to attempt to minimize problems stemming from
the administration of the questionnaire. These procedures provided for the
use of experienced professional interviewers, intensive in-person training of
the interviewers, and close supervision of the entire data collection effort.
In addition, to conclude the survey administration effort, the team also con-
ducted an in-person session to debrief the interviewers about data collection.
This section describes the training methods, highlights the quality control pro-
cedures, summarizes the outcome of the data collection, and concludes with a
review of the information provided by the interviewers in the debriefing.
9.4.1 Interviewer Training
Because interviewer training had played such a critical roie in earlier
contingent valuation surveys (e.g., see Desvousges, Smith, and McGivney
[1983]) and because of the complex nature of the hazardous waste question-
naire, the project team developed a detailed training agenda. This agenda
consisted of preparing a comprehensive manual tailored to the questionnaire, a
home study of the manual, a 2Vday training session, and four to six practice
interviews accompanied by intensive debriefing. All five of these elements
played an important part in helping the interviewers understand not only what
they were supposed to do, but why they were doing it.
The interviewer training manual consisted of eight chapters. Topics in-
cluded a description of the overall research objectives, the sampling protocol,
procedures for securing the interview, general questionnaire administration,
question-by-question specifications with detailed explanations and examples,
and general administrative procedures. Interviewers studied the manual prior
to the training session and referred to it throughout the data collection.
The in-person training session covered topics ranging from enumerating
the household to using the visual aids to represent risk. Figure 9-3 presents
the agenda for the training session. During the session, the project team
stressed the importance of developing a thorough understanding of the logic
of the questionnaire and, therefore, carefully explained the rationale for each
9-8
-------
Contingent Valuation Survey to -
Estimate Benefits of
Hazardous Waste Management Regulations
field interviewer laining Agenda
Match 19-21,1984
Monday
March 19,1984
8:30 a.m.
8:45 a.m.
9:00 a.m.
9:45 am.
10:00 a.m.
10:45 a.m.
12:00 p.m.
1:00 p.m.
2:15 p.m.
3:15 p.m.
3:30 p.m.
5:00 p.m.
Tuesday
March 20,1984
8:30 a.m.
9:15 a.m.
10:00 a.m.
10:15 a.m.
Introduction of Trainers, Trainees, and Observers
Review of Training Agenda
Background and Purpose of the Regulatory
Benefits Survey
Break
The Benefits Questionnaire
• Overview of major sections
• Versions and variations
• Visual aids
Demonstration Interview (A simple simulated
interview designed to illustrate administration
of the direct question version D111)
Lunch
Mock Interview ri (The trainees will be divided
into two groups to ocpedita interviewing through
all sections of version 0711)
Discussion of Mock Interview #1
Break
Mock interview #2 (The trainees will be divided
into two groups to expedite roLnd-tJbin
interviewing of version D824)
Adjourn fer the Day
Discussion of Mock Interview #2
Demonstration interview of the ranking
version R111 (ranking section only}
Break
Mock Interview <#3 (The trainees will be divided
into two groups for round-robin interviewing
of the ranking section only of version R513)
Kirk i
Kirk Pate
Bill Desvousges
Kirk Pate
Kirk Pate
Bill Desvousges
Trainers
Trainees
Full Group
Discussion
Trainers
Trainees
Group Discussion
K;rk Pate
Bill Desvousges
Trainers
Trainees
(continued)
Figure 9-3, Interviewer training apnda.
9-9
-------
Contingent Valuation Survey to
Estimate Benefits of
Hazardous Waste-Management Regulations (continued)
Field Interviewer Training Agenda
March 19 - 21, 1984
Tuesday
March 20,1984 (continued)
11:00 am
12:00 noon
1:00 p.m.
1:30 p.m.
1:45 p.m.
2:30 p.m.
3:00 p.m,
3:15 p.m.
4:30 p.m.
4:45 pm
5:00 p.m.
Wednesday
March 21,1984
9 00 a m.
10:15 a.m.
10:30 a.m.
Discussion of Mock Interview #3
Lunch
Mock Interview #4 (ranking section only of
version R424)
Discussion of Mock Interview #4
Locating Sample Segments and Housing Unite
Completing the Household Control Form
• Record of contacts
• Enumeration and "eporting unit formation
• Reporting unit selection
• Eligibility rules for interview respondents
• Sample indivicual selection
Break
Continue Topic
Quality Control Procedures
• Field editing
• Telephone review of first administration
• Observations
• Validations
Pass out Assignments
Adjourn for the Day
Administrative Procedures
• Completion of Interviewer Production. Time
and Expense Report
• Preparation of assignments
• Reassignment procedures
• Visual aids and questionnaire supply
and replacement
• Disposition of completed questionnaire
and householc control forms
• Disposition of adrrinistration forms
• Scheduled weekly telephone reports
Break
Final Discussion, Clarification and Wrap-up
Group Discussion
Bill Desvousges
Group Discussion
Annette Born
Kirk Pate
Annette Born
Kirk Pate
Annette Born
Kirk Pate
Annette Born
Kirk Pate
Annette Born
1:00 p.m.
Adjourn
Figure 9-3. Interviewer training agenda (continued).
9-10
-------
element in the hypothetical market. Following this review, the project team
divided the interviewers into small groups and conducted mock interviews with
both the ranking and direct question versions of the questionnaire. The ses-
sion continued with the interviewers administering the questionnaire at home
and again with team members.
The final element in the training, full-scale practice interviews, proved
very successful. The interviewers conducted four to six practice interviews
with respondents, A member of the project team and the field supervisor ob-
served the first practice interviews and reviewed them with the interviewers.
These same project team members critiqued the final practice interviews on a
question-by-question basis in telephone conversations with each interviewer.
At the end of these sessions, the interviewers were familiar with both the logic
and purpose of each section of the questionnaire.
9.4.2 Quality Control Procedures
The field supervisor monitored all field activities on a daily basis. The
monitoring consisted of both telephone conversations and in-person review
supervision. Interviewers discussed problem cases as they arose and reported
the status of each case on a weekly basis to the field supervisor. The field
supervisor transmitted an updated computer file to the project team each week
for review.
During the data collection, two problems arose which required additional
discussion. First, the interviewers experienced unexpected problems in ob-
taining an enumeration of respondents. These difficulties stemmed from the
fact that a substantial number of the sample housing units were in limited
access apartments and from the fact that many professional persons were not
at home even after five attempts to interview them at various times of the day
and night. Second, the sample contained at (east 30 respondents who did not
understand English. (The majority were Portuguese.) The language barrier
problem proved impossible to solve without expensive (and of uncertain value)
translations of the questionnaire. However, a certified mailing with a letter
providing a strong appeal for cooperation signed by each interviewer proved
a very cost-effective way of gaining access to difficult-to-reach respondents.
Indeed, had the mailing been attempted sooner in the data collection period,
it is likely that the interviewers could have reduced the number of "no re-
9-11
-------
spondent at home" because several of the interviewers had not mailed all their
letters.
The final quality control measure consisted of telephone verification of a
randomly selected sample of the interviews. These calls all indicated that the
interviewer has completed the interviews with the respondent and that certain
selected items were accurately recorded,
9.4.3 Data Collection Summary
The data collection yielded 609 completed interviews from a total sample
of 953 sample housing units. The sample size increased over the earlier figure
cited above because of the addition of eight housing units that were missed in
the counting and listing activity. The household enumeration was the first
element of the data collection. Enumeration consisted of listing the names and
ages of the household members, determining the economic reporting units within
the household, and randomly selecting an economic decisionmaker (as defined
earlier) from the reporting unit. The interviewer attempted the initial contacts
for enumeration in-person but left notification when the respondent was not
home.
Table 9-2 summarizes the status of the household enumerations. Inter-
viewers successfully enumerated the household for slightly more than 75 per-
cent of the sample housing units. Respondents refused to be enumerated in
11 percent of the households, while no one was home in about 5 percent of
the households. A sizable percentage of the refusals gave "illness" or "too
elderly" as the reason they refused. The remaining nonenumerated households
consisted of vacant units, nonhousing units, respondents with language barri-
ers, respondents on vacation, and respondents with a physical or mental limi-
tation . These latter 15 respondents did not refuse to be enumerated but were
incapable of providing the information.
The final stage of the data collection consisted of the interview stage.
Table 9-3 provides the summary of outcomes for this stage. Interviewers ob-
tained fully completed interviews for 609 (84.58 percent) of the 720 successfully
enumerated housing units. Only three interviews were not completed after
initiation. This statistic is encouraging because it suggests that the inter-
viewers were effective in communicating the material. It also suggests that
few respondents found the interview (which lasted an average of 83 minutes)
9-12
-------
TABLE 9-2. ENUMERATION RESULTS
Desvousges, Smith, and
Present study and McGivney [19831
Result at the household Percentage Percentage of
Result code enumeration stage Number of sample Number of sample
1
Successfully enumerated
720
75.55
347
87.41
2
No enumeration eligible home
49
5.14
9
2.27
.3
Household absent during study
period
4
0.43
-
-
4
Enumeration respondent refused
105
11.02
17
4.28
5
Language barrier
23
2.41
-
-
6
Vacant housing unit
22
2.31
18
4.53
7
Not a housing unit
15
1 .57
3
0.76
8
Mentally/physically incapable
15
1.57
3
0.76
Number of sample housing units
953
100%
397
100.ooa
Total may not add to 100 percent due to rounding.
-------
TABl L " i. INTERVIEW RESULTS
Present study
Desvousges, Smith,
i i code
Result at the
interview stage
Number
Percentage of
enumerated
housing
units
Number
Percentage of
enumerated
housing
units
20
Fully completed interviews
609
84.58
303
87.32
22
Partially completed interviews
3
0,42
2
0.58 .
23
Sample individual not at home
21
2,32
14
4.03 '
24
Sample individual refused
69
9.58
24
6.92
25
Language barrier
7
0.97
1
0.29
26
Mentally/physically incapable
11
1.53
3
0.88
Enumerated housing units
720
100%
347
100.00
-------
either too difficult or disconcerting that they failed to complete it. The re-
fusal rate was a relatively modest 9,58 percent, which was also encouraging,
indicating that respondents were not discouraged by the subject area. The
remaining 33 cases consisted of incompletions because respondents were not at
home or because the respondents had language barriers or physical or mental
limitations.
The project team computed two different rates to express the results of
the field data collection process: an enumeration rate and "an interview rate.
Each rate may be calculated in two ways, depending upon how eligibility for
the survey is defined. In the strictest sense, ineligible housing units included
only those that were occupied by persons who were temporarily absent for the
study period, those that were vacant, or those that were discovered riot to
be housing units at all (for example, demolished or used as a business). In
the less strict sense, ineligible housing units also included those that occupied
by non-English-speaking persons or by persons who were physically or mentally
incapable of providing meaningful responses,
In the strict sense, the enumeration rate was computed as follows; num-
ber of enumerated housing units divided by sample size minus result codes 3,
6, and 7;
??0
= 78.95 percent .
in the strict sense, the interview rate among successfully enumerated housing
;
units was computed as follows; number of completed interviews divided by
number of enumerated housing units:
809 a A C
-J2Q - 84.5 percent ,
tn the less strict sense, the enumeration rate was computed as follows: num-
ber of enumerated housing units divided by sample size minus result codes 3,
5, 6, 7 and 8:
720 -
=82.3 percent .
In the less strict sense, the interview rate among successfully enumerated
housing units was computed as follows: number of completed interviews divided
by number of enumerated housing units minus result codes 25 and 26:
9-15
-------
j02 = 88.75 percent .
In summary, then, completed interviews were obtained from 66.78 percent
of all eligible sampling housing units under the most conservative definitions.
Under more generous but realistic assumptions, 71.46 percent of the sample
housing units yielded completed interviews.
9.4.4 Comparison With Other Contingent Valuation Studies
This section compares the results of our fieldwork with those of two other
contingent valuation studies--Desvousges, Smith, and McGivney [1983] and
Mitchell and Carson [1984]. These studies were selected because they both
elicited valuations of water quality changes, which we would expect to be an
"easier" commodity for the respondent to understand, and because both pro-
vided sufficient documentation of the fieldwork--in their respective reports--to
enable the comparison.
Table 9-2 summarizes the fieldwork results from the present study and
from the Desvousges, Smith, and McGivney [1983] study. As shown in Table
3-2, the water quality study has substantially more (12 percent) successfully
enumerated households than our study. The difference can be attributed to
higher rates of "not at homes" and refusals in our present study. However,
differences at the enumeration stage also likely reflect the difference in atti-
tude toward household surveys in the two areas (Pittsburgh versus suburban
Boston) or, more simply, the differences between the time periods--198l versus
1934--during which the two studies were conducted. Finally, our interviewers
did encounter more apartment buildings with limited access in the Boston area. '
For additional perspective on the final disposition of our sample, Table
9-4 gives the disposition of the national sample in the Mitchell-Carson [1984]
survey of individuals' willingness to pay for improving the nation's water qual-
ity. In reaching their 1,042 eligible respondents, they encountered 409 indi-
*Conversations with interviewers also suggest that the Pittsburgh inter-
viewers were very effective at using people's concern over water issues as a
way of getting a foot in the door. Boston interviewers did not have any com-
parable comments about hazardous wastes. Alternatively, the suburban Boston
residents had experienced several other surveys prior to our survey, including
one on environmental issues. Area residents couid simply have become weary
with surveys.
9-16
-------
TABLE 9-4. ENUMERATION RESULTS--MITCHEL L-CARSON [1984]
Result at the household Percentage of
enumeration stage
Successfully enumerated
No enumeration eligible home
Listing areas not assigned
Enumeration respondent refused3
Language barrier
Vacant housing unit
No information
: Number of sample housing units
Number
sample
1,042
51.20
454
22.31
33
1,62
383
18.82
28
1.28
83
4,08
14
0,83
2,035
100.00%
Source: Mitchell and Carson [1984}.
'jncli
mation.
0 .
Includes 27 respondents classified as too busy to give enumeration infor-
9-17
-------
vtduals out of a total sample of 2,035 (20 percent) who either refused or were
unable to complete the screening questions. In addition, 48? households were
not contacted because no one was home when the interviewer called, and no
information was obtained from 14 of the sample households due to administrative
reasons. Summing these numbers gives a total of 993 households; or 48 per-
cent of the total, that could not be screened for eligibility. By contrast,
about 25 percent of our households did not provide enumeration information.
Comparing refusals of enumerated households also shows that the field experi-
ence with the hazardous waste questionnaire was somewhat better. Specifically,
Mitchell and Carson [ 1984J had a 16-percent refusal rate while this study ex-
perienced about a 10-percent refusal' rate. However, differences in field pro-
cedures and in the survey designs account, at least in part, for these differ-
ences in field results. For example, in an attempt to improve our enumeration
results, our procedures required a greater number of callbacks to corny'ete
the household numeration. In addition, while the Mitchell and Carson sample
was drawn from households across the United States, our sample, as noted
earlier, is taken from a much smaller geographic area. Finally, the limitations
imposed on fieldwork procedures by the severe cost constraint in the Mitchell-
Carson study should also be acknowledged.*
In summary, our field experiences with the hazardous waste questionnaire
fare well when compared with those of two other recent contingent valuation
studies. And, although the Desvousges, Smith, and McGivney water quality
study did better in terms of enumerating households, our survey performed
as well at the more critical interview stage. When compared to Mitchell and
Carson's [1984] study, our present survey performed at least as well in both
stages of the fieldwork.
Finally, the data on successfully enumerated households in Table 9-5 sug-
gest several other illustrative comparisons. In particular, our survey compares
very well with the earlier Desvousges, Smith, and McGivney study in both
percentage of interviews completed and in the refusal rates. Both rates dif-
fered by less than 3 percent. Because the hazardous waste interview was
*!t should clearly be noted that these activities are costly, A budget
constraint for survey activities often requires that enumeration rates be opti-
mized subject to that constraint.
9-18
-------
TABLE 9-5. INTERVIEW RESULTS — MITCHELL AND CARSON [1984]
Percentage of
(Result at the
enumerated
interview stage
Number
housing units
Fjuliy completed interviews
Partially completed interviews
Sample individual not at home
Sample individual refused
Interviewed wrong respondent
Other
! Enumerated housing units
813
33
171
11
14
1,042
78.02
3.17
16.41
1 .06
1 .34
100.00%
Source: Mitchell and Carson [1984]
9-19
-------
longer and more complex, we would have expected larger differences in these
two summary statistics if our respondents had experienced difficulty with the
subject area.
9.4.5 Interviewer Debriefing
After the completion of the data collection,* the project team conducted a
debriefing session with the field interviewers and the field supervisor. This
session, which followed the precedent set in Desvousges, Smith, and McGivney
[1983], proved informative. One of the most encouraging dimensions of the
session was simply listening to the interviewers describe how they handled
various questions that arose during their interviews. Without exception, the
interviewers described solutions that were completely consistent with the goals
and procedures for the survey;" This was not self-serving behavior on their
part because, frequently, the questions involved issues not explicitly covered
in training. Their matter-of-fact delivery reinforced with concrete examples
also suggested that their answers were rooted in experiences and not their
imagination. This impression was shared by all members of the project team
but was especially apparent to the team members who had not participated in
the earlier training sessions.
In addition, the session yielded important information about the interviews,
the questionnaire, and the training. This information ranged from general
impressions to detailed suggestions for improvements. General impressions
included the following:
Even though it was not easy for some respondents, both ver-
sions of the questionnaire "worked." The interviewers ex-
pressed a slight preference for the ranking version as being
easier to administer, ;
The visual aids all contributed to the success of the interview.
Respondents tended to use them extensively.
Respondents frequently expressed genuine gratitude to the
interviewers at the end of the interview, often indicating that
they "had not thought about these things quite like that
before,"
The interview "needed more interaction with respondents.
~Technically, about 1 week remained for collection activities.
3-20
-------
All the training activities were useful, but the practice inter-
views and the intensive debr-efings really made a difference.
(Interviewers were almost unanimous on this point.)
Flipping through the visual aids in the job risk section was
the most dreaded portion of the interview. The materials were
too cumbersome. (This was noted in the pretest but the proj-
; ect team did not have time to make the extensive changes that
would have been required to make it work easier.)
The question asking for individuals to indicate a.distance that
would assure a risk reduction (#P4.b.) was probably the least
reliable question.
Specific suggestions for improvement included the following:
The colors on the risk ladder were a big plus. There were
some questions about actual numbers, but most interviewers
indicated that the ladder "usually worked" for purposes of this
study.
The introduction to the risk circles needed more pointing/inter-
action with respondents. Some respondents would have pre-
ferred shorter explanations, but many found the explanation
helpful .
A smaller cleanup slice was needed on first part of payment
vehicle card. An additional reminder of product prices would
also have been an improvement, though the card did help.
The circumstances card helped respondents keep the hypothet-
ical situation in mind.
The presence of children in a family seemed to influence re-
sponses. (This was also noted with focus group participants.)
There were not many bid revisions, and where revisions did
take place they went both ways rather than one way or the
other.
In the ranking version of the questionnaire, people frequently
mentioned having their budgets in mind. Respondents seemed
to separate highest and lowest first and -then pick between the
other two.
Respondents' values for the question on intrinsic values were
in addition to their earlier bids.
Some respondents disliked the property rights reversal in Sec-
tion G of the questionnaire. Some may have had a hard time
separating Section G from Section F,
9-21
-------
People genuinely liked the housing distance/risk reduction
(Question HI.) Several respondents indicated that this was an
"easier game to play."
Respondents seemed to use metrics of 10 in answering distances !
question (H2). They felt the responses were more ordinal than
cardinal.
In the wage risk section, some respondents expressed fairly
large dollar amounts that the interviewers thought were unrea- -
listic.
Card 1, which contained the list of common products and cor- ,
responding wastes, was the most effective card in the inter- I
view. (This was also noted in the review of videotape sessions i
and focus groups.)
9.5 SUMMARY
This chapter has highlighted the sampling procedures and the administra-
tion of the contingent valuation survey. The key points in the chapter can
be summarized as follows:
The sample design called for a two-stage, stratified, clustered
sample of economic reporting units in the suburban area around
Boston. The two primary geographic strata consisted of the
town of Acton and the remainder of the suburban Boston area.
The sample size of 953 housing units yielded 609 interviews--9 :
more than planned. The final sample was well distributed across
the various versions of the questionnaire.
The study rate, which includes enumeration and interview rates,
was about 61 percent under conservative assumptions and about
71 percent under more realistic assumptions. The interview
rate was very satisfactory, with almost 85 percent of enumerated
housing units completing the 54-minute (average) questionnaire-
Only three respondents broke off an interview after initiation.
Certified mailing provided a cost-effective means of contacting
difficult-to-reach respondents.
Intensive interviewer training using in-person sessions, at home
study, and practice interviews proved very successful. Inter-
viewers strongly endorsed practice interviews supplemented by
debriefing.
The interviewer debriefing session yielded very encouraging
information. Interviewers felt that the questionnaire worked
and that the visual aids were effective. This session also pro-
vided pertinent suggestions for improving the questionnaire.
9-22
-------
CHAPTER 10
PROFILE: THE SURVEY AREA AND ITS POPULATION i
10.1 INTRODUCTION
While there has been a growing awareness among the general public of
the potential environmental problems associated with the disposal of hazardous
wastes, it is not clear what level of information on these problems has been
acquired by the average household, Contamination incidents such as those
that occurred in Love Canai, New York, in Times Beach, Missouri, and, most
recently, in Bhopal, India, have clearly heightened the attention given both
to the production processes involving hazardous substances and to the prac-
tices and procedures used for their disposal. However, from the perspective
of the analysis of household behavior in these circumstances, it is fair to sug-
gest that analysts do not have a full understanding of either the level of avail-
able information or the public's perception of the problem. Consequently, an
important aspect of this chapter, which describes the features of the population
in the area chosen for our contingent valuation study, is a discussion of the
availability of information on hazardous wastes in the survey area and the per-
ception of the risks of exposure to hazardous wastes in comparison to other
risks faced by individual households in our sample. In addition to providing
background for the results, this description also compares the features of our
sample in relation to those of the overall population of the survey area.
Throughout our conceptual analysis to this point, hazardous waste dis-
posal practices have been described as imposing both risks of exposure and
risks of death, if exposure occurs. Regulating these disposal practices deliv-
ers a risk reduction to house. As a result, it is also important to understand
the knowledge, perceptions, and attitudes toward risk. Therefore, this chap-
ter also briefly highlights survey responses that are especially relevant to
these issues.
10-1
-------
10.2 GUIDE TO THE CHAPTER
Section 10.3 of this chapter provides a brief description of the geographic
area for the survey, an explanation of how and why it was chosen, and a brief
Socioeconomic profile of the target and sample populations. To characterize
the available information on hazardous wastes in general and the experiences
of area households in particular, Section 10.4 briefly examines a series of
Incidents involving hazardous wastes in one survey-area town and reflects on
>
the various reactions in the local population. Section 10.5 focuses on our sam-
ple and describes respondent knowledge, perceptions, attitudes, and personal
Actions concerning hazardous wastes. Finally, Section 10.6 concludes the chap-
ter with a summary of its major points.
10.3 THE SURVEY AREA AND POPULATION
This section briefly describes the survey area and its population. In
particular, after describing the geographic area--its character and its indus-
trial development--the following subsections describe how and why the area
was chosen and offers a brief socioeconomic profile of both the overall popula-
tion in the survey area and our sample.
" 0.3.1 The Survey Area
The geographic area chosen for our contingent valuation survey was the
greater Boston area--specifically, the Boston, Massachusetts, standard metro-
politan statistical area (SMSA), exclusive of the City of Boston itself. As
Shown in Figure 10-1, this target geographical area consists of more than 100
smaller communities (i.e., towns) of varying distances from Boston and of vary-
ing population sizes. Many of these smaller communities have been absorbed
by Boston as suburbs--!.e. , they are without any recent major industrial, com-
mercial, or residential development that is truly independent of the city, al-
though many others are largely self-contained communities with their own in-
dustries. Irrespective of distinctions in the economic base, however, it should
be acknowledged that many residents of these smaller communities commute to
work in the Boston central business district.
Historically, New England, and especially Boston, have been noted for
Several traditional industries--fishing, merchant shipping, textiles, and the
shoe industry. Due to a number of sociological and economic factors, however,
10-2
-------
E S € N 0
Figure 10-1. Survey area.
10-3
-------
this industrial base has been considerably broadened to include organic and
inorganic chemicals manufacture and the microelectronics and other high tech-
nology industries, including firms that specialize in the development and manu-
facture of computer hardware and software products.* Table 10-1, for exam-
ple, displays 1977 employment in selected key industries in Boston and several
other large U.S. cities. The brief socioeconomic profile provided below seems
to suggest that this broadening of Boston's industrial base, coupled with Its
considerable cultural, social, and educational resources, has helped to draw
and retain a largely white, young, well-educated, reasonably well paid popula-
tion .
There were three reasons for selecting the Boston SMSA for the survey.
First, the role of hazardous wastes in economic decisionmaking through resi-
dential housing choices has been the subject of a detailed study by David
Harrison (see Harrison [1983] and Harrison and Stock [1984]) as part of the
U.S. Environmental Protection Agency (EPA) co-operative agreement with Har-
vard University. It was recognized at the outset of this research that selec-
tion of this location offered an unusual opportunity to compare the measured
benefits associated with policies reducing the risks of exposure to hazardous
wastes. In fact the Harrison work not only offered the potential opportunity
for a comparison study similar to the Brookshire et a!, [1982] and Smith,
Desvousges, and Fisher [1984] studies, but also provided considerable back-
ground information on the nature of some specific contamination incidents in
the area.
Second, as discussed in Chapters 7 and 8, our research design called
for the completion of 600 contingent valuation interviews with economic deci-
sionmakers in local households. Obviously, successfully completing this many
interviews at a resonable per-interview cost required the selection of a target
area with a relati vely large — and preferably compact—population.
Third, the residents of the area have recently had substantial experience
with hazardous waste problems from contamination incidents. It may be, in
fact, that the greater Boston area is the prototypical urban area in this re-
*See Hekman [1980] for an interesting discussion of the historical evolu-
tion of the industrial base in New England, with special reference to the tex-
tile industry.
10-4
-------
TABLE 10-1. COMPOSITION OF 1977 CIVILIAN LABOR
FORCE FOR SELECTED U.S. CITIES
1377 Civil
an labor force (percentage)
City
Manufacturing
Wholesale and
retail trade
Professional and
related services
Atlanta
13.2
20.1
23.2
Boston
14.3
16.8
31 .6
Chicago
26.6
18.6
20.1
Dallas
18.7
23.4
16.8
Los Angeles
23.0
19,9
!
20.1 ;
New York
17.4
18.1
23.1
New Orleans
9.7
21.6
25.1
Philadelphia
20.9
19.2
2.4.5
San Francisco
10.3
19.9
23.0
Seattle
16.4
21.1
:
25.8 |
;
Washington, D.C.
4.5
11.7
;
27.8
Source: County and City Data Book, U.S. Department of Commerce, Washing-
ton, D.C. , 1983.
10-5
-------
spect. Its long and varied history of industrial development--particularly with
manufacturing industries whose production processes generate hazardous waste
byproducts--make it well suited for the study, especially since this long experi-
ence usually implies rather substantial and prolonged media coverage. Indeed,
as indicated in Table 10-2, which summarizes news items taken from two Boston
area newspapers, The Boston Globe and The Acton Beacon, several different
communities in the greater Boston area have experienced major problems with
hazardous waste management.
In summary, the greater Boston area was chosen as the location for this
survey not only because it offered easy, cost-effective access to the numbers
of respondents required by the experimental design, but also, and more impor-
tantly, because it offered the opportunity to develop a comparative analysis of
alternative methods for measuring the values of hazardous waste policy and
because it offered the chance to study an area whose residents have recently
been forced to deal with the problems of hazardous wastes and who are there-
fore likely to be interested and well-informed on the problem. This last dimen-
sion is quite important to our study, since, as implied by the reference oper-
ating conditions proposed by Cummings, Brookshire, and Schulze [1984], fam-
iliarity and experience with the circumstances involved in contingent valuation
experiments may well be quite important to the ability of the method to elicit
reasonably accurate valuation estimates,*
10.3.2 Socioeconomic Profile
As noted earlier, the social, cultural, educational, industrial, and economic
opportunities offered by Boston and its surrounding communities have attracted
a fairly well paid, predominantly white, young, and fairly well-educated popu-
lation . Based on the population information from the 1980 Census, Table 10-3
provides an economic and demographic comparison of the overall population
and the sample we acquired. The two sets of descriptive statistics compare
remarkably well because of the sample design. For example, based on 1984
dollars, the median income for the target population is $32,723; the medium
income of the sample is only slightly lower, at $32,500. Similarly, the target
*This is especially important in our case because the commodity being
valued is a risk change.
10-6
-------
TABLE
10-2, NEWS SUMMARY; GREATER BOSTON AREA COMMUNITIES
EXPERIENCING PROBLEMS WITH HAZARDOUS WASTES
Greater Boston
community
Date
News item
Woburn
May 1979
State officials announce: the discovery of 187 barrels of
abandoned chemical wastes containing polyurethan# resin.
East Woburn
May 1979
Two wells are shut down when tests reveal they have
been contaminated with trichioroethylene.
Da rivers
May 1979
Oil spill at Danvers State Hospital teaks 2,500 gallons of
oil into brook.
Lowell
May 1979
Bankrupt owner of hazardous wast# storage facility
abandons 15,000 barrels containing several million gallons
of chemicals. Fires and explosions ignite hundreds of
barrels.
Kingston
March 1980
Several hundred barrets of abandoned toxic chemicals are
discovered at mar of property by owner of trucking firm.
Sommervilie
April 1980
Thousands of residents are affected by fumes from phos-
phorus trichloride spilled in a railway accident.
Salem
April 1980
State and local police raid a warehouse and confiscate
350 cardboard arid steel drums containing illegally stored
chemical wastes.
Canton
January 1981
State officials order removal of 40 barrels of polychlon-
nated tipheniys (PCBs) from farm site.
Lowell
February 1981
Based on complaints of fumes and noxious odors by resi-
dents , Federal, State, and local officials begin sampling
wastes at a barrel and drum company.
Mlddlebo rough
July 1981
Junkyard owner found in contempt for failing to clean up
300 barrels of hazardous wastes buried on his property.
Woburn
Ashland
Tyngsborougb
October 1981
Three Greater Boston communities appear on EPA list of
114 priority hazardous waste locations.
Acton
Ashland
Bridgewater
Groveiand
Holbrook
Lowell
Plymouth
T yngsborough
Westbo rough
Woburn
December 1982
Ten greater Boston area comntun t e appear on EPA list
of the nation's most dangerous che 1 Sump sites.
Boxborough
May 7983
City officials seek source of Clapp well contamination,
contemplating lawsuit,
Acton
Boxborough
Bedford
May 1983
State officials identify three communities whose drinking
water supplies are susceptible to contam nation from local
hazardous wastes.
Source: The Boston Globe, Boston, Massachusetts, 1979-1983; The Acton Beacon, Acton,
Massachusetts, 1979-1983,
10-7
-------
TABLE 10-3. CHARACTERISTICS OF THE TARGET
POPULATION AND THE SAMPLE
Characteristic
Target
population'
Sample
Income
Median
Per capita
Race
Percent white
$32,723
$11,447
97,0
$32,500
$12,185
37,2
Age
Median years
Percent 85 years or over
Sex
Percent male
39,4
12,0
46,6
42.5
17,2
39.2
Education
Median school year completed
. Percent high school graduates
; Percent college graduates
Family status
8
¦ Percent single
Percent of households with children
less than 18 years old
Persons per household
Mobility
.. Percent living at the same address
for the last 5 years
; 6
Labor force participation
Male
Female
12.9
79.4
25.8
32.6
48.4
2,8
83,0
78.0
54.2
14.0
89.2
38.3
35.9
36.0
2,7
73.3
78.4
63.3
Source: 1980 Census of Population and Housing, U.S. Department of
Commerce, Bureau of the Census, Washington, D.C., 1382.
aThe target population is defined as individuals within the Boston SMSA and
outside the city of Boston. All values for the target population are derived
from the 1980 Census of Population and Housing.
bThese Census data values were converted from reported 1979 dollars to 1984
dollars using the GNP implicit price deflator.
c
These populations include only those individuals more than 17 years old.
^These populations include only those individuals more than 25 years old.
e :
As the Census data uses include individuals more than 15 and the survey was
administered to those 18 years or over, these populations are not identical.
10-8
-------
and sample are primarily white and young: While the target population is
97 percent white/ has a median age of 33.4 years, and is only 12 percent 65
years old and over, the sample is 97.2 percent white, has a slightly higher
median age, 42.5 years, and is slightly more than 17 percent 65 years old and
over. The largest discrepancy between the target and the sample arises in
comparing sex composition. Both are less than 50 percent male, with the tar-
get population at 46.6 percent and the sample at 39.2 percent.
The target population and sample exhibit fairly high levels of education,
though the sample is slightly more well educated. Specifically, while the tar-
get population has on average completed nearly 13 years of school and has
graduated from high school more than 79 percent of the time and from college
almost 26 percent of the time, the sample has on average completed 14 years
of school and has graduated from high school almost 90 percent of the time
and from college more than 38 percent of the time. The sample has slightly
more single persons than the target population (35.9 versus 32.6 percent) and
has fewer children less than 18 years old than the target population (36.0 ver-
sus 48.4 percent). Somewhat surprisingly in view of these other family sL.itus
statistics, however, the sample and target population have virtually identical
family sizes--2.7 and 2.8, respectively. Finally, the sample has been living
at the same address longer than the target population (73.3 versus 63.0 per-
cent, respectively) and has a greater number of female workers (63.3 ve-sus
54.2 percent, respectively) in its labor force.
10.4 A SURVEY FOCUS: HAZARDOUS WASTES IN ACTON
The experimental design of our survey deliberately called for oversampling
in a single community in the Boston SMSA--Acton, a small town of approximate-
ly 19,000 people about 45 minutes northwest of the City of Boston. The deci-
sion to oversample in Acton was made for several reasons. First, one compo-
nent of the Harvard University study [Harrison, 1983; Harrison and Stock,
1984] mentioned earlier had alreaay developed detailed analyses of the conse-
quences of recent contamination incidents in Acton, including a risk assess-
ment, an evaluation of household and community averting costs, and the appli-
cation of an area-wide hedonic property value model. Thus, since an important
objective of our study was to develop information that could enable a comp,3ra-
10-9
-------
tive analysis of estimates from a variety of benefits measurement approaches,
we felt detailed contingent valuation estimates for Acton were necessary. Sec-
ond, and equally important, the experience in Acton is itself quite interesting.
As shown in Table 10-4, Acton has over a decade of experience with hazardous
waste problems. It is therefore reasonable to expect that local households wilt
have considerable familiarity with the issues encompassed by these kinds of
problems. This prior knowledge should facilitate the process of communicating
the contingent valuation questions. It has, as we have noted throughout the
report, been argued to be an important factor in determining the plausibility
of contingent valuation estimates in past studies (see Cummings, Brookshire,
and Schulze [1984]).
Although Acton is in many respects primarily a bedroom community whose
residents, predominantly professionals and skilled technicians, commute to work
in Boston or in the high-technology companies that have sprung up in other
larger communities around it, the town has experienced more serious environ-
mental pollution and hazardous waste contamination incidents than most towns
with considerably greater industrial development. In 1982, in fact, EPA listed
Acton as the site of one of the nation's most dangerous chemical dump sites,
and, in 1383, that dump was listed as one of 38 top priority sites eligible for
cleanup under the "Superfund" Act.
Given Acton's small size and relatively modest industrial development, it
is not surprising that its major environmental/hazardous wastes contamination
problems are almost synonymous with those of its largest industrial resident—a
large chemical firm that has operated a battery separator plant and variously
manufactured organic chemicals, synthetic rubber, and plasticizers in Acton
since 1945. Based on news items that appeared in the local newspaper, The
Acton Beacon, Table 10-4 summarizes the history of major environmental and
hazardous waste contamination incidents in Acton during the last 15 years.
As shown in Table 10-4, Acton experienced five major incidents involving
hazardous and/or toxic substances during the past 15 years--all related di-
rectly or indirectly to operations at the chemicals plant.* The most important
*lt should be noted that Acton has also experienced environmental pollution
and hazardous waste contamination incidents that are not related to operations
at the chemical plant. However, none of these had the impact of the five inci-
dents listed in the table.
10-10
-------
TABLE 10-4. SUMMARY OF MAJOR ENVIRONMENTAL POLLUTION AND HAZARDOUS WASTE
CONTAMINATION EPISODES IN ACTON
Date
Contaminant
Description of incident(s)
1973 to 1978
Organic chemicals
Acton residents repeatedly complain of chem-
ical odors, paint peeling from houses, and
effects on vegetation. Local chemical plant
institutes odor screening program in 1973
and extends it in 1974 and again in 1977.
Company denies in 1978 that odors result
from plant chemical emissions.
November-December 1978
Organic chemicals
Acton Water Supply District detects several
organic chemicals in two municipal water
supply wells and shuts them down immedi-
ately, decreasing the Acton municipal water
supply by 35 to 40 percent.
August 1981 Styrene
Fumes leak from an underground storage
tank at local chemical company, requiring
an emergency evacuation of 100 Acton and
400 Concord residents.
August 1982 Oil
January 1983 Hexane
Oil (6,500 gallons) leaks from an under-
ground storage tank at local chemical com-
pany, risking contamination of Sinking Pond
Aquifer. Test wells show oil reaches depths
of 14 to 40 feet.
Hexane (1,400 gallons) leak is discovered
in an underground storage tank at local
chemical company. Leak actually occurred
in November 1982. Eventually, hexane
dilutes oil spilled in earlier incident, further
risking contamination of aquifer.
Source; The Acton Beacon, Acton, Massachusetts, 1973-1933.
-------
of these incidents is the water supply contamination that occurred in November-
December 1978, which involved the contamination of the water in two municipal
wells that constituted 40 percent of Acton's water supply, A year-long hydro-
geologic study in 1979 concluded that the aquifer supplying the two wells, the
Sinking Pond Aquifer, had been contaminated by liquid wastes from two chem-
ical company lagoons and a landfill located 2,500 to 3,000 feet north of the
two municipal wells, Assabet No. 1 and No. 2. This study also identified a
contamination plume with chlorinated hydrocarbon concentrations as high as
10,000 ppb within 1,000 feet of Assabet well No. 2.
While the loss of 40 percent of a municipal water supply is a serious dis-
ruption, the other long-term consequences of this major hazardous waste con-
tamination incident may well prove to be more serious. Primary among these
were the questions that quite naturally arose concerning the condition of the
town's remaining water supply and its susceptibility to future contamination
incidents. In addition, this incident raised questions as to the chemical com-
pany's ability to manage its operations--and its hazardous waste byproducts — in
a- manner that would ensure the health and safety of the citizens of Acton.
Perhaps most important of all, the long-term implications of the potential health
effects of exposure to the six contaminants identified in the two municipal wells
are unclear. Table 10-5 lists these contaminants along with their potential
health effects. Although no data have yet shown that any Acton resident has
experienced any of these health effects as a direct result of the water supply
contamination in 1978, the long latency periods associated especially with car-
cinogens leave the ultimate impact of the contamination incident an open ques-
tion .
Several aspects of Acton's contamination incidents are especially relevant
to the objectives of our research: (1) the sources and types of information
provided on the first undertaking and three subsequent contamination episodes,
(2) the nature and types of activities the citizens who responded to these inci-
dents, and (3) the actions taken by local government in response to the prob-
lems posed by the incidents, It is difficult to reconstruct specific events that
would answer all the questions related to these areas. However, in an effort
to develop a better understanding of the events that surrounded these contam-
ination episodes and how residents learned and responded, we undertook a
10-12
-------
TABLE 10-5, CONTAMINANTS FOUND IN ACTON WATER SUPPLY
AND THEIR POTENTIAL HEALTH EFFECTS
Contaminant
Potential health effect(s)
T richloroethylene
Suspected carcinogen ;
Neurological effects--dizziness, loss of
appetite, loss of motor coordination
Causes cell mutation and liver damage;
Dichioroethane
Suspected carcinogen
Central nervous system "damage
Liver damage
Kidney damage
Dichloromethane
Central nervous system damage ;
T richloromethane
Carcinogen
Central nervous system damage
Blood chemistry effects
Kidney damage
Liver damage
Heart damage
Ethylbenzene
Suspected carcinogen
Benzene
Carcinogen ;
Blood chemistry effects
Fatigue
Anorexia
Central nervous system disorders ;
10-13
¦
-------
search of 6 years of news items reported in The Boston Globe and The Acton
Beacon. Composed of summaries of these news items, Tables E~1 through E-3
in Appendix E present the results of our search by describing the chronology
of events in three ways. Table E-1 describes the nature of the information
available, specifically from October 1978 to May 1983; Tabie E-2 highlights this
information as well as additional sources of the record of community responses
tjo the incidents over the same approximate time span; and Table E-3 describes
the actions taken by the local Acton town government. Because a substantial
overlap among these three sets of information is almost inevitable, each dimen-
sion of these three interrelated issues is not discussed in detail. Rather, we
summarize the most important elements that appear to characterize each.
Based on the summary information in Appendix E, it appears fair to con-
elude that the local news coverage was excellent. Indeed, the public had con-
tinual and immediate access to substantial amounts of political, technical, and
other factual information concerning hazardous wastes In genera! and their
water supply and other waste contamination problems in particular. In addi-
on, the public had other important sources of public information, including
e reports released after several hydrogeologic studies of the contaminated
Sinking Pond Aquifer and the information assembled and dissiminated by vari-
ous citizens groups, such as the Acton League of Women Voters (LWV). The
availability of all these studies and other information sources was also reported
in The Acton Beacon. The following summary lists the most important sources
and types of public information available to Acton residents during the hazard-
ous waste contamination controversy:
Newspaper coverage--The Acton Beacon and The Boston Globe
together covered the full range of events surrounding the con-
tamination incidents in Acton as well as their implications for
Acton and the larger greater Boston community. Particularly
important is the appearance of a "water" column in The Acton
Beacon to provide public information on water-related issues on
a regular basis.
Technical Reports--Several different studies by several differ-
ent engineering consulting firms were commissioned by Acton,
the chemical company, and others — all resulting it detailed tech-
nical information on the incidents and their implications.
:
10-14
-------
Public Meetings--Several different forums for public discussions
existed during the incidents, including meetings of local con- i
cerned citizen groups and public meetings of the Acton Board
of Selectmen (ABS) and others.
Publications of Environmental Groups--Several different envir-
onmental groups--including the Audubon Society and the Sierra ,
Club--published accounts of the incidents in Acton and their
implications for the local citizenry and for hazardous waste man-
agement generally.
Pamphlets—The Acton League of Women Voters published several
pamphlets on the history of the incidents and on ways to cope :
with their effects.
Based, once again, on the new items summarized in Appendix. E, the citi-
zens of Acton reacted quickly to protect themselves and to monitor the actions
being taken both by the chemical company and by the local government to
address the problems posed by the contamination incidents. In particular,
the Thoreau Group (a local Sierra Club chapter), the Acton Committee for
Environmental Safety (ACES), the Metropolitan Area Planning Counci I ( MA PC ),
the Acton League of Women Voters (LWV), the Water Land Management Advis-
ory Committee (WLMAC), the Citizens Association for Preservation of the Envi-
ronment (CAPE), the West Concord Citizens (WCC), the Concerned Citizens
Coalition of Billerica (CCC), and numerous individual Acton citizens took the
following specific actions:
Speaking to each other and to town government officials (e.g. ,
the ABS) about the planned chemicals company plant expansion *
and the contamination incidents.
Submitting citizen petitions on contamination to the town
government.
,
Joining in on the suit filed against the chemical company by j
EPA and others.
Gathering, organizing, and distributing information on contami-
nation and mitigation strategies.
Ensuring large turnouts at ABS meetings on contamination prob-
lems.
Seeking reimbursement from chemical company of evacuation
costs due to contamination. :
10-15
-------
; * Questioning ABS on delays in chemical company "actions to com-
ply with the Consent Decree.
Serving on the Technical Advisory Committee (TAG) appointed
by ABS to review materials on contamination incidents.
Supporting each other in position of mutual interest against
the chemical company or ABS.
Pressing for a "Hazardous Waste Day," during which hazardous
household products could be collected.
Finally, also based on the news items summarized in Appendix E, the town
government acted quickly and repeatedly to protect its citizenry and to address
the problems posed by the contamination incidents. In particular, specific
actions by the Acton Water Supply District (AWSD) and the Acton Board of
Selectmen (ABS) include the following:
Identifying the well contamination and its contaminants
Closing the contaminated wells
Demanding the chemical company fund a hydrogeologic study
Controlling local water use through bans
Investigating and reporting on mitigation strategies for the con-
tamination
Appropriating funds for technical studies by the town
Filing suit against the chemical company
Locating and tapping additional water supply sources from sur-
rounding communities
Inspecting the chemical company plant site and forcing it to
comply with State and local requirements
Meeting with and critiquing the action of Federal environmental
protection personnel
Meeting with and criticizing the actions of chemical company
officials.
30.5 RESPONDENT KNOWLEDGE AND PERCEPTIONS OF HAZARDOUS
WASTES AND THEIR RISKS
This section describes respondent knowledge, perceptions, and attitudes
concerning hazardous wastes and the risk(s) associated with them. Specific-
10-16
-------
ally, it summarizes what respondents knew about hazardous wastes, how they
had learned it, how serious they thought the problem was, how they perceived
their risks from it in relation to other sources of environmental pollution,
whether they had taken any action to try to mitigate that risk, and, finally,
how effective they thought the government and other organizations were in
dealing with the hazardous waste problem,
10.5.1 Respondent Knowledge
Tables 10-6, 10-7, and 10-3' describe how many respondents had recently
read or heard about hazardous wastes in the media, the frequency with which
they had read or heard about them, and the subject geographic area of the
information they had read or heard about, respectively. As shown in Table
10-8, 551 respondents, 90.6 percent of the sample, had recently read or heard
about hazardous wastes in the media, while only 57 respondents, or 3,4 per-
cent, had not. In addition, although nearly 40 percent of the respondents
did not know on how many occasions they had heard or read about hazardous
wates, most did know. Table 10-7 summarizes these results, showing that 10
percent indicated a frequency of 1 time; 2 percent, 2' to 5 times; 29 percent,
6 to 10 times; and 20 percent, 11 times or more. As indicated in Table 10-3,
most respondents, 74 percent, said that the geographical area associated with
the hazardous waste information they had recently read or heard about was
their own state; nearly 40 percent said it concerned their own town;, and 87
percent said the information concerned the entire nation. Therefore, it
appears that the survey respondents had almost invariably read or heard about
hazardous wastes in the recent past; that most of the respondents had read
or heard about them on 6 or more occasions; and that most of the respondents
saw or heard about information concerning their own state or town.
10.5.2 Respondent Perceptions
This section summarizes how serious an environmental problem the survey-
respondents thought hazardous wastes were and how effective they thought
various organizations — Federal, State, and local governments and other organi-
zations—were in dealing with the problem. It should be noted that we elicited
these ratings from the respondents by using a scale card that we explained
should be used to provide a basis for scaling the issue addressed. Thus,
10-17
-------
TABLE 10-6. NUMBER AND PERCENTAGE OF TOTAL
RESPONDENTS WHO HAD RECENTLY READ OR
HEARD ABOUT HAZARDOUS WASTES
Responcents
g
Status Number Percent
Had read or heard about 551 90,6
hazardous wastes
Had not read or heard 57 9.4
about hazardous wastes
"I don't know"'3 4
a"Percent" column may not total 100 due to rounding.
b
Respondents giving "I don't know" answers are excluded from
the population from which percentages are calculated.
TABLE 10-7. FREQUENCY WITH WHICH RESPONDENTS HAD
RECENTLY READ OR HEARD ABOUT
HAZARDOUS WASTES
Respondents
Frequency
Number
Percent3
1 time
57
* 9.6
2 to 5 times
12
2.0
6 to 10 times
171
28.7
11 times or more
119
20.0
Don't know how many times
236
39.6
M b
No answer
17
-
3
"Percent" column may not total 100 due to rounding.
Respondents giving "I don't Know" answers are excluded from
the population from which percentages are calculated.
10-18
-------
TABLE 10-8. SUBJECT OF HAZARDOUS WASTE
INFORMATION RECENTLY READ OR HEARD
BY RESPONDENTS
Respondents3
Frequency
Number
Percent
Respondent's
town
244
39.9
Respondent's
state
453
74.0
Entire nation
414
67.6
3
Columns do not total because the "subjects" are not mutually
exclusive--'!.e., because the information read or heard by
respondents could concern some combination, or all three, of
the subject areas.
10-19
-------
the numerical values reported in the following two subsections are intended to
provide an index of the degree of harm associated with hazardous wastes or
the degree of effectiveness of the governmental unit, respectively.
Severity of the Problem
Table 10-9 displays respondent ratings of the degree of harm posed by
pollution from eight current environmental problems, including hazardous
wastes. As shown in the table, respondent perceptions of the harm of the
eight environmental problems is fairly evenly distributed, with a few impor-
tant exceptions at each end of the scale of harm. Specifically, 79 percent,
considered pollution from strip mining "not harmful," while 40 percent thought
that pollution from nuclear and other radioactive wastes is "not harmful."
However, perhaps because of their recent and, for the most part, frequent
encounters with information on hazardous wastes, 18 percent of the respond-
ents felt that pollution from hazardous waste is "very harmful." In addition,
respondents clearly felt most pollution sources--e.g., sewage, nuclear wastes,
acid rain, strip mining—are less harmful than hazardous wastes,
Especially interesting for our case and for the likely performance of con-
tingent valuation questions involving the risks of exposure to hazardous wastes
are the numbers of "I don't know" answers given by respondents for each of
the eight pollution sources. In particular, as shown in Table 10-3, only 27
respondents did not know how to rate the relative harm of pollution from haz-
ardous wastes, compared to 36 for sewage, 66 for nuclear wastes, 76 for acid
rain, and 94 for strip mining. Only automobiles, manufacturing, and solid
wastes had fewer "I don't know" answers than hazardous wastes. Given the
detailed information that has been provided in this area over the past 6 years,
this enhanced degree of knowledge concerning hazardous wastes is not surpris-
ing.
Organizational Effectiveness
Table 10-10 summarizes the respondent effectiveness ratings of six key
types of organizations that have responsibilities for dealing with hazardous
wastes, including Federal, State, and local governments, local water districts,
and both waste-producing and waste-disposal firms. Based on this information,
few respondents rated any of these organizations as "very effective." Local
10-20
-------
TABLE 10-9. RESPONDENT HARMFUL NESS RATINGS OF £NViRONM£N f AI POLLUTION SOURCES
Number and percentage of total respondents by pollution sources
Degree of
harrnfulness
Automobiles
Mjnuldtlur iriq
Solid
wastes
Sewage
Nuclear
wailes
Hazardous wastes
Acid
rain
Strip
miniria
Number
Pert ©nl
Number
Percent
Number
Percent
Number
Percent
Number
Percsnl
Number
Percent
Number
Percent
Number
Percept
1
21
§ 3.5
113
19.0
91
IS 2
121
21.0
222
40.7
112
19.2
107
20.0
411
79.3
2
31
5, I
60
10.1
84
14,0
62
10.8
78
14.3
49
a.a
26
4.9
29
5.6
3
5?
9,1
52
8.7
88
14.7
65
11,3
36
6 6
49
8.4
37
6,9
14
2.7
4
€4
10.6
45
7.6
62
10,3
55
9.6
ZS
4,6
35
6.0
40
7 5
9
1.7
5
111
18.3
64
10,?
76
13.0
73
12.7
M
6 6
45
7.7
60
11.2
19
3.7
6
67
11.1
49
8.2
56
9,J
«
7.5
tl
2 4
42
7.2
40
7,5
10
1.9
1
73
12.0
49
8.2
48
8.0
39
6.8
Z6
4.8
51
8.7
SO
9.3
7
1.-1
8
70
11.6
52
8.7
43
7.2
47
8,2
19
3.S
58
9.9
61
11.4
>0
1.9
9
35
5.8
39
6.5
17
2.8
21
3.6
1?
3.1
38
6.5
4.2
7.8
1
0.2
10
77
10.?
73
33
5.5
SO
B. 7
74
13.8
106
18.1
73
13.6
8
1.5
,-b
6
.
16
-
12
-
36
-
68
.
27
-
76
-
94
.
O
NJ
Not harmful
Very harmful 10
"I don't know
^'Percent" columns may 101 total 100 due to rounding.
^Respondents giving "I don't know" answers are excluded from the population from which percentages are calculated -
1
-------
TABLE 10-10. RESPONDENT EFFECTIVENESS RATINGS OF ORGANIZATIONS THAT DEAL WITH HAZARDOUS WASTES
Number and percentage of total respondents fay rated organization*
Federal Slate Local local Waste- Waste-
government government government water district generating firms disposal firms
effectiveness Number Percent Number Percent Number Percent Number Percent Number Percent Number Percent
Not effective
1
69
11.9
37
6.6
46
8.8
45
9,0
92
10,
.2
75
18.0
2
S3
10.2
SI
9,0
35
6.7
24
4.8
83
16.
.4
57
13,7
3
90
15.8
75
13.3
47
9,®
27
5.4
78
15
.4
61
14,8
4
87
15.1
84
14.9
52
9,1
34
6.8
65
12
.9
49
11.8
5
123
21,3
133
23.6
91
17,4
80
18.0
86
17,
,0
80
14,4
«
59
10.2
71
12,6
42
B.O
44
a.8
35
6,
.9
36
8.6
7
42
7.3
64
11.3
70
13.4
58
11.0
29
5,
.7
38
8.1
§
34
5.9
31
5.5
78
14.S
89
17,7
22
4.
,4
20
4,8
i
4
0.7
e
1.4
38
7.3
45
9.0
§
1,
,2
13
3.1
Very effective
10
11
1.9
10
1.8
2?
5.2
59
11.8
8
1
i
S
1.9
"f don't know"'1
-
34
.
48
-
88
-
110
.
107
195
-
^''Percent" columns may not total 100 due to rounding.
^Respondents giving "I don't know" answers are excluded from the population from which percentages are calculated.
-------
water districts received the greatest number of high ratings (59 "very effec-
tive" ratings, or 11.8 percent of all respondents who rated the water districts),
and four of the five remaining organizations are approximately comparable in
their small number of "very effective" ratings (between 8 and 10, for 1.8 to
1.9 percent of all respondents who rated those organizations). On the other
hand, few respondents (from 7 to 11 percent) rated any of the organizations
as "not effective," although approximately twice as many (about 18 percent)
rated waste-generating and waste-disposal firms as "not effective." Respond-
ents gave ratings of five on the effectivness scale of 1 to 10 more often than
any other rating for all the organizations but one — local water districts. Con-
sistent with the "very effective" rating results reported earlier, respondents
who rated the local water districts gave them more ratings of 8 on the scale
of 1 to 10 than any other rating and also gave them more ratings of 6 and
higher.
(t is also interesting to note the varying numbers of respondents who
answered "I don't know" in response to the request to rate each of the six
types of organizations. For example, 34 respondents declined to rate the effec-
tiveness of the Federal government, 48 declined to rate their State government,
88 declined to rate their local government, 110 declined to rate their local water
district, 107 declined to rate waste-generating firms, and 195 declined to rate
waste disposal firms.
10.5.3 Respondent Awareness of Risk
Focusing on comparisons of annual risks of death from a variety of
sources, on the levels of risk associated with various hazardous waste expos-
ure pathways, and on specific likely causes of death, this section summarizes
the extent to which respondents were aware of their actual risks of exposure
to hazardous wastes and the health effects that might be associated with that
exposure.
Annual Risks of Death
Table 10-11 reports how respondents' perceptions of their annual risk of
death from exposure to hazardous wastes compare to their annual risks of death
as a result of an automobile accident, heart disease, and exposure to air pollu-
tion. These data were collected from the respondents using a risk ladder that
10-23
-------
TABLE 10-11. RESPONDENT-RELATED ANNUAL RISKS OF DEATH FROM SELECTED
CAUSES USING RISK LADDER
Number and percentage of total respondents by cause of death8
Automobile Hazardous
accident Heart disease Air pollution wastes
Number Percent Number Percent Number Percent Number Percent
Flood
0.05
53
8,8
159
26.2
223
36.9
191
31.7
Poisoning
§.6
40
6.6
99
16.3
120
IS,9
123
20.4
Airplane
0.8
56
9.3
60
9.9
56
9.2
57
9.5
Home fire
2.8
72
11.9
54
8.9
54
8,9
55
9.1
Insurance agent
4.0
46
7.6
33
5.4
12
5.3
40
6.6
Banker
6.0
42
7.0
25
4.1
27
4.5
41
6.8
Home accident
11.0
32
15.2
22
3.6
22'
3,6
24
4.0
Diabetes
15.1
32
5,3
17
2.8
10
1.7
IS
2.5
Police
22.0
28
4.6
14
2.3
12
2.0
5
0.8
Home builder
47.0
49
8.1
1?
2.8
16
2.6
17
2.8
Stroke
77,0
29
4.8
65
10.7
8
1.3
11
1.8
Truckdriver
99.0
33
6.5
11
1.8
9
1.5
12
2.0
Skydiver
200.0
11
1.8
3
0.5
0
0.0
1
0.2
Smoker
300.0
?
1.2
21
3.5
12
2,0
9
1.5
Slunlmart
2,000.0
8
1.3
7
1.2
3
0.5
2
0.3
"1 don't know"b
-
6
-
5
-
8
- ¦
9
-
a"Porcent" columns may not total 100 duo to rounding.
Respondents giving "I don't know" answers are excluded from the population from which percentages are calculated.
Annual risk of death
Anchor from Chance of death
risk ladder In 100,000
-------
was developed in the focus group sessions (see Chapter 8). Figure 10-2 pre-
sents the ladder,*
in general, respondents who did not decline to answer the questions rated
their chances of dying from hazardous waste exposure as considerably more
remote than that of dying as a result of heart disease or of an automobile acci-
dent, t Almost 32 percent (191) of the respondents selected the most remote
possibility--0.05 chance in 100,000 (the same as the chance everyone faces of
dying in a flood)--as their own risk of dying during the next year as a result
of exposure to hazardous wastes. Only air pollution was selected by more re-
spondents--223 f or approximately 37 percent—as the most remote cause of death
possible. Also as shown in the table, another 123 respondents (a little more
than 20 percent) selected the second most remote possibility—0.6 chance in
100,000 (the same as the chance everyone faces of dying of poisoning)—as
their risk of dying doing the next year as a result of exposure to hazardous
wastes. Approximately the same number of respondents—120, or nearly 20 per-
cent—chose air pollution exposure as the second most remote cause of death
possible.
In contrast to the annual risk results for hazardous waste exposure; many
fewer respondents considered their chances of dying as a result of an auto-
mobile accident or as a result of Heart disease so remote. Only 53 respondents
(or not quite 9 percent) selected the most unlikely possibility (0.05 chance in
100,000) as their annual chance of dying as a result of an automobile accident.
Exposure Risks by Exposure Pathway
Table 10-12 summarizes respondent perceptions of the various fevefs of
their risk of exposure to hazardous wastes for five different exposure path-
ways—drinking water, breathing air, touching soil, eating food, and eating
fish (from contaminated waters). The respondents generally felt themselves
more at risk for hazardous waste exposure through drinking water, breathing
air, and eating fish than through touching soil or eating foods other than fish.
~The actual form of the ladder used in the interviews identified the seg-
ments of the ladder with different colors.
fA small number of respondents declined to rate their risk of dying in
each category.
10-25
-------
Card 5
Risk Ladder: Comparing Annual Risks of Death
15
Stuntmsn
2.000 of 100.000
14
13
12
11
10
09
08
07
06
OS
04
03
02
01
Smokar"
Skvdivw
Shiobuildw/T ruckdriver
Stroke
Hometaui ld«r
Polica Officar
Oia betes
Horn# Accident
Lj
LJ
Banker/En<|im«r
Insurance Adam
Horn* Fire
Aircians
Poiwninq
Flood
300 of 100,000
200 of 100.000
99 of 100,000
77 of 100,000
47 of 100,000
22 of 100.000
IS. 1 of 100,000
11 of 100,000
6 of 100,000
4 of 100,000
2.3 of 100.000
0,8 of 100.000
0.6 of 100,000
.OS of 100,000
•At !***t on* pack pw day.
Figure 10-2. Final version of risk ladder incorporating
suggestions from participants.
10-26
-------
TABLE 10-12. RESPONDENT-RAT ED CHANGES OF EXPOSURE THROUGH
TYPICAL EXPOSURE PATHWAYS
Chance
of exposure
Number and percentage of total respondents by exposure pathway'
Breathing air
Touching soil
Eating food
Eating fish
Drinking water
Number Percent Number Percent Number Percent Number Percent Number Percent
No chance at all
"I don't know1
1
116
19.4
50
8.2
190
31.9
80
13.5
74
12.3
2
61
10.1
48
7.9
105
17.6
54
9.1
39
6.5
3
49
8.2
57
9.4
73
12.3
52
8.8
37
6.2
4
41
6.8
46
7.6
42
7.1
47
7.9
38
6.3
5
55
9.2
85
14.0
5&
9 7
90
15.2
81
13 5
6
33
5.5
38
6.3
29
4.9
51
8.6
49
8.2
7
19
3.2
47
7.7
33
5.6
41
6.9
49
8.2
8
55
9.2
63
10.4
17
2.9
62
10.5
85
14.2
9
29
4.8
30
4.9
7
1.2
27
4.6
33
5.5
10
141
23.5
144
23.7
41
6.9
88
14.9
115
19.2
13
_
4
_
17
20
12
"Number11 columns exclude respondents who answered "I don't know." "Percent" columns may not total 100 due
to rounding.
Respondents giving "I don't know" answers are excluded from the population from which percentages are
calculated.
-------
Indeed, 141 respondents, or more than 23 percent, felt that they risked "cer-
tain exposure" through drinking water; 144 respondents (nearly 24 percent)
felt they risked certain exposure through breathing air; and 115 respondents
Cor more than 19 percent) felt they risked certain exposure through eating
contaminated fish.
The varying numbers of respondents who answered "I don't know" may
indicate their knowledge of the various exposure pathways, or at feast their
level of confidence in their knowledge. For example, that'fewer respondents
declined to rate their exposure risk for drinking water, breathing air, and
eating fish (13, 4, and 12, respectively) than declined to rate their exposure
risk for touching soil (17) or eating food (20).
Likely Causes of Death From Exposure
Table 10-13 summarizes the causes of death that respondents had in mind
when they responded to the contingent valuation questions for reductions in
their risks of exposure to hazardous wastes. As shown in the table, most
respondents (nearly 54 percent) indicated that they had no specific cause of
death in mind. However, of the remaining respondents, 226 indicated that
they believed cancer to be the health effect that would result from exposure,
and 21 indicated they had either lung disease or leukemia in mind. These
findings, together with the fact that only 14 respondents answered "I don't
know," indicate that a large number of respondents (nearly half) had definite
ideas about the potential health effects of hazardous waste exposure when they
responded to the valuation questions associated with reducing their exposure
risks.
10.5.4 Respondent Actions to Reduce Risks
Finally, Tables 10-14, 10-15, 10-16, and 10-17 provide a summary of sev-
eral types of respondent action to reduce their risk of exposure to hazardous
wastes. Table 10-14 provides a kind of overview of the then current respond-
ent actions. Only 1 respondent (0.2 percent of the total surveyed population)
was taking no action whatsoever. However, 41 respondents (nearly 7 percent)
were using water filters; 175 (or nearly 29 percent) were using bottled water;
and 48 (or nearly 8 percent) were attending public meetings. Table 10-15
reports total respondent expenditures for water filters during the last 5 years.
10-28
-------
TABLE 10-13. NUMBER AND PERCENTAGE OF TOTAL
RESPONDENTS WITH PARTICULAR CAUSE OF DEATH
IN MIND, FOR WILLI NGN ESS-TO-PAY BID
Number and percentage
of total respondents
Cause of death
Number
Percent'
None
321
53.7
Cancer
228
37.8
Lung disease
10
1.7
Lu kemia
17
2.8
Poisoning
4
0.7
All others
20
3.3
"1 don't know"*3
14
_
3
"Percent" column may not total 100 due to rounding.
^Respondents giving "I don't know" answers are excluded from
the population from which percentages are calculated.
TABLE 10-14, CURRENT RESPONDENT ACTIONS TO REDUCE
RISK OF EXPOSURE TO HAZARDOUS WASTES
Action
Number
of total
Number
and percentage
respondents
Percent
None
1
0.2
Using water filters
41
6.7
Using bottled water
175
28.6
Attending public meetings
48
7,8
10-29
-------
TABLE 10-15, TOTAL RESPONDENT EXPENDITURES
ON WATER FILTERS DURING LAST 5 YEARS
Number and percentage
of respondents
Amount, $
Number
Percent3
4 to 10
6
16.2
11 to 20
8
22.0
21 to 30
13
35.1
31 to 40
4
10.9
41 to 100
4
10.3
101 to 300
0
0
301 to 500
2
5.4
a"Percent" column may
not total 100 due to
rounding.
TABLE 10-16. TOTAL RESPONDENT EXPENDITURES
ON BOTTLED WATER DURING LAST 5 YEARS
Number and percentage
of respondents
who bought bottled water
Amount, $
Number
Percent3
0,10 to 10
43
2.3.2
11 to 30
27
14.8
31 to 70
30
16.2
71 to 160
30
16.2
161 to 300
20
10.8
301 to 500
15
¦
8.1
501 to 750
6
3.2
7S1 to 1,000
8
4.3
1,001 to 3,000
4
2.2
3,001 to 10,000
2
1.1
a"Percent" column may not total 100 due to rounding,
10-30
-------
TABLE 10-17. TOTAL PUBLIC MEETINGS ATTENDED
BY RESPONDENTS DURING LAST 5 YEARS
Number and percentage
of respondents who
attended public meetings
Number of meetings
Number
Percent
1 to 2
26
36,6
3 to 5
21
29.6
8 to 10
11
15.5
11 to 15
9
12.7
16 to 25
1
1.4
28 to 50
1
1.4
300
1
1.4
500
1
1.4
3,1 Percent" column may not total
100 due to
rounding,
10-31
-------
As shown, the 41 respondents who used water filters spent between $4 to $500
an water filters during the 5-year period, although most spent a total of less
than $30. Table 10-16 summarizes total respondent expenditures for bottled
water during the last 5 years. The 175 respondents who used bottled water
spent between $0.10 to $10,000 on bottled water, although 80 percent spent a
total of less than $300.00.* Finally, Table 10-17 reports the total number of
public meetings attended by respondents during the last 5 years. The 48 re-
spondents who attended public meetings to learn more about hazardous wastes
attended a total of anywhere from 1 to 500 meetings during the last 5 years,
though most respondents attended only 5 meetings or fewer. Here too, there
appears to be a few implausible responses. For example, two respondents indi-
cated that they had each attended 300 town meetings during the past 5 years--
an average of 100 meetings per year, or about one meeting every 3.5 days,
10.6 SUMMARY
This chapter has described a diverse array of background material impor-
tant to the interpretation of our contingent valuation estimates for household
valuations for risk reductions. Four general themes that emerge from this
overview. First, the economic and demographic characteristics of our sample
respondents closely parallel those of the target population. Second, the resi-
dents of the Boston SMSA have had substantial experience with incidents
involving hazardous wastes. As a result, there has been nearly continuous
coverage of issues associated with hazardous wastes in the Boston Globe and
several local newspapers. There are several external indications of citizen
concern and involvement with the problem. In addition, the local governments
in the area have also had to deal with contamination episodes. As a conse-
quence, citizens have a performance record on which to base their expecta-
tions of government involvement in any future incidents.
Based on both of these considerations we would expect that the circum-
stances described in our contingent valuation question would be familiar. They
were structured, in part, based on experiences in this area. Respondents
can be expected to have had the equivalent of valuation experience as a result
*
*The upper end of the range seems rather implausible and does not, as
e table indicates, involve many respondents.
10-32
-------
of these past episodes and the respective roles of town government and citizen
groups.
Third, our review of the actual information of our survey respondents
confirms what was expected based on the record. On the whole, they do ap-
pear to be aware of the problem. Their perception of the risks involved do
not appear to be irrational responses to these incidents. Rather, their percep-
tion of relative risks seems quite sensible. This is not to suggest that the
risk of exposure is not regarded as a serious problem. Rather, it indicates
that we have not selected a case where a set of frenzied or irrational responses
to recent contamination incidents would condition ail responses to the contin-
gent valuation questions.
Finally, residents have themselves undertaken tangible actions and expen-
ditures that should also provide a basis for gauging their respective valuations
for risk changes associated with regulations governing the disposal of hazard-
ous wastes.
While much of this information has been informal, it is also quite consist-
ent in its message. The degree of familiarity with the problem and past ex-
perience with contamination incidents serve to aid in satisfying the conditions
we expected would need to be satisfied for the development of plausible valu-
ation estimates involving risk reductions within a contingent valuation frame-
work .
10-33
-------
PART 111
PRELIMINARY EMPIRICAL ANALYSIS
-------
PART Hi
PRELIMINARY EMPIRICAL ANALYSIS
Part III of this draft interim report consists of six chapters that describe
the preliminary findings of the empirical analysis of the contingent valuation
survey data for individual's values of changes in hazardous waste risks, Part
iil consists of the following six chapters:
Chapter 11 - Option Price Results: The Framing of the Commodity
and an Analysis of Means
Chapter 12 - Option Price Results: Preliminary Regression Analy-
ses Using Unrestricted Models
Chapter 13 - Valuation Estimates for Risk Reductions: Using Re-
stricted Models
Chapter 14 - The Use of Contingent Ranking Models to Value Expo-
sure Risk Reductions: Preliminary Results
Chapter 15 - A Comparison of Contingent Valuation and Hedonic
Property Value Models for Risk Avoidance
Chapter 16 - Policy Implications and Research Agenda
The objective of this part of the report is to summarize the status of the em-
pirical results at the end of the first phase of the research effort. It should
not be interpreted as completed empirical analyses. The findings are sugges-
tive but require further analysis before they will be regarded as final. In-
deed, based on the work to date, it seems clear that we have only begun to
scratch the surface of the complex issues involved in our research plan.
Some examples of the empirical issues may help to illustrate the prelimi-
nary nature of our analysis. The treatment of "outliers" Is probably the most
illustrative case. In our previous work, we have emphasized the importance
of the treatment of "outliers," or influential observations. Use of regression
diagnostics and judgments about thresholds for influential observations, we
ill-1
-------
suggested that it was possible to identify outliers in a more systematic way.
In this earlier research we expected a relationship between the contingent val-
uation bids and income. In the present effort the issues are much less clear-
cut. Both theoretical and econometric problems must be resolved before con-
sidering the treatment of outliers in a final set of models. This problem is
addressed in Chapter 13.
Equally important, our design provides an alternative basis for dealing
with any survey respondent's incomplete understanding or' acceptance of the
contingent valuation framework. It is possible to estimate the variance in the
error associated with each individual's valuation response in relation to the
models used to explain marginal valuations for risk. Using generalized least-
squares procedures, differences among individuals are explicitly taken into
account. The models with this adjustment differ from the unadjusted models.
Thus, our treatment of outliers is incomplete in this report. It also implies
that our mean values presented in Chapter 11 are presented with "outliers"
included because these responses are not yet identified. Clearly, as suggested
in Chapter 13, this is a crucial area needing more research.
Several other chapters are preliminary for different reasons. For exam-
ple, Chapter 12, which presents some preliminary regression results using un-
restricted models, contains no adjustments for outliers or unequal variances.
In addition, some variable specifications attempted in this chapter, and found
unfruitful, have not been attempted in the more robust restricted models pre-
sented in Chapter 13 due to time limitations. Clearly, this is another issue
for additional consideration.
In addition, the focus of the contingent ranking analysis presented in
Chapter 14 is directed toward attempting to understand exactly how the re-
spondents reacted to the ranking task. In this regard, it explores several
new dimensions of contingent ranking to evaluate the effect of the four ver-
sions in the ranking research design on how people processed the information
provided. However, the models used in this specification testing to provide
preliminary benefits estimates are somewhat ad hoc. There is an obvious need
to evaluate models using more theoretical considerations. This is another area
for further research.
111-2
-------
The comparative analysis in Chapter 15 is preliminary for a somewhat dif-
ferent set of reasons. The most important of which is the preliminary nature
of the models estimated with our survey response and the analysis of the prop-
erty value data performed by David Harrison. Clearly, more detailed analysis
will be required before any final judgments will be possible on the performance
of contingent valuation compared to the hedonic property value model.
The empirical analysis is organized within a specific framework. The first
three chapters--Chapters 11, 12, and 13--form one specific body of analysis.
Chapter 11 begins the investigation of the survey data under the assumption
that survey respondents are homogeneous. It describes the framing of the
contingent commodity, changes in hazardous waste risks, and then examines
the responses classified as "protest" bidders throughout the remainder of the
report. The remainder of the chapter examines many of the key features of
our research design--levels of risk, the role of the conditional risk, the
assignment of property rights, the certainty effect and intrinsic values--by
using tests for mean responses and analysis of variance procedures.
Chapter 12, the second empirical chapter, relaxes the assumption that
individuals are homogeneous in their response to risk changes. It uses multi-
variate regression techniques which specifically examine the influence of indi-
viduals' characteristics on the responses to the design issues. However, this
chapter is primarily exploratory in nature. It provides some preliminary in-
sights about the effect, of different variable specifications. Since the results
of this chapter were somewhat discouraging, they suggested the need for add-
ing more theoretical structure to the analysis.
Chapter 13 is the most detailed empirical chapter. It interprets the option
price bids as providing information for deriving point estimates of the individ-
uals' marginal valuations of a risk change. In other words, it approaches the
analysis of the bids from the perspective of the incremental value for an incre-
mental change in the risk of exposure to hazardous wastes. This chapter also
addresses the nature of the distribution of bids and the treatment of outliers
and provides some generalized least-squares estimates that adjust for the sub-
stantial differences in the estimated variances among individual respondents.
III-3
-------
Chapter 14 is the second unit of empirical research. It follows the same
logical structure developed with the analysis of the contingent valuation bids.
Ranks were first analyzed as if respondents' characteristics had no effect on
how they responded to the combinations of risk and payment presented to them.
This assumption was relaxed and the rank-logst maximum likelihood model used
to estimate random utility functions. Finally, these estimated models were then
used to estimate their implied valuations for risk changes and to develop a
reliminary comparison with the contingent valuation results.
Chapter 15 summarizes the comparison between Harrison's hedonic prop-
erty value model and the position of the survey devoted to developing compar-
able information. It is also preliminary because the hedonic model used was
an early version of the Harrison models and because further consideration will
need to be given to the assumptions used in developing the comparison.
Finally, Chapter 16 offers some discussion of the implications of these
very preliminary analyses and begins the process of outlining an agenda for
further research.
:
;
:
.
III-4
-------
CHAPTER 11
OPTION PRICE RESULTS: THE FRAMING OF THE j
COMMODITY AND AN ANALYSIS OF MEANS 1
11.1 INTRODUCTION
This chapter, the first of five empirical chapters, describes our prelimi-
nary empirical analysis of the features of the valuation responses. Before pre-
senting these results we highlight the role that framing plays in contingent
valuation and describing how the contingent commodity was framed in our
study.
Framing, or the process of describing the contingent commodity in a ques-
tionnaire, is always an important part of a contingent valuation analysis.; In
our case it is especially important because the commodity we describe to each
respondent is a reduction in the risk of exposure to hazardous wastes. And
ultimately, many of the inferences that will be drawn from our research will
depend on this framing of the commodity. However, our framing discussion is
not an exhaustive synthesis of the literature. Rather, it draws on the existing
literature only to the extent it is necessary to provide the rationale behind
the framing of our contingent commodity.
The empirical analysis presented in this chapter has several dimensions.
First, it describes our evaluation of the sample respondents who refused to
participate in the valuation of the contingent commodity, usually termed "pro-
test bidders." This evaluation includes a description of the procedure used
to identify these respondents, a profile of their characteristics, and the effect
of our research design. This evaluation is important because it affects the
subsequent analyses of the valuation responses. Second, it considers the user
and intrinsic values that have been elicited in the survey. To aid the presen-
tation of our user and intrinsic value results, we describe the valuation re-
sponses for each of a set of questions individually. The first set summarizes
our preliminary empirical results for user values measured as incremental option
price bids for risk decreases. Following these findings, the chapter describes
11-1
-------
the effect on option price bids of alternative property rights, certainty as a
risk end point, and alternative risk outcomes. Finally, the chapter presents
estimates for intrinsic values which are viewed as incremental option price bids
for reduced risks to the ecosystem.
11.2 GUIDE TO THE CHAPTER
Section 11,3 presents a preliminary classification of potential biases that
can arise from framing the contingent commodity. Section 11.4. describes
specifically the process of framing our contingent commodity--changes in the
risk of exposure to hazardous wastes. It addresses the development of the
context, character, and the question format to elicit individuals' values. Sec-
tion 11.5 explains our procedure for classifying protest bids, examines the
potential determinants of the likelihood of someone being classified a protest
bidder, and compares our results with those of another recent contingent valu-
ation study. Section 11.6 presents the mean option price bids for reductions
in hazardous waste risk and analyzes the influence on those bids of the initial
levels of risk and the conditional risk--two key elements in our research
design. Section 11.7 presents the mean option' price bids for avoiding an in-
crease in the risk of exposure to hazardous wastes and examines the influence
of risk levels, the conditional risk, and the assignment of property rights on
these bids. It also compares the mean bid for avoiding a risk increase with
those for obtaining a risk decrease. Section 11.8 describes the summary re-
sults for a reduction in the risk of exposure to zero. Section 11.9 considers
the effect of changing the outcome at risk on the mean option price bids for a
decrease in risk. Section 11.10 presents our estimated mean values for intrin-
sic benefits. Section 11.11 concludes the chapter with an overall summary of
its principal findings.
11.3 FRAMING AND CONTINGENT VALUATION
The objective of this section is to provide an interpretive and selective
overview of the literature on the role of framing in a contingent valuation anal-
ysis, The Cummings, Brooks hi re, and Schulze [1984] state-of-the-art assess-
ment , along with research by Mitchell and Carson [1984] and Bishop and
Heberlein [1984], have reconsidered and, as a consequence, revised the previ-
ous conclusions concerning contingent valuation. However, this new research
11-2
-------
I. Conventional Classification II. Mitchell-Carson Classification
A. Genera! Biases
• Strategic
• Information
• Hypothetical
B. Instrument Related Biases
• Starting point
• Payment vehicle
C. Procedural Biases
• Sampling
• Interviewer
A.
B,
D.
Incentives to Misrepresent Responses
• Strategic bias
• Compliance bias
— Sponsor bias
— Interviewer bias
Multiple Valuation
Vehicle bias
Method of provision bias
Implied Value Cues
Starting point bias
Range restriction bias
Yea-saying bias
Relational bias
Misspecification of Market Scenario
Vehicle misspecification
Budget constraint misspecification
Amenity misspecification
Probability of provision misspecification
Context misspecification
Aggregation Bias
Sampling design bias
Nonresponse bias
Item nonresponse bias
Sequence bias
111. Post Palo Alto Classification
A. Framing Biases
• Situation or context
— Interview situation
— Mental image
— Strategic effects
• Commodity specification
— Perceptions
— Property rights
— Implied linkage to behavioral
activities
• Elicitation
— Question format
— Payment vehicle
— Sequence
B. Procedural Biases
• Sampling and non responses
• Interviewer
Figure 11-1. Classifications of potential biases in contingent valuation.
-------
of the contingent commodity itself. Arrow {1984] suggests that information
"bias" is not a bias at all. It could imply explaining the commodity to be
valued in greater detail to make the entire contingent valuation exercise more
realistic. Our classification endorses this view by emphasizing the importance
of context and commodity specification under the framing umbrella. These two
facets of the contingent valuation method provide a more tangible notion of
the potential effects of information on the elicitation of people's values. This
position on information bias is also consistent with the conclusion of the state-
of-the-art assessment for contingent valuation (see Cummings, Brookshire,
and Schufze (1984]).
For example, they observed that:
The information bias ruble seems to serve no useful purpose for
assessments of CVM [contingent valuation method]; indeed, it may
be counter productive, [p. 253]
By eliminating hypothetical bias from the revised classification, the revised
taxonomy accepts the position of Mitchell and Carson [1984],
We conclude that hypothetical bias is a misnomer since there is no
one bias which uniquely results from the hypothetical character of
CV surveys. The hypothetical character of a CV survey may make
it vulnerable to one or more biases and/or it may affect the reliabil-
ity of its findings, [p. 43].
However, they are not alone in this conclusion. Arrow [1984] also has
noted there is nothing inherently wrong with the hypothetical character of con-
tingent valuation. To support this view, he cited the number of new products
that are introduced each year that likely were evaluated for the market by
potential consumers when they involved hypothetical elements. Yet, Arrow
does add caution about drawing conclusions solely from contingent valuation.
That is, without the discipline provided by "real" payments for commodities,
there is potential for inaccuracy. This point is also consistent with Bishop
and Heberlein [1984] who argue that their simulated market experiments pro-
vide better estimates of value because actual cash transactions are involved.
However, Mitchell and Carson [1984] in their reinterpretations of both
the early Bishop-Heberlein [1373] study and the Bohm [1971] study, challenge
the position that actual cash transactions are necessary for eliciting "accurate"
estimates of willingness to pay, Cummings, Brookshire, and Schuize [1984]
11-5
-------
concur with the Mitchell-Carson view by citing the experimental results of Ver-
non Smith and the results of various comparisons as providing additional evi-
dence of the accuracy of contingent valuation for public goods that satisfy
their reference operating conditions. Unfortunately, as both Smith [1984c]
and Freeman [1384b] have observed, these conditions are usually satisfied
when one would be least likely to need contingent valuation.
Therefore, our conclusion is that treating the hypothetical character of
contingent valuation as a "bias" is confusing. Instead, it is both the strength
of the approach and its greatest weakness. Because it can be based on hypo-
thetical commodities and circumstances, contingent valuation offers a wide
range of possibilities for addressing many different problems, in effect, it is
a malleable approach that can be shaped to meet the needs of the problem at
hand. Yet, this malleability and its basically hypothetical character expose
contingent valuation to the pitfalls associated with describing the hypothetical
situation (whether commodity or circumstances governing the provision of a
known commodity) in sufficient detail to make it tangible and believable for
respondents to a contingent valuation survey. Unfortunately, the existing
body of research is inadequate for obtaining a definitive answer to questions
raised by the hypothetical character of contingent valuation. Nevertheless,
this does not imply it should be treated as "bias"; it is an attribute of the
method itself.
11.3.2 Context
Context is an important element in the framing section of Figure 11-1.
In our use of the term, context consists of the physical setting in which the
interview takes place and the mental setting or milieu (see Mitchell and Car-
son [1984]) that is created by the survey questionnaire.
Context: Physical Setting
The contingent valuation literature contains little information on the effect
of physical setting on the outcome of an interview. For example, Mitchell and
Carson {1984] note that the survey research literature has indicated some con-
cern that the respondent may feel the need to accommodate the "visitor" (the
interviewer) and try to provide responses that he feeis the interviewer wants
to hear if the interview is conducted in his home. To minimize the potential
of this "bias" occurring, Mitchell and Carson [1984] and Desvousges, Smith,
11-6
-------
and McGi vney [1983] have told interviewers to emphasize the notion that there
were no right or wrong answers. In both surveys, interviewers prefaced each
interview with that philosophy.
Another aspect of the possible influence of the physical setting is that
respondents are "on their own turf" when they give their responses in most
contingent valuation surveys. In effect, they respond to the "hypothetical"
situations in the same setting in which they are likely to make many of their
household decisions. Moreover, the respondents can set certain basic ground
rules for the interview. For example, they can simply ask the interviewer to
leave if they find the questions annoying or troublesome.
While it is unclear exactly what effect the setting has on an interview, it
does differ considerably from the setting, usually a laboratory or a classroom,
in which the majority of psychology and experimental economics data are col-
lected. Although, the survey questionnaire still sets the terms under which
the contingent commodity is offerred, the home setting may put the respondent
more at ease in answering questions. This does not imply that empirical evi-
dence from a laboratory setting is not relevant to contingent valuation. What
it does suggest is that the differences in physical setting and frequently in
the types of respondents--college students are the usual respondents in the
laboratory setting--may complicate the transfer of learning between the two
environments. *
Context: Mental Setting
Mitchell and Carson (1984] suggest that the mental setting created within
the contingent valuation survey is even more important than the physical set-
ting. This aspect of context is the atmosphere or milieu that the contingent
valuation questionnaire establishes. Poster boards with pictures of different
vistas from the Grand Canyon, questions about familiar household activities
like recreation, and general attitudinal questions on a particular theme are all
examples of how a mental image can be established in questionnaires. Again,
there appears to be little or no research that has systematically tested for the
*lt would be interesting to compare the results of Charles Plott and Ver-
non Smith who have conducted experiments using nonstudent subjects with
those with student subjects. To our knowledge, this has not been done in a
wide variety of problem settings.
11-7
-------
effect of this dimension of the context for individuals' valuation responses for
contingent commodities.
As noted in Chapter 8, our experiences with the focus groups and video-
tape interviews also indicated the importance of context. In fact, the sessions
themselves elicited ideas that aided in creating the mental setting and later in
qualitatively evaluating their effectiveness. Yet, this is not a substitute for
a well-designed empirical test of context effects.
The Post Palo Alto classification (PPAC) in Figure 11-1,still retains a pos-
sible role for strategic effects in contingent valuation as an element in context
effects. This concern for strategic effects emanates from sources other than
the usual ones. That is, while it is possible to agree with Cummings, Brook-
shire, and Schulze's [1984] conclusion that there is virtually no evidence of
strategic behavior in almost all previous contingent valuation surveys, there
is nonetheless a type of strategic response in certain contexts. For example,
using contingent valuation in the siting of undesirable facilities or for some
other highly emotional issue, respondents can attempt to engage in strategic
behavior to try to influence the outcome. While the possibility of strategic
behavior may be a limited one, it may not be prudent to conclude on the basis
of past studies where the issues may not have been as closely tied to the local
interests of the respondents that strategic bias would not arise tr» other con-
texts ,
The specification of the contingent commodity is a prominent part of the
overall framing process. That specification must consider the procedures used
by individuals to form perceptions and their ability to process the information
provided. It should also be cognizant of the implied links that are presented
between the valuations elicited and the behavioral actions described. As a
rule, the conceptual foundations for these links come from economic theory.
Thus, a description may have implicit maintained hypotheses concerning feasi-
ble responses available to the household. In fact, we consider these features
as the basic elements of the contingent commodity itself. That is, it is diffi-
cult, if not impossible, to interpret the values elicited in contingent valuation
independent of the specification of the commodity. This seems to be consistent
with both Randall [1984] and Arrow 11384], This is one reason why our con-
tingent valuation commodity, changes in the risk of exposure to hazardous
11-8
-------
wastes, was described under two different property rights allocations to assess
the relative importance of this aspect of the specification.
The perceptions component of the commodity specification refers to how
people perceive and process the information used to describe the commodity.
While some of the perception issues were discussed in Chapter 7, several other
aspects of perception deserve attention. For example, Tversky and Kahne-
man's [1981] mental accounts concept,, or the notion that people process infor-
mation and make allocation decisions by grouping items into "aggregate accounts
like recreation, food, and housing, can be interpreted as falling under the
perceptions component of the commodity specification.* in effect, they are
suggesting that the way people process information to make decisions may
affect how they perceive the commodity.
In addition, Mitchell and Carson's [1984] part-whole bias can be viewed
as being part of how people perceive the specification of the contingent com-
modity, In their words, part/whole bias arises when the respondent views
Lhe contingent commodity differently than the researcher. For example, the
researcher might have attempted to elicit a value of improved water quality in
all the nation's water bodies while the respondent may very well be providing
a value for only part of that, i.e., for a particular water body. Thus, part/
whole bias occurs when people's perceptions are quite different from those of
the researcher who designs and then interprets the results derived from a
contingent valuation analysis. The PPAC classification considers part/whole
bias as a part of framing the contingent commodity that deals with people's
perceptions of the commodity.
Another key element of commodity specification in our classification is the
elicitation process or, more specifically, the various parts of the eiicitation
process. This process can be viewed as consisting of the question format,
the type of question used to eiicit the value, and the payment vehicle that
denotes the terms in which the hypothetical payment would be made. Chapter
7j has discussed the importance of question format as part of the elicitation
*This concept should not come as a surprise to economists. It is nothing
lore than a specification of additional structure on the utility function com-
pletely consistent with the budget decomposition assumptions made in the theory
ssociated with developing aggregate price or quantity indexes (see Blackorby,
'rimont, and Russell [1978] for example..
11-9
-------
process. Therefore this discussion will not consider these issues beyond a
recognition that the existing evidence seems to suggest that question format
can have an impact on the individuals' valuation responses for contingent valu-
ation commodity.
The payment vehicle is also a crucial part of the elicitation process. In
this case, the research on the effect of payment vehicle does not seem as, well
established or as well documented as that for question format. For examp'e,
Cummings, Brookshire, and Schulze [1984] were unable to "establish any defi-
nite problem arising from payment vehicles. Yet, they were unwilling to dis-
miss it as a potential problem in contingent valuation. Mitchell and Carson
(1984] point out that perhaps one of the most sensitive aspects of the payment
vehicle may be the implied value that results from the payment vehicle. That
is, the vehicle itself may imply a specific starting point to people. In effect,
it may provide them an implicit anchor for their responses. For example, when
one is asked to make a hypothetical payment in the form of a utility bill, what
comes to mind is one's typical monthly bill, from either the gas or electric util-
ity. However, if one is offered a payment vehicle that is a user fee, e.g., a
pass to use a lake during a year, then a range of comparable user fees like
$5 to $10 (per person) more than likely comes to mind. Thus, the Mitere l/
Carson position is that payment vehicles may create problems for the elicitation
process similar to the anchoring problems that arise with different question
formats, particularly the bidding games. Ultimately, more research is needed
to verify this position.
In addition, Arrow (1984] has noted that he does not find it surprising
that different payment vehicles would result in different values but for reasons
other than anchoring. Arrow's position is that the institutional arrangements
by which payments are to be made are an integral part of the contingent valu-
ation commodity itself. Our classification seems consistent with the Arrow posi-
tion in that we have placed vehicle bias underneath the commodity specifica-
tion .
The lower part of the PPAC classification deals with procedural issues.
Two of the most important procedural questions considered in regard to contin-
gent valuation are sampling bias and interviewer bias. Unfortunately, there
is little evidence on the potential effects of alternative sampling procedures on
11-10
-------
contingent valuation estimates. In fact, with the exception of Mitchell and
Carson [1984], the literature is almost devoid of any consideration of interac-
t ons between the procedures used to select the sample and other research
design considerations.
The second form of procedural bias, interviewer bias, has received some
attention in the contingent valuation literature. This bias results from indi-
vidual interviewers affecting the valuation process. Desvousges, Smith, and
Fisher [1984] have observed from a survey to measure the benefits of improved
ater quality that a couple of interviewers seemed to have a differential effect
people's bids. However, these effects were not widespread and seemed
t to have a significant overall impact on the valuation estimates. On the
.her hand, Boyle and Bishop [1984] do find some evidence of interviewer
ects in their recent study of scenic beauty on the Wisconsin River.
In summary, what has been defined as the PPAC classification is an
tempt to provide a brief description of the evolution of thought on the prob-
lems in using contingent valuation. It is also a preliminary attempt at synthe-
sis that is intended to pre ide background for the specific framing decisions
made as part of the present c_.itingent valuation. Clearly, a final classification
awaits both more research and more thorough reflection.
11.4 FRAMING THE COMMODITY: REDUCTIONS IN HAZARDOUS WASTE
RISKS
This section describes how we framed the contingent commodity for this
research — reductions in the risk of exposure to hazardous wastes. Four key
aspects of our contingent commodity are examined:
Behavioral actions implied by our conceptual linkages for indi-
viduals' responses to risk
Context for the contingent valuation questions
Procedures used to specify the commodity
Procedures used to elicit valuation.
This section also describes the framing of the initial commodity that was
presented to the respondents. Later sections of the chapter will highlight
the variations on this initial on"iodity that follow from our research design.
For example, the variation used to examine the effects of property rights is
11-11
-------
deferred until the section on empirical results that presents these estimates.
This pattern is maintained in the remaining empirical chapters,
11.4.1 Conceptual Linkages
The theory of welfare measurement generally provides the basic guidelines
for the overall definition of the valuation concepts used in a contingent valua-
tion survey. For example, the debate over Greenley, Walsh, and Young's
[1981] estimates of the values of changes in water quality relates to the con-
ceptual basis for their framing of the contingent commodity, They have inter-
preted their description as providing measures of option value while others
(Mitchell and Carson [1982] and Desvousges, Smith, and McGivney [1383])
have argued that they elicited two slightly different measures of option price.
Thus, the importance of the conceptual foundations in contingent valuation
follows from the intended use of the results: to obtain measures of individ-
uals1 values of commodities that are not exchanged in conventional markets.
Without its conceptual linkages, contingent valuation estimates can be difficult
if not impossible to interpret. An additional aspect of what we have referred
to as conceptual linkages concerns the behavioral responses that are described
or implied by the question to be feasible actions available to the respondent.
The conceptual foundations for our contingent valuation survey were
developed in Part I of this report. Our objective in this section is a more
limited one; to highlight the measurement guidelines that the conceptual analy-
ses of both user and intrinsic benefits provides for our survey questionnaire.
For more detailed explanations of the rationale underlying the guidelines, the
reader is referred to Part I, This section first considers the implications of
the conceptual analysis of user benefits and then discusses the same topic as
it applies for intrinsic benefits,
"User" or Household Benefits
The basic valuation measure used in our conceptual analysis to define
what we refer to as the user benefits from a change in the household's risk
of exposure to hazardous wastes as the increment to planned expenditures re-
quired to maintain the constant expected utility. There are several aspects
of this valuation concept that are important to our contingent valuation analysis
and to the specific process we used to implement the framework. First, it is
11-12
-------
ojf si
Our
ah ex ante welfare concept, defined as if the individual were capable of making
payments prior to knowing the outcome of the events at risk. This is an ana-
lytical method for describing the individual as planning consumption choices,
contingent upon the events at risk, rather than explicitly making those
choices. To use this framework within a contingent valuation setting we must
describe the institutions that organize (or restrict) how contingent payments
are made, in effect, is it possible to precommit to different payments now
that would vary based on the events that do take place? Or must one specify
a payment now for a desired outcome (in our case a change in the likelihood
some event) that must be made regardless of what the actual events are?
description selects the second and implies that the change in planned
expenditures will correspond to the option price payment that would be made
for the risk change,*
The individual's valuation response, or hypothetical payment, is then the
maximum constant payment the individual would be willing to make to obtain
the risk change. The indiv dual is willing to pay the option price because
the payment, by reducing his risk of exposure to hazardous wastes, wilt enable
him to obtain the same expected utility level with a lower risk of the detrimen-
tal event--exposure to hazardous wastes. In effect, the option price is the
d fference between the individual's planned expenditures made before the risk
change and those made afte- ^e risk changes. However, this difference is
an option price, only if the mdi 'iduat has no other avenues of state-dependent
adjustments. If these avenues are perceived to be available, then the valua-
tion response is conditional on these perceived opportunities for state-depend-
ent adjustments.
In fact, an individual who has different avenues of adjustment available
may well have different valuation responses for risk reductions as the level of
the risk changes than would be predicted under the assumption of option price
payments. This suggests caution in interpreting the payment amounts in our
survey. An alternative interpretation may be that we have not fully reflected
the perceived "planning process" individuals think they can use when faced
w th uncertainty. Several parts of the questionnaire elicit information about
*For more discussion of this result see Chapters 4 and 5.
11-13
-------
some of these avenues of adjustment to aid in interpreting both the plausibility
of the responses themselves and their implications for benefits measurement.
Intrinsic Benefits
As we noted earlier in this report, we are using intrinsic values and
existence values as synonyms. This departs from earlier benefit taxonomies
(including our own adaptation of the Mitchell-Carson [1381] work presented in
Desvousges, Smith, and McGivney [1983)), It is a deliberate departure be-
cause these past efforts mixed an ex ante and an ex post perspective for wel-
fare analysis. We have argued that it is probably not desirable to try to dis-
tinguish other nonuser components of valuation from existence values. Institu-
tions for ex ante adjustment also affect intrinsic values. Nonetheless, our
conceptual analysis in Chapter 6 assumes that the option price is the appropri-
ate welfare measure. That Is, the focus of our contingent valuation question
that poses reductions in the risk of exposure to hazardous wastes that "crit-
ters" experience in their natural ecosystem should be to elicit constant ex ante
payments for these risk reductions. As with user values, the payments are
independent of the state of the world that actually occurs.
However, there are three important features that distinguish these values
from the user values. First, the outcome at risk is the risk of exposure and
possible death for the creatures themselves, not the household. Second, the
events at risk are specifically described as not implying the extinction of a
species. Third, the endpoint of the risk change is described in somewhat
vague terms: to the level these creatures face in their natural habitats. It
is likely that different individuals will perceive this endpoint differently. As
noted in Chapter 7, both the outcome at risk and the risk endpoint may influ-
ence individuals' values for changes in hazardous waste risks. Thus, it is
not possible to develop estimates of the values for reductions in these risks
on a per-unit basis.
11.4,2 Context
This section briefly describes the physical setting or locational context
in which the interviews were conducted and the "mental image" or context that
the questionnaire attempted to set for respondents.
11-14
-------
;
; Physical Setting
'
Generally, our interviewers conducted the interview within the confines
of the respondent's residence, often as not seated at the respondent's dining
room or kitchen table. The interviews usually were conducted at a prear-
ranged time. However, in some cases they were completed at the same time
that our interviewer compiled the list of household decisionmakers. On aver-
age, the interviews lasted 53 minutes, though some lasted as long as I5! hours.
Despite the length, oniy three of the interviews were not completed after initi-
ation (see Chapter 9 for details).
The interviewers prefaced each interview with a statement that there were
no right or wrong answers and that the respondent could refuse to answer
any question or simply reply "I don't know." During the training sessions
and practice interviews, the interviewers were reminded of the importance of
this preface. Thus, the main intent was to keep the physical context as com-
fortable as possible for the respondent and to minimize the opportunity for
implying that the interviewer was interested in any particular response. The
interviewers identified themselves as employees of the Research Triangle insti-
tute. No mention was made of the sponsoring agency, either before or after
the session.
Context: Mental Setting
As noted in Chapter 8, the focus group and other questionnaire develop-
ment activities consistently pointed out the importance of establishing an effec-
tive mental setting with the survey questionnaire. Mitchell and Carson [1334]
argue this point quite persuasively based on their efforts to develop their
questionnaire for eliciting values of improvements in national water quality.
The final context established by our questionnaire resulted through a
trial and error process documented in Chapter 8. The questionnaire opened
with a general question asking the respondent to rate the potential harm to
people and the environment itself from different sources of pollution, including
hazardous wastes. This was the first mention of hazardous wastes during the
interview with the intent being to-elicit a relative rating of hazardous wastes
to other pollution sources prior to providing any information about the main
questionnaire topic.
11-15
-------
Next, the interview turned the focus to hazardous wastes by defining
them and then differentiating between radioactive wastes and hazardous wastes
with the stress on factories and landfills to help make the distinction. Figure
11-2 shows the visual aid that the interviewer gave to the respondent to hefp
create the mental image. While the respondent was looking over the card, the
interviewer said the following:
To give you an Idea of what hazardous wastes are and where they
come from, here's a list of some products we use every.day and some
wastes that are left over after they're made. For example, a common
waste is the chemical solution used to tan the leather in shoes, wal-
lets, or purses. After the chemical solution is used, it must be
thrown away. Because the solution contains chromium, it's consid-
ered a hazardous waste. Hazardous wastes are left over after mak-
ing a wide range of other consumer products — from the gasoline and
batteries for cars to the plastic containers used to package and store
food. Some companies put these wastes in their own special facili-
ties ; others pay companies to dispose of their wastes in special
dumps called hazardous waste landfills. Some products that we use-- ;
like paint, varnish removers, and weed kiilers--are themselves con-
sidered hazardous wastes when we throw them away. Although haz-
ardous wastes often have been handled carefully, sometimes the
practices have been inadequate.
The next two building blocks for context involved eliciting the frequency
at which the respondent had obtained information about hazardous wastes and
the name of the nearest factory that produced hazardous wastes and its dis-
tance from the respondent's residence. The second block involved obtaining
a rating of the respondent's perceived effectiveness, at the time of the inter-
view, of different organizations in dealing with hazardous wastes. Among
those included in the list were several levels of government and generators of
hazardous wastes.
The interviewer* next moved into the crucial section of the questionnaire,
which included several perception questions. These included the respondent's
perceived likelihood of exposure — using a 1-10 scale card -- to hazardous wastes
from various environmental media and the use of the risk ladder to obtain the
respondent's perceived annual risk of dying from different causes during the
next year. The interviewer used the multicolor ladder, shown earlier in Figure
8-1, to ask about death from auto accident, heart disease, illness caused by
air pollution, and an illness caused by hazardous wastes. Thus, these percep-
tion questions constituted the last links in the chain of context in the ques-
11-16
-------
Products and Their Hazardous Wastes
Consumer Products
Automobile batteries
Dry cleaning fluid
Paint/textiles
Shoes and other
leather goods
Glass/electronics
Steel
Plastics
Pesticides—afdrio, dieldrin
DDT, chlordane
Chemical and
petroleum products
Pharmacy products
Discarded Hazardous Substances
Lead
Carbon tetrachloride
Chromium, chlorinated organic
compounds
Chromium
Selenium
Manganese, phenols,
benzene
Vinyl chloride
Chlorinated organic compounds
Phenols, benzene, organic ,
compounds, brines
Organic solvents
Figure 11-2, Hazardous waste information card.
-------
tionnaire before the specific commodity of hazardous waste risks were intro-
duced .
11.4.3 Contingent Commodity Specification
The questionnaire specified the contingent commodity in four steps. The
steps include describing hazardous waste as a risk, explaining the payment
vehicle, specifying the ground rules for the valuation process, and, finally,
highlighting the character and circumstances of hazardous waste risks.
Hazardous Wastes and Risk
The first step in specifying the commodity is describing the concept of
hazardous wastes as a situation involving risk. Table 11-1 shows the; text
the interviewer used to introduce the concept of hazardous waste as risk. In
this text, the interviewer also explains the risk circles or probability wheels
that are the visual aid used to communicate risk. Figure 11-3 shows one card
with risk circles that were described.
To aid the respondent in using the risk circles, and to provide a link to
the hedonic property value study, the interviewer handed the respondent a
second card with risk circles and asked him to translate the risk change; into
a distance, in miles, that would provide an equivalent risk reduction. The
question was posed as a hypothetical situation involving a chemical contamina-
tion of the local drinking water supply.
The Payment Vehicle
The second step in specifying the commodity is explaining the payment
vehicle that would be used in the elicitation of values. The interviewer first
introduced the general idea of the payment vehicle;
Next, I would like you to think about the costs of more controls on
hazardous wastes. When the government decides to clean up aban- j
doned dump sites, place stricter controls on landfills, or stop some I
very toxic wastes from being generated, these actions would reduce ;
the risk of exposure. However, they cost someone. As consumers
and as taxpayers, we pay for the costs of better control of hazard- ;
ous waste.
After the introduction, the interviewer handed the payment vehicle card,
shown in Figure 11-4, to the respondent and then described it by saying:
11-18
-------
TABLE 11-1. THE INTRODUCTION TO HAZARDOUS WASTE
AS A RISK
Another way to think about hazardous wastes and risk is with this card. It
uses circles to stand for two different kinds of risks we face from hazardous
waste.
Since risk involves chance, we can also think of risks by putting pointers
that would spin easily on each of the circles. A pointer has an equal chance
of landing at any spot on its circle. The larger the portion of the circle that
is "cut out" by the blackened area--that is, the bigger the slice—the more
likely the pointer would land there. On the first circle on Card A, for exam-
ple, 20 percent of this circle is blackened. There is one chance in 5, or 20
percent chance, the pointer will land in the blackened area. This means that,
on the average, for every 100 spins the pointer would land in the blackened
slice twenty times.
The numbers on the cards are hypothetical because even experts disagree
about the sizes of the risks. However, in the rest of this interview, J_ want
yoti to think of these numbers as actual risks you face.
Look at the differences between each circle. The first circle shows the risk
cr chance that you (and your household members) would be exposed to haz-
ardous waste. By exposed, I mean touching, breathing, eating, or drinking
a large enough amount of a hazardous waste over a period of time so that it
could be harmful. Exposure through the pathways we have discussed could
be a brief/ one-time thing, or it could happen over several months or years.
The importance of the middle circle is that it stands for the second, and dif-
ferent, type of hazardous waste risk--the chance of dying after being ex-
posed . This means that even if you're exposed, there's a separate chance-
not a certainty--that you would die. For example, some people are healthier
or have better resistance. Whether or not you're actually harmed is based
upon your physical makeup, heredity, and overall health. An important thing
to remember about the first two circles is that you would never have to spin
the pointer on the second circle as long as the pointer on the first circle
never landed in the blackened area. In other words, there's no chance you
wouId die from the effects of hazardous wastes if you're never exposed to
them.
The third circle combines the two types of risks into a person's overall risk.
It shows the bottom line: your chances both of being exposed to hazardous
wastes and, once exposed, dying. The combined risk of exposure and death
is found by multiplying the chance you see in the first circle by the chance
in the second circle.
11-19
-------
Risk of Exposure
Risk of Death
if Exposed
1
10
110 percent)
1
i
10
50
(20 percent)
Possible
Pathways
Heredity
and Health
Figure 11-3, Risk circles.
Card A
Combined Risk: '
Exposure and Death
Personal
Risk
-------
Card 8
How We Pay for Pore Control of Hazardous Waste
Products We Buy
Automobile and
petroleum, products
Shoes and other
leather goods
Chemicals, plastics,
carpet and other
floor coverings
h
A
Pesticides in the
home and yard
Existing Controls
Existing Exposure Risk
and Product Prices
More Controls
Lower Exposure Risk
with Higher Product Prices
Higher Tax Bills
Existing Tax Bills
Investigations Education investigations PuMc Education
„ , \ and Information \ and Information
Enforcement \ —V——
Research
Enforcement
Research
Buyouts and
Relocations
Figure 11-4, Payment vehicle card.
11-21
-------
The top part of this card shows how we would pay for lower expo-
sure risks through higher prices for the products we buy. If the
government puts stricter regulations on car makers, shoe companies,
or chemical companies, it would cost them more to make their prod-
ucts. Then if you buy a pair of shoes or a pesticide, you would
pay a higher price than you would without the regulations.
The lower part of this card shows how we would also pay for lower
exposure risks through higher local, state, or federal taxes. The
card shows the higher tax bills providing more money to investigate
and enforce the regulations and to clean up places like Times Beach
or Love Canal.
We chose higher prices and taxes as the payment vehicle for several rea-
sons. First, it has no implied starting value like a utility bill or user fee.
Second, it corresponds closely with how people actually pay for regulations on
hazardous wastes. Even though our hypothetical situation was structured in
terms of a local company located 3 miles from the respondent's house, our
focus group experiences suggested that the general vehicle was more tangible
to respondents., making it easier to comprehend than trying to develop a
hypothetical vehicle that would be tied directly to the local situation. Finally,
this payment vehicle had proved effective in several previous contingent valu-
ation studies, in particular Mitchell and Carson [1981, 1984] and Desvousges,
Smith, and McGivney [1983],
In summary, our payment vehicle is a practical compromise between the
need for credibility and comprehension and the need for consistency with the
hypothetical situation. The effectiveness of this compromise is an empirical
issue that is considered later in this chapter in the evaluation of the reasons
for protest bids. If, there was insufficient correspondence between the com-
modity, the circumstances under which it occurred, and the method of pay-
ment, we anticipate there would be a sizable percentage of participants who
would reject the terms of the market.
Valuation Ground Rules
Explaining the ground rules for the valuation exercise to the respondent
is the third step in specifying the contingent commodity. This step consists
of three key parts: informing the respondent in advance of the sequence of
valuations, offering the opportunity to review this sequence, and benchmarking
the valuation perspective.
11-22
-------
The interviewer explained the sequence of the valuation exercise by re-
tewing the three risk circle cards with the respondent and describing how
they would be used. The interviewer said:
Now, think about these cards and about paying higher prices and
taxes. Based on a hypothetical situation, I'm going to ask you some
questions about paying to reduce your (and your household mem-
bers') risk of exposure from the level on Card A to the levels on
Card B, and Card C, As you can see on the cards, the risk of
exposure decreases in the first circle from 1 chance in.5 on Card A,
to 1 chance in 10 on Card B, to 1 chance in 25 on Card C. It also
means your combined risk of exposure and death gets smaller each
time.
After asking about paying for these risk reductions for people, I
am going to ask about paying an adcitional amount to reduce risks
for fish, wildlife, and plants only--not for humans. Do you have a
question about how I am going to continue?
This prenotification is important for three reasons. First, it enabled the re-
spondent to be informed in advance of what he was going to be asked to do.
This was necessary because focus group participants pointed out that they
likely would have divided their bid differently between the two risk levels, A
to B and B to C, had they known they were going to be offered a second
level. Thus, we added the advance notice- and explained the incremental na-
ture of the intrinsic value question.
The second reason accounts for how some of our respondents were able
to refuse to pay anything for the first change but were willing to make an
option price bid for the entire change from A to C later. These bids of these
respondents have been evaluated separately and are discussed later in this
chapter.
Finally, the prenotification gave the interviewer the opportunity to ask
the respondent if he would like to review these terms prior to the actual valu-
ation exercise. This proved useful because it minimized the need to review
even more material if the respondent experienced difficulty later in this part
of the interview.
The second part of the ground rules for the valuation exercise involved
the interviewer establishing two mental benchmarks for the respondent to use
in developing his responses. The interviewer said:
11-23
-------
Before I go on, there are two things to keep in mind. One, please
decide how to respond as though you actually were facing this hypo-
thetical situation. In other words, 1 would tike you to keep in mind :
your (and your household members') income, how you budget your
money, the kinds of products you buy and the taxes you pay. Two,
any amounts that you're willing to pay would be in addition to what
you're now paying for hazardous waste controls and would affect
only hazardous waste problems. The amounts are not to reduce acid
rain or any other environmental problem.
These benchmarks are important because they help to orient respondents'
thought processes for this crucial part of the interview.
Character and Circumstances of Risks: The Hypothetical Situation
The last step in the sequence for specifying the contingent commodity
consisted of the interviewer explaining the hypothetical situation. This situa-
tion is important because it described the specific circumstances under which
the respondent would experience the risk and outlined the features, or attri-
butes, of that risk. Figure 11-5 shows the card that the respondent received
as a reminder of the exact circumstances.
There are three important aspects of the hypothetical situation. First,
it describes the commodity—the change in the risk of exposure to hazardous
wastes — from the level shown on Card A to the level on Card B--and it links
this exposure to tangible actions, the use of liners and a monitoring system,
in response to government regulations.* Our focus group experiences contrib-
uted substantially to the use of this specification. Participants suggested that
such concrete terms were necessary to make it tangible. The importance of
concrete terms and examples seems consistent with the psychology literature,
in particular Slovic, Fischhoff, and Lichtenstein [1982] and Wallsten and
Budescu [1983 j.
The second important aspect of the hypothetical situation is that it
describes the timing of the risk. That is, death from the exposure would not
occur until 30 years later. This aspect is important for two reasons. First,
it may account for some of the differences among individuals in their responses
*The only other attempt to use contingent valuation to value regulations
involving hazardous wastes emphasized the inherent uncertainty in the process
and asked for the valuation of a regulation in the presence of this overall un-
certainty. See Burness et al. [1983] for further discussion.
11-24
-------
Card 7
Exposure Risk Circumstances
• Electronic parts company
• Located 3 miles from your home
• Generates 2,000 gallons of hazardous waste
each day
• Company disposes of the wastes in a landfill at
company site
• If you are exposed, there is a chance you will
die in 30 years
Figure 11-5. Description of hypothetical situation.
-------
to the risk. For example, older people may well view the risk as less relevant
to them because of its timing. Second, the timing makes it difficult, if not
impossible, to compare values from our situation with many of those in the
existing literature (see Violette and Chestnut [1983] for a review). For exam-
ple, the outcome of risks in labor markets that are estimated with hedonic wage
models relate to annual risk. While we present a simple and crude attempt to
compare these valuations in Chapter 16, it is prudent to view this as an area
for further research. This comparison is intended to highlight the issues in-
volved. As noted in Chapter 7, there is substantial evidence that the types
and attributes of risk are likely to be important considerations in how people
value risk changes.
The third important aspect of the hypothetical situation is that it does
not specify a particular cause of death. Consequently, we asked people if
they had a cause of death in mind, and if so, which one. As noted in Chap-
ter 10, a sizable majority of the people, not surprisingly, envisioned cancer
as the cause of death. This provides a good indication of individuals' per-
ceived character of hazardous waste risks.
In addition, not specifying a cause of death allowed us to modify the situ-
ation to describe a particular cause--immune system damage--and a different
type of risk--birth defects--and ask the respondents if they would like to
change their bids if the nature of the event at risk was modified in either of
these ways. Finally, the hypothetical situation provided a baseline for the
intrinsic values question and the reduction in risk to zero to address the
importance of these two parts of our research design.
11,4.4 Elicitation of the Option Pr ce Bids
The final task in framing the commodity is eliciting the individual's value
for the contingent commodity. While we have noted the importance of this step
throughout this report, Cummings, Brookshire, and Schulze [1984] add another
dimension to the importance of the elicitation process. They suggest that this
part of the framing of the contingent commodity provides the opportunity to
mitigate attitudinal bias as in contingent valuation. This bias occurs when an
attitudinal response is interpreted as an indication of behavior, a potential
problem noted by Bishop and Heberlein [1979, 1984]. Cummings, Brookshire,
11-26
-------
and Schulze [1984] suggest that by carefully structuring the elicitation proc-
ess, contingent valuation can be interpreted as "intended behavior,"
Our elicitation procedure consisted of the following:
Interviewer statement:
Purpose:
Interviewer statement:
Purpose:
Interviewer statement:
Purpose:
Interviewer statement:
Purpose:
Think about your monthly income and what
you spend it on in your budget.
Reminds respondent of budget constraint
and how they typically make expenditures.
How much would you be willing to pay each
month..,
Gives specific action/time frame
in higher taxes and in higher prices for
products you buy
Explains specific action
to lower your (and your household mem-
bers1)* risk of exposure from the level on
Card A to the level on Card B?
Identifies a specific target.
In this procedure, the interviewer used the direct question format that was
discussed in Chapter 7. The same procedure was also followed for the second
risk change. Chapter 14 will describe the contingent ranking format and how
we used it to elicit individuals' responses.
Finally, the interviewer offered the respondent an opportunity to revise
his bid. By transforming the monthly bid into its annual equivalent and veri-
fying its accuracy with the respondent, the interviewer enabled the respond-
ent to reconsider if he thought his bid was either too high or too low. This
rechecking helped to minimize problems if the respondent did not fully appre-
ciate the magnitude of the monthly amount, which Mitchell and Carson [1384]
refer to as the "easy monthly payments" syndrome, tt also allowed the re-
spondent to reconsider his response with a minimum of pressure. Equally
important, the procedure does not change the terms under which the risk re-
duction is to be provided, a potential problem we discussed in Chapter 7 with
the iteration process that has been used in several past studies.
~Parentheses imply that interviewer only read if more than a single per-
n household.
11 -27
-------
This opportunity for revisions was provided for each valuation question
that appeared in the questionnaire. Table 11-2 profiles the bidders who re-
vised their bids and the questions or valuation circumstance in which the revi-
sions occurred. Somewhat surprisingly, the revisions were distributed almost
evenly between increases and decreases. Of the total of 30 revised bids, 13
bids were increased and 17 were decreased. We had expected individuals
would primarily lower their bids because of underestimating the annual equiva-
lent of monthly amounts. Another important feature is the infrequent occur-
rence of the revisions. If the direct question format for the valuation ques-
tions for risk changes are totaled across all sections of the questionnaire,
2,038 valuation responses were obtained, which implies a revision rate of 0.001,
or about 1 out of every 1,000 bids. :
There does not seem to be a particular pattern to the revisions. Most of
the revisions appear plausible, afthough there is one potential exception.*
The only other characteristic that seems prevalent among the revised bidders
is that they had rated the potential harm of hazardous wastes toward the more
harmful end of the scale. Yet, this characteristic does not seem to correlate
with the direction of the revision.
Finally, we are somewhat unsure of how to interpret the inf requent y of
revisions, but it does not seem to indicate any implausible behavior on the
part of the majority of the respondents. Nevertheless, it does offer another
check on the consistency of responses that has not been widely used in prior
studies. Our finding also seems consistent with Mitchell and Carson [1984j
who used a slightly different procedure but with the same objective.
11.5 PROTEST BIDDERS
This section describes the protest bidders from various sections of the
contingent valuation questionnaire. In addition, it examines characteristics
*There is one individual who revised his bid from $80 to $500 a month
for the reduction to zero who merits a specific explanation. Our first thought
was that the accurate response was $50 and that we had made a coding or key-
ing error. However, when we reviewed the questionnaire, we found that the
revision was indeed to $500. The respondent was a creative writer making
$22,500 annually. Thus, this response likely will be reflected when our analy-
sis of outliers is performed on the certainty questions. Since the treatment
of outliers was not resolved as part of Phase I's research activities, we have
left this response in the mean values that are presented later in this chapter.
11-28
-------
TABLE 11-2. PRO* lit Of RE^PONDEN1 3 WiTH REVISED BrDS
Willingness-to-pay
quesnon
initial
bid
Revised
bid
Question-
naire Version
format number
Respondent
revised
bid--more
than one
question
House •
hold
income
woa)
Age
Education
(years)
Sex
Hom«-
ownership
Years
lived
in
town
Number
or
children
under
18
Rated
har m or
hazardous
waste .
pollution
Rated
ef fectiveness*
of .
guvfcrn-
roeM in
dealing with
hazardous
wastew
©oitied
water
is a
current
activity
Public
reeling*
are a
current
activity
Risk decrease front
a-ard A so card 8d
0
25
till ei 1
b
No
47.5
45
18
f em«s!e
Own
7
0
10
1
Yes
Ha
1
5
Direct
8
No
NA
38
18
Female
Own
11
1
2
3
Yes
No-
5
2
Direct
1
No
17 S
40
16
J-eniil*
Own
6
0
2
3
No
Na
5
2
Duet 1
6
No
32.5
27
14
Male
Rent
2
0
4
3
No
No
5
10
Direct
7
NO
ST. 5
42
16
Mala
Own
12
1
9
%
Yes
No
10
4
Direct
3
No
82. S
40
18
MA
Own
1
1
NA
3
No
No
IS
25
Direct
7
No
62.5
38
16
Male
Own
7
2
4
4
No
Yes
20
10
Direct
7
No
02.S
38
14
Female
Own
10
0
5
8
No
No
20
IS
Direct
6
Yes
27.5
25
14
Female
Rent
s
1
10
5
No
No
30
20
Direct
S
No
37.5
44
4
Mate
Rent
2
1
NA
7
Yes
No
100
SO
Direct
5
No
47.5
55
12
Female
Own
20
0
10
a
Yes
Yes
Risk
decrease from
'
card
8 to card Cd
10
S
Direct
6
Yets
27.5
25
14
Female
Rent
£
1
10
5
No
No
10
5
Direct
7
No
52. S
57
18
Male
Rent
19
0
8
i
No
No
50
150
Direct
J
No
52.5
32
16
Female
Own
8
1
10
i
No
Mo
Risk increase froin
card X to card Y
10
10
20
30
£5
120
25
20
Ranking
Direct
Rrfnkmg
Ranking
No
No
No
No
27.5
7.5 -
17 5
27.S
m
33
63
31
12
12
14
16
Own
Rem
ftent
Rent
10
10
10
NA
Yes
Yes
No
No
NO
HO
No
No
Risk decrease from
card A to *ero risk*5
20
5
Ranking
1
No
12.5
64
12
M4|#
Own
25
0
5
10
No
No
40
20
HAJIklllQ
2
No
NA
35
12
Female
Rent
6
0
S
4
Yes
No
50
S3
Hanking
1
No
67.5
35
18
Male
Own
6
0
1
2
No
No
80
500
Ranking
4
No
22.5
53
18
Male
Own
14
0
8
1
Ho
NO
100
25
Ranking
4
Yes
17.5
29
12
Male
Rent
2
2
10
4
Yes
No
100
75
banking
2
Yes
17.5
46
18
Female
Rent
12
0
10
3
Yes
Yes
150
100
Hankmg
3
NO
27.5
70
14
fsmale
Own
70
0
5
10
No
No
Risk decrease to
wildlife^
10
20
Direct
2
No
27.5
46
18
Male
Own
fi
0
¦0
4
Yes
Yes
20
30
Rjuikino
]
No
72.5
67
16
Male
Own
25
0
a
2
No
No
25
18
Direct
4
No
57.5
46
14
Female
Own
10
0
9
&
Yes
Yes.
SO
25
Ranking
4
Yes
17.5
29
12
Male
Rent
2
2
10
4
Yes
No
50
75
RenkiriQ
2
Yes
17,5
46
10
Female
Rent
u
0
10
3
Yes
Ye&
T h ree respondent had revised I heir bids lor two questions] hence, these respondent b am counted twice within Lhis table.
Represents a scale card response from 1 to 10 with I = not harmful and 10 = very harmful.
Represents a scale card response from 1 to 10 with 1 not at alt effective and 10 = very effective,
a
Applicable only to the direct-question formal.
applicable only to the ranking-question format.
includes onjy non*ero bidders for a reduction in household risk; consequently( the posited level or household risk is equal lo the level purchased m
previous questions.
-------
that influence the likelihood of someone being a protest bidder. It also pro-
vides a brief comparison of our results with those of Mitchell and Carson
[1984], It is important to consider the reasons underlying protest bids be-
cause this process helps to assess whether or not they are influenced by the
nature of our contingent commodity. Finally, we evaluate whether or not any
of the features of our research design--e.g., property rights assignment and
the low probability vectors for the exposure risks — had any effect on the like-
lihood that someone would reject the terms of the contingent market.
Examining the role and influence of protest bidders has taken on increas-
ing importance in contingent valuation research over the last few years, Ran-
dall, Hoehn, and Tolley [1981] stressed that the elimination of protest bidders,
as well as outliers, enables one to obtain a solid core of data from a contingent
valuation survey. A number of studies, including Desvousges, Smith, McGiv-
ney [1983] and Mitchell and Carson [1984], have argued for detailed examina-
tion of the characteristics of protest bidders as one mechanism for evaluating
the framing properties of a contingent valuation instrument.
An important consideration for protest bids is the procedure used to
classify them. In our case, the protest bids were determined using ex post
classification of the reasons that people gave for zero bidding. If people gave
any reasons other than that's what it's worth to them, or they cannot afford
anything, they were considered a protest zero. In addition, we classified non-
respondents to the valuation question as protest bidders. Table 11-3 shows
the frequency of reasons for zero bids for questions eliciting option prices
for risk decreases. The results in Table 11-3 are quite interesting. Only 15
percent of the sample were protest bidders for the questions associated with
the valuations for risk reductions. Within this group there are several inter-
esting subsets. For example, 11 of the 55 total protest bidders, or 20 per-
cent, felt that companies or government should bear the cost of controlling
hazardous wastes. This would imply that the reason most frequently given
for protesting the terms of the market was related to our payment vehicle--
higher taxes and product prices. People did not accept our explanation that
when companies or government do pay the cost, they ultimately pay part of
the share. Alternatively, they could simply feel that they have an implicit
property right to reduced exposure to hazardous wastes. That is, they should
11-30
-------
TASL6 11-3. FREQUENCY OF REASONS FOR ZEF-'Q BIDS BY LEVEL OF RISK, DECREASE
DIRECT QUESTION FORMAT
Decrease in risk of exposure per conditional risk
Reasons for zero bids
ViJgJLZZL-
1/10 1/20
1/10 to 1/50
1/10 1/20
1/30 to 1/150
1/10 1/20
1/300 to 1/1,500
1/100 1/200
Total for
all versions
Protest bids
Not enoygh information
Did riot want to piece •
, doner value on chang®
Objected to the presentation
: of the question
Multiple riiponx'
Costs should be born# by
b
, companies or government
1 ODjBCted to more tlxil'1
' Objected to the existing b
t of Btwtrnment
3 1
C
of payments
Further coot
imposed with no costs"
Obiected to the distribution
o
Further controls could be
.6'
Nonresponse
Other6
Total protest bids
Nonprotest bids
That Is what it is worth
Cannot afford anything
Other6
0
1
s
0
11
0
of sample size
1
0
11
0
3
0
e
•!i
3
3
55
0
41
5
TOM non prat est zero bids
2
4
?
5
7
7
11
3
48
Total of all zero bids
6
10
13
11
13
18
16
14
101
Total sample sizes
42
47
47
46
49
45
S3
42
371
Protest bids as a percent
.14
.13
,13
,13
.12
.20
.12
,26
, 15
'Respondent stated a reason chat was a combination of the reasons formatted within me survey-
expense was not formatted within the survey
11-31
-------
sample size appears sufficient to support further analysis of certainty! as a
risk endpoint.
11.9 RISK OUTCOMES
This section describes our design for, and preliminary analysis of, plter-
native risk outcomes on the mean option price bids for reducing the risk of
exposure to hazardous wastes. Recall our framing of the initial commodity did
not describe a particular risk outcome. Instead, it asked each respondent
what cause of death he envisioned. However, the design subsequently called
for changing the framing of the commodity to elicit responses for two alterna-
tive outcomes: death caused by damage to the body's immune system and a
risk of severe lifetime birth defects.
The interviewer asked the respondent if he would like to change hi]s bid
for the risk change (decrease) for each case. This sequence was introduced
by the statement:
Think about this [hypothetical] situation. Most experts agree that
exposure to hazardous wastes may cause different kinds of health
problems. . . You might decide that you would be willing to pay
something different [from the total for risk reductions that the inter-
viewer had just mentioned] if you thought about different kinds of
health problems.
Table 11-16 provides summary statistics for the bidders who revised their
responses for the immune system damage and birth defects as risk outcomes.
These results and subsequent evaluation of a profile of the respondents are
interesting. For example, only two respondents out of a total of 172 respond-
ents who changed their bids lowered - their responses. In one case, the
respondent lowered his bid from $175 a month to $50 a month for each of the
alternative outcomes. However, the reliability of this respondent is somewhat
suspect given that his initial bid is high relative to his income. General!y,
the mean revisions are quite sizable, ranging from $12 to $30 per month for
immune system damage and from $10 to $20 per month for birth defects, A
sizable number of these means are significantly different from zero. This is
especially true for birth defects where the mean for only one design point is
not significantly different from zero at the 0.01 level of significance.
In addition, the means for birth defects appear much less skewed than
those for immune system damage. Indeed, there are a sizable number of large
11-62
-------
TABLE 11-18. SUMMARY STATISTICS Of THE CHANGE IN OPTION PHfCI BIDS GIVEN A SPECIFIC ILLNESS,
PROTEST BIDS EXCLUDED, OUTLIERS INCLUDED
Chinotd bid*
Chang* In option price to avoid immune >ystew daawge Chmm »n option pfic» to «voM birth defects
Condi I tonal
risk
Exposure
risk
change*
Mean
Median
Standard
deviation
Mumber
of
obser-
vations
Minimum
value
Maximum
value
t" b
statistic
Mean
Median
Standard
deviation
Number
of
obser-
vations
Minimum
value
Maximum
value
»- K
statistic
1/10
1/5 to 1/25
12.00
10 00
7.53
10
5,00
25.00
5,04"
16,0?
10.00
14,®?
14
2.00
50.00
4.27»*
1 m
l/S to 1/25
40.09
5,00
113.89
11
-125.00
306.00
1.1?
20 33
64.38
10.00
15
-125,90
200,00
1.22
1/10
1/10 to 1/50
28.6?
10.00
40,SO
9
1.00
100.00
2.10*
15.32
7.50
25.15
14
0.5®
100.00
2 22**
1/20
1/10 to 1/50
30.25
•/.SO
52,46
6
o.so
135.00
1,41
11,23
10.00
14,03
11
O.SO
SO.00
2.85**
1/10
1/30 to 1/50
15.64
10.00
17,51
11
2,00
50.00
2.96**
18.69
10,00
27,01
19
-15,00
100.00
3 01**
1/20
1/30 to 1/50
19.00
10.00
18 17
5
5 00
50.00
2.34*
15,30
10.00
13.78
11
2,00
SC. 00
3,68**
1/100
1/300 to
1/1,506
15.23
S.MJ
24.87
10
0.25
80.00
1.96*
10 66
S 00
13.SI
14
1.00
SO.00
2.87"
1/200
1/300 to
1/1,500
14.30
10.OS
IS 48
5
O.SO
. 40.00
2.06
11.15
IS.00
7 91
?
0.25
20.00
3.S3**
All versions
combined
2z.es
10.00
51.46
6?
-125.00
300.00
3.54"
15.48
10.00
29.4?
105
-125.00
200 00
5,37**
"•'Significant at the 0,01 percent level using a one-tsil test.
•Significant at the 0,05 percent (eve) using a one-tail test.
aAll risk changes represent a mouwitnt (r» card A to card C.
bfor (he null hypothesis that the population mtin Is «ro.
-------
revisions, with the largest being $300/month for immune system damage. This
occurred when a 66-year-old male with an income of $27,500 doubled his month-
ly bid. A 55-year-old female with household income of $52,500 also changed
her bid substantially for the immune system outcome when she increased it
from $8/month to $200/month. (However, she did not change for the birth
defects.) The largest change for birth defects occurred when a 59-year-old
male with an income of $42,500 doubled his monthly bid of $200. Thus,, it
appears that the outcome of death by damage to the immune system affected
somewhat fewer respondents (67 vs. 102) than the birth defects outcome, but
some of the responses were quite large. Clearly, further analysts of these
responses is warranted in order to judge their plausibility.
Table 11-17 provides some additional information on the influence of the
specific risk outcomes on the option price bids. It compares the summary sta-
tistics of respondents who changed their bid in response to either or both out-
comes with those respondents who did not change. In general, the mean bids
for the "changers" exceeds those for the nonchangers. This is true for both
birth defects and the immune system effects. While this is not surprising
given their willingness to change, the substantial size of some of the mean
bids—over $80/month in one case--was somewhat surprising. It should be
noted that these results relate to the sum of the option price bids for risk
reductions from A to B and B to C.
In summary, the effects of risk outcomes offer one clear area for addi-
tional research. The description of specific outcomes generally resulted in
about one-third of the respondents altering their bids. The research issue
that must be considered is to develop a framework that provides a better
understanding of the influence of individuals1 characteristics on this process.
The size and number of changes should provide sufficient sample for these
additional analyses.
11.10 INTRINSIC VALUES
This section presents our results on intrinsic or existence values. It
describes the framing of the commodity--risk reductions for critters--and high-
lights the summary statistics for the option price bids. Finally, it considers
the implications of differences in the initial levels of risks posed to individuals
11 -64
-------
TABLE 11-17 SlIMMARV STATISTICS OF OPTION PR I CI B»DS FOB A RISK DEC R£A5b GIVLN A SPECIFIC ILLNESS,
PROTEST BIDS EXCLUDED, OUTLIERS INCLUDED
Nonprotest bids
Respondents who charmed their bids Respondents who maintained their original bid
Condi- Queslion-
Stated illness
tional
risk
Exposure
risk change
naire
version
Mean
Median
Standard
deviation
Number of
observations
t- h
statistic
Mean
Median
Standard
deviation
Number of
observations
1- h
statistic
Immune system
1/10
1/5 to 1/25
3
37.50
30.00
21.51
10
5.51**
23.65
12.30
29.87
26
4. 04 **
damage
1/20
1/5 to 1/2S
1
79.27
20.00
136.24
11
1.93*
42.63
14.00
59.05
30
3.36**
1/10
1/10 to 1/50
1
61.44
35.00
67.28
9
2. 74*
24 56
10.00
33.89
32
4.10**
1/20
1/10 to 1/50
2
66.42
27.50
82.17
6
1.98
55.85
27.50
83,81
34
3 89**
1/10
1/30 to 1/150
5
72.36
35.00
68.23
11
3.52**
22.46
12.50
. 26.23
32
4.85**
.
1/20
1/30 to 1/150
6
35.00
30.00
21.79
S
3.59**
43.35
15.00
92.95
31
2.60**
1/100
1/300 to 1/1.500
7
46.73
20.00
51.22
10
2.88**
33.13
15.00
53.05
38
3.85**
1/200
1/300 to 1/1,500
8
34. SO
20.00
37.88
s
2.06
27.42
15.00
27.59
26
5 07**
Brith defects
1/10
1/5 to 1/25
3
37.28
27.50
29.11
14
4.79**
26.04
15.00
31.87
22
3.83**
1/20
1/5 to 1/25
4
84.87
22.00
116.04
IS
2 83**
28.54
11.50
39.35
26
3 70**
1/10
1/10 to 1/50
1
42.11
35.00
49.18
14
3 20**
26.15
10.00
36,44
27
3. 72**
1/20
1/10 to 1/150
2
44.77
40.00
39.48
11
3 76**
60.24
25.00
90.11
29
3.60**
1/10
1/30 to 1/150
5
54.32
30.00
57.10
19
4.15**
27.75
10.00
37.88
24
3.59**
1/20
1/30 10 1/150
6
35,85
30.00
22.37
11
5.32**
47.92
15.00
102.85
25
2.33*
1/100
1/300 to 1/1,500
7
38.68
17.50
46,98
14
3 10**
34.60
IS. 00
54.46
34
3 71**
1/200
1/300 to 1/1,500
8
30.75
30.00
17.52
7
4.64**
28.46
15.00
29.23
24
4.77**
**Significant at the 0.0} level.
•Significant at the 0,05 fevel.
3AII risk changes represent a movement from card A to card C.
bFor the null hypothesis that the population mean is zero.
-------
for their valuations and described to be relevant to other elements in the eco-
system (i.e., critters) for the option price bids.
As noted both in Chapter 7 and in the discussion of the conceptual link-
ages in Section 11.4, the intrinsic value question was also framed in terms of
state-independent option price bids for risk reductions to be experienced only
by critters. The bids are state independent because they are ex ante amounts
that are made without prior knowledge of the eventual outcome of the risk
reduction. The interpretation of the bid as an option price requires that the
individual does not have any other avenues for adjusting the risks to critters.
For example, he does not participate in a community hazardous waste col lection
near some ecosystem, which could reduce the risk of exposure to hazardous
wastes for the critters.
The text to describe the framing of risk reductions for critters is shown
below.
Now suppose that the risk of exposure to you (and your household
, members) has been reduced to the level on Card .*
Suppose that the government adds regulations on this landfill,
These additional regulations would not lower your (or your househo o
members') risk, but would lower the risk of exposure to hazardous
waste for fish, wildlife, and plants only. Their combined risks
would be lowered to the levels they face in nature. Suppose also
that none of them is in danger of becoming extinct.
In addition to the (READ TOTAL OF F.6.a + F.6.b OR AMOUNT
FROM F.6.c ON REMINDER SHEET) you have said you would be will-
ing to pay, how much more in higher product prices and taxes per
month would you be willing to pay for these regulations that woulc;
reduce risks of exposure for fish, wildlife, and plants only?
The framing of this risk reduction affects the interpretation of the empirical
results for the risk reductions. For example, the outcome at risk differs from
all the previous outcomes for risk changes: it is only for critters and does
not affect the household's risk in any way. Not only does the outcome differ,
the endpoint for the risk is less specific than in the earlier risk changes.
The endpoint is to levels the critters face in nature.
*Blank line refers to the level of risk that each respondent had purchased
for their household. The interviewer supplied this value.
11-66
-------
In addition, the framing also affects the attributes of the population (i.e.,
fish, wildlife, and plants) experiencing the risks. It states that none of the
members of the ecosystem are in danger of becoming extent. Our intent here
was to suggest that the population did not include snail darters, or Indiana
bats, or some other creatures on the endangered species lists. Clearly, a
more comprehensive design would have varied this attribute to see if it affected
the option price bids. Nonetheless, it is important to recognize that our con-
ceptual analysis for these values has not provided a specific description of
the potential importance of the attributes of the creatures at risk.
Instead, the design considers only differences in the initial levels of risk
for option price bids for intrinsic values. This design feature followed logical-
ly from our treatment of intrinsic values as an increment to the user values.
Because respondents differed in the amount of household reductions they pur-
chased—a. g, f Level A (zero bidders), Level B, Level C, or zero--the initial
levels of exposure risk for creatures also varied. The framing reflects this
feature by requiring the interviewer to remind the person of the endpoint for
the household risk.
The summary statistics, shown in Table 11-18, provide some insights as
to the importance of our research design for Intrinsic values. The monthly
option price bids are statistically significant from zero for three of the five
initial levels of risk. The two values that are not different are those for zero
bidders. Nevertheless, it seems premature to conclude that providing these
respondents an opportunity to relfect intrinsic values was not useful. The
summary statistics include responses for all bidders, including protest bidders
and potential outlying bids. A final assessment of our attempt to include zero
bidders for household risks in the design for intrinsic values will require more
analysis.
The results in Table 11-18 also suggest that the starting level for the
risk reduction affected the mean values of option prices for risk reductions to
fish, wildlife, and plants. In particular, the bidders who purchased a zero
risk reduction had larger, and statistically different, mean option price bids
than any of the means in the first three rows of the table. This is surprising
given the vague specification of the initial level of risk in this case.
11 -67
-------
TABLE 11-18. SUMMARY STATISTICS OF OPTION PRICE BIDS FO
INTRINSIC VALUES (RISK REDUCTIONS TO CRITTERS), ALL BI DDE
RS
Initial level of risk
Version
Mean
Standarc^
deviation
N
t- statistic
Zero bidders
Bid for A-*B only
Bid for A->B and B-*-C
Bid to zeroC
Zero bidders
D
D
D
R
R
7.55
2.77
6.72
13.6
5.77
11.99
6.02
13.10
17.91
28.36
11
53
196
187
49
2.OS
3.35**
7.1$**
3.85**
1.46
•"~Significant at the 0.01 level using a two-tail test
a
Standard deviation
-u
l(Xi-X)2 where X is the sample mean and Xi
is the
observation for individual i and N is the sample size.
3For the null hypothesis that the population mean is zero.
"In this case the initial level of risk was vaguely defined. It was described
as positive, greater than their natural state but not specified. This outcome
resulted from the effects of our design and the sequence of responses to it
that individuals could make with each design point.
11-68
-------
Differences in the information provided to respondents in the two ques-
tionnaire versions could account for the significant different option price means
for intrinsic values. That is, the direct question version notified respondents
in advance that they would be asked to bid for two distinct risk decreases
for their household and then an additional amount for reducing the risk of
exposure to hazardous wastes for fish, wildlife, and plants. In effect, these
respondents might have been able to mentally allocate their respective total
valuations between risk reductions for their household and "those for critters,
resulting in lower amounts for critters than in the ranking version where there
was no prior notification. This could imply a variation on Tversky and
Kahneman's [1981] "mental accounts" concept is operating.
Other possible explanations are possible for these results. For example,
a simultaneous equation model of the decision process for valuation responses
for the respondents in the direct question part of the design who were pre-
notified of the bids to be requested may help to explain their behavior. In-
deed, both Smith [1384] and Hanemann [1385] have argued for the need of
such models in analyzing contingent valuation responses.
An alternative explanation may lie in a basic assumption of any analysis
of mean values; differences in the characteristics among individuals are not
important. Relaxing this assumption is a high priority for future research
activities, especially given the sensitivity of the "use values" for risk reduc-
tions that is discussed in Chapter 13. While many of our explanations are
very speculative at this juncture in the research, the quality and diversity of
information on intrinsic values merits more intensive investigation than is now
possible,
11,11 IMPLICATIONS
Given the objectives of this chapter, a summary of our results seems in-
appropriate. Essentially, its purpose is to initiate the empirical analysis of
the option price bids for changes in the risk of exposure to hazardous wastes,
in achieving its purpose, the chapter has stressed the importance of the fram-
ing of the contingent commodity for interpretations of the contingent valuation
results,
Overall, the results described in this chapter indicate that further re-
search is clearly warranted. Our examination of protest bidders revealed an
11-69
-------
overall rate of protest bids that is most encouraging. These bids accounted
for only about 15 percent of our valuation responses. Had respondents been
completely unable to deal with our interpretation of hazardous waste as a risk,
we would have had a much higher rate of protest bids in our sample. = While
the findings on protest bidders do not necessarily provide evidence on the
quality of the valuation responses in the nonprotest component of the sample,
they do seem to indicate that individuals did not reject the framing : f our
commodity as one involved with a risk change. Whether oi~ not they experi-
enced difficulty in processing the framing information is a crucial objective of
any subsequent empirical analyses.
While the option prices for risk changes do not appear to be consistent
with our a priori expectation that risk changes from a higher initial level would
be valued more highly, they are not implausible, especially if one accepts the
view that individuals may perceive that state-dependent adjustments are feasi-
ble. Nearly all mean bids are significantly different from zero. Clearly, addi-
tional analysis of the outlying responses and the differences among individuals
that may affect the mean bids should help to clarify some of the relationships
between the values for risk changes and the initial levels of risk.
Our preliminary investigations into the option prices motivated by intrinsic
values are also encouraging. The relative sizes of the means compared to the
use value means suggest that respondents understood the incremental nature
of oUr design. The preliminary nature of our research in this area precludes
further general conclusions. The plausibility of the responses seems to sug-
gest that efforts to model the nature of individuals' responses to these ques-
tions may be beneficial.
Additionally, our preliminary results on the effect of certainty as a risk
endpoint offer encouragement that the responses to this design question merit
further attention. This implication also appears to hold for the questions that
elicited changes in option price bids when specific risk outcomes, death from
immune system damage and birth defects, are posed.
Thus, the main implication to be drawn from these results is that consid-
erably more research is required to discern the patterns and processes that
underlie these responses. Yet, the data seem capable of fulfilling at least
some of the requests from both economic and psychological analyses,
11 -70
-------
CHAPTER 12
OPTION PRICE RESULTS: PRELIMINARY REGRESSION
ANALYSES USING UNRESTRICTED MODELS
12.1 INTRODUCTION
This chapter presents our statistical analysis of how differences in char-
acteristics among individuals may affect their values for reductions in the risk
of exposure to hazardous wastes. The basic assumption underlying this analy-
sis—that these differences can influence responses—is an extension of the
analysis of means presented in Chapter 11, which assumes that the only
sources of differences in valuation responses are related to the specified fea-
tures of the risk changes. That is, the analysis in Chapter 11 assumes that
the level of the exposure risk, conditional risk, and size of the risk change
are the only potential sources of differences in the estimates. Unfortunately,
however, the results of our examination using the simple model are largely
uninformative. Consequently, detailed interpretations of these findings are
not presented. Instead, the chapter focuses on summarizing our attempts to
develop measures of several characteristics that our conceptual analysis, other
literature, and the focus groups suggested would be important to understand-
ing the valuation responses. The two main characteristics which organized
this empirical work are the household's available avenues for adjustment and
its health status. Following a discussion of the information available on these
issues, the chapter presents some illustrative regression results based on our
use of the simple model.
12.2 GUIDE TO THE CHAPTER
Section 12,3 of this chapter presents the simple model that provides an
organizational structure for the chapter. Section 12.4 discusses the role of a
household's avenues of adjustment in our analysis and describes several vari-
ables that are used to represent these avenues. Section 12.5 considers the
influences of a household's health status on the valuation responses and details
12-1
-------
our various alternative specifications of measures of health status. Section 12.8
presents our illustrative regression results for risk increases and decreases,
Section 12.7 summarizes some implications that can be drawn from this chapter,
12.3 SIMPLE MODELS
This section develops the underlying rationale for a simple model to ana-
lyze individuals' valuations of changes in hazardous waste risks. The model
is primarily a heuristic device to reflect several major points from our concep-
tual analysts (see Part I) and some of the elements of our research design (see
Chapter 7). This basic structure is then varied to attempt to reflect factors
other than the risk change for the valuation responses. Nonetheless, all of
the models considered are simplified in three respects: they are assumed to
be linear in variables and parameters, they are used to examine values both
for risk decreases and for avoiding risk increases, and they pertain only to
use values.
12.3.1 The Model's Rationale
The object for starting with a simple model is to guide the process of
examining how differences in particular characteristics across individuals can
influence their valuation responses. There are several reasons to test for the
influence of these differences in respondents' characteristics for their respec-
tive valuation responses. First, our conceptual analysis clearly indicates that
on economic grounds individuals should differ in how they value changes in
risk, This same conclusion can also be inferred from psychologists who sug-
gest that either differences in perceptions or differences in the ability to proc-
ess the information presented in our questionnaire should lead to variations in
the valuation responses. Equally importantly, our experiences in the focus
group sessions (see Chapter 8) suggested that variations in individual's valua-
tions could frequently be linked to certain attitudes or perceptions. For exam-
ple, individuals who expressed concern over the effectiveness of the govern-
ment in delivering the risk change frequently gave low, or zero, values for
reductions in hazardous waste risks. Additionally, the presence of children
at hcme--and especially younger children — seemed to have a positive effect on
valuation responses. Finally, participants who perceived that their genetic
make-up or overall health status made them more susceptible to experiencing
12-2
-------
ie health effects if exposed to hazardous wastes frequently expressed higher
aluation responses.
In summary, our conceptual analysis, the findings of psychologists related
3 risk perception, and our focus group experiences point the way toward a
lodel that can account for differences among individuals" attributes and per-
eptions in the formulation of their valuation responses,
2.3.2 The Model
The basic model used to begin our evaluation of the option price results
s described in Equation (12,1);
AE = f(R, q, Z) , (12.1)
here
AE = the contingent valuation response, which, based on the form
of the contingent valuation question, is a constant, state-
independent payment (i.e., an option price).
R = postulated initial level of the individual's (or a family mem-
ber's) probability of being exposed to hazardous wastes suf-
ficient to imply a second-stage risk, q, of death.
Z = a vector of measures of the individual's socioeconomic char-
acteristics, measures of attitudes toward risk, effectiveness
of government, and information on the subject.
his basic model is a first step in reflecting the implications of our conceptual
lalysis. It suggests that the individual's valuation, which is interpreted as
i option price, wilt be affected by the initial level of risk posed to the re-
pendent in the framing of the commodity.
Generally, our conceptual analysis implies that there is a positive relation-
lip between the option price bid and the initial level of risk including both
ie risk of exposure and the conditional risk of an effect given an exposure.
hile the analysis of means presented in Chapter 11 finds that the conditional
sk has a strong effect on the option price amounts, the direction of the
'feet is the opposite of our a priori expectations. Our basic model, and the
¦gression analyses that will be used with it, may help to evaluate whether
lis effect stands up when the framework controls for differences in character-
tics among individuals.
12-3
-------
M
CP
*Not covered 111 survay quMtionrutir*.
Figure 12-1. Potential avenues for adjusting to risk exposure.
-------
fied in our conceptual analysis, but we were unsure how to elicit this informa-
tion in the survey.
Instead, the survey requested information on activities individuals were
currently undertaking (or had undertaken in the past). Some of these actions
were associated with perceived risks of exposure to hazardous wastes and some
with the types of risk. Consequently, this chapter assumes that these re-
sponses provide an indirect indication of each respondent's potential for seek-
ing to adjust to the circumstances described in our contingent valuation ques-
tions, Clearly, the specific activities undertaken in the past and the other
variables used to measure this potential could not have been in response to
our contingent valuation questions.
Another limitation of our attempt to reflect these effects on individuals
valuation responses stems from our understanding of what each individual's
perceived avenues for adjustment might be. It is somewhat vague at best.
Nevertheless, comments from our focus group participants can be interpreted
as giving an approximate idea of how at least a few of these avenues mignt
work. For example, participants frequently mentioned that they purchased
bottled water as a way of avoiding exposure to hazardous wastes and other
possibly harmful materials. Others said that they attended public meetings
and workshops and sought out other information sources to better understand
hazardous wastes and the ways of limiting their exposure. In particular, the
Acton, Massachusetts, residents who participated in our focus groups stressed
the relevance of both bottled water and better information as ways of avoiding
risks. (See Desvousges et al. [1984b] for details on the Acton sessions.
Also, see Chapter 10 for summary statistics on bottled water and information
acquisition,) The regression analysis uses a qualitative variable to reflect
the presence or absence of these two adjustment avenues for this household.
Occupation selection and residential focation are also avenues for adjusting
to several types of risk including the possibility of being exposed to hazardous
wastes. For example, white collar workers are likely to have very small risks
of being exposed on their jobs, while the chances of exposure are probably
higher for certain types of blue collar workers. Thus, individuals could re-
duce their exposure risk by their occupation choice.
Additionally, it is possible that workers in technical occupations may have
a better understanding of, and access to, information about risks. This may
12-6
-------
be another way that occupation influences the perceived avenues for adjusting
to risks. In our empirical analysis we devHaped several occupational classifi-
cations to try to account for these potential information differences. The ra-
tionale for this approach follows our earlier argument. These occupational and
the diflerentials in risk and information on risk may have required some indi-
v duals to consider adjustments to risk Therefore, they have provided exper-
ience and familiarity with the process This could as a result influence how
they responded to the risk changes posed in our contingent valuation ques-
tions. These were tried in the model both as an additive term and as an
interaction with the conditional risk. Unfortunately, neither was significant
in any of the regression models.
As noted in Chapter 1 and in the discussion of the property value model
in Chapter 15, changing the location of one's dwelling could alter the risk of
exposure to hazardous in several ways. For example, a move could change
the level of air quality and the source and subsequent quality of the drinking
water. It also could change the flow of information available to the household
should it move into a town in which the town council or newspaper provides
iniormation about hazardous wastes. As noted in Chapter 10, anyone moving
to the town of Acton, Massachusetts, after it experienced a series of hazardous
waste contamination incidents would probably have experienced a considerable
increase in the flow of information on hazardous wastes.
As a very crude attempt to account for the influence of the residential
location as an avenue for information on risk and familiarity with adjusting to
it we have included qualitative variables for several of the towns in our sam-
ple. The approximate nature of these qualitative variables is attributable to
the possibility that they could also reflect some other town characteristic or
household characteristic related to the town that are omitted from our model.
Nevertheless, improved modelling of this avenue may yield some cub tantial
payoffs in future research because of the pervasiveness of residential location
in the household's risk of exposure to hazardous wastes.
12.5 HEALTH STATUS
This section describes the potential effect of health status on individuals'
valuations, 11 also provides summary information on the health status of our
respondents. 11 concludes by discussing the analysis variables that were con-
12-7
-------
strutted to measure the effect on differences in health status on individuals'
valuations.
12.5.1 The Role of Health
A household's value for reductions in hazardous waste risks is likely to
be influenced by its health status. Differences in value could be attributed
to perceptual or economic factors, or both. For example, a household that
has experienced the consequences of a disabling disease may place a very dif-
ferent value on risk reductions than one who has not. However, the implica-
tions of differences in health status are not clear on a priori basts. A house-
hold with lower health status may perceive itself more predisposed to experi-
encing the health consequences of hazardous wastes and therefore have a high-
er value for reducing these risks. Conversely, it may be willing to pay less
for a risk reduction because it has already contracted a major disease, and
any effects from hazardous wastes are viewed as of secondary importance.
Or its poor health may have resulted in lower earnings for the household and,
therefore, reduced its ability to pay for reducing exposure to hazardous waste
risks,
A behavioral model that includes a household's health status would seem
a logical way to improve our understanding of the effects of this characteristic
on valuation responses. Unfortunately, such a model has not been developed
for this phase of the research. Instead, we have tried several ad hoc specifi-
cations that include health status in our simple model. Nevertheless, our sur-
vey questionnaire does provide a substantial amount of information on the health
status of our sample individuals. Figure 12-2 depicts the main health-related
questions that were included in the questionnaire.
Both self-assessed health status measures and objective health indicators
were elicited in the questionnaire. The self assessment included the respond-
ent's rating of his health and a comparison of his health with others of the
same age. Table 12-1 shows the respondent's ratings for these two perceived
health indicators. Generally, our respondents considered themselves to be in
good health. Only 12.7 percent of the respondents rated their health as fair
or poor, while less than 6 percent considered their health worse or much worse
than average for someone their age.
12-8
-------
Figure 12-2. Overview of health Information.
-------
TABLE 12-1. SELF-ASSESSED HEALTH STATUS
Own health compared
Own health to others of same age
Rating
Number of
respondents
Percent
of total
sample
Rating
Number of
respondents
Percent
of total
sample
Excellent
278
45.8
Much better
97
16.1
Good
252
41 .5
Better
'211
35.0
Fair
60
9.9
Same
261
43.3
Poor
17
2.8
Worse
31
5.1
Much worse
3
0.5
Total
607
100.0
Total
603
100.0
TABLE 12-2. QUANTITATIVE INDICATORS OF HEALTH STATUS
Workdays missed in last 2 weeks
Percent
of total
sample
Number Number of
of days respondents
0
1-2
3-4
5-7
8-13
14
Total
560
25
5
7
3
12
612
91 .5
4.1
0.8
1.1
0.5
2.0
100.0
Overnight stays in hospital
Number
of nights
0
1-2
3-4
5-7
8-14
15-30
31-70
Number of
respondents
538
16
15
15
16
12
612
87!. 8
21.6
as
2:,5
2j.S
2U
100
12-10
-------
The questionnaire included the objective or quantitative indicators of
!
health status shown in the right side of Figure 12-2 to supplement the per-
ceived health information. These quantitative measures could be used either to
verify the self-assessed health status or simply as alternative measures. We
have used them to try to provide a set of more discriminating distinctions in
health status. Nonetheless, future investigations may well consider the corre-
spondence between perceived and objective measures of health and its implica-
tion for the valuation of risk reductions using the two types of health infor-
mation .
Table 12-2 provide summary information on our respondent's health status
using the quantitative measures. Again, the respondent's appear to be in rea-
sonably good health based on workdays missed and overnight hospital stays.
Less than 10 percent of the sample had missed a day of work in the 2 weeks
prior to their interview, while only 12.2 percent had spent any time in the
hospital during the last year. Table 12-3 provides additional information on
the incidence of six common diseases or ailments — heart disease, hypertension,
diabetes, kidney trouble, cancer or leukemia, and the effects of a stroke.*
12.5.2 Health Analysis Variables
To account for differences in health status among individuals we con-
structed a variety of proxy variables. As noted earlier, these are largely
ad hoc measures. The majority of the variables were qualitative, or dummy
variables. These variables were used in regression models either to test for
intercept changes or as interactions with the conditional risk variable to reflect
possible influences on individuals' perceptions of the events at risk. These
health variables! are as follows:
A qualitative variable equal to 1 for status categories excellent
and good and equal to 0 if respondent rated health either fair
or poor.
A qualitative variable equal to 1 if respondent rated health at
least average in his age and equal to 0 if rated worse than
average for his age.
~Another possible issue for further research is to compare these quanti-
tative measures with comparable measures from other health surveys to better
appriase the health of our respondents.
t A11 listed variables were used both as intercept shifters or as interactions
with conditional risk.
12-11
-------
TABLE 12-3. INCIDENCE OF SIX DISEASES
AMONG SAMPLE RESPONDENTS
I ncidenee
Diseases Yes No Total
Heart disease
Number 47 565 612
Percent 7.7 92.3 100.0
Effects of stroke
Number 4 608 612
Percent 99.3 0.7 100.0
Hypertension
Number 98 514 612
Percent 16.0 84.0 100.0
Diabetes
Number 15 597 612
Percent 2.5 97.5 100.0;
Kidney trouble
Number 28 584 612
Percent 4.6 95.4 100.0
Cancer/Leukemia
Number 17 595 612
Percent 2.8 97.2 100.0
12-12
-------
A set of qualitative variables interacted with the assessed health
categories,
A qualitative variable equal to the respondent had spent ? or
more days in the hospital during the last year.
A set of qualitative varibbies interacted with days spent in
hospital.
:
A qualitative variable equal to 1 if the individual was presently
, a smoker and equal to 0 if nonsmoker .*
j ¦ Qualitative variable equal ru of respondent or family member
had experienced cancer or leukemia and equal to 0 if no inci-
dence of cancer.
Qualitative variable equal to 1 if respondent had indicated inci-
dence of disease among family members and equal to 0 other-
wise. (Measures were constructed for each disease and across
diseases.)
Uhfortunateiy; these efforts yield measures that were statistically insignificant,
at conventional significance levels, determinants of option price. In fact, the
vast majority of the variables showed virtually no relationship across all model
specifications. t Presently, we are unsure whether or not this is attributable
to the ad hoc nature of our variables.* If so, then developing a more formal
model to reflect health status may be warranted. If not, the poor performance
may be suggesting that there is inadequate variability in health status among
our sample individuals for the differences to be significant,
12.6 REGRESSION RESULTS
This section presents some illustrative results from regression analysts
using the simple unrestricted model. The results are presented for both risk
decreases and risk increases.
*The questionnaire contained detailed smoking histories. Future research
may include constructing more thorough measures to indicate intensity of smok-
ing activity.
tSmoking was the only exception. In a few models, this variable was
significant at about 0.15 levels. This suggests that attempts to improve the
smoking variable may be more fruitful than any with the other health variables.
^Resolution of this issue will require a review of the literature on health
status and other behavioral decisions, such as participation in the labor force.
12-13
-------
12.6,1 Option Price Results for Risk Decreases
The option price regression results based on the simple model in Equa-
tion 12-1 are presented in Table 12-4. Separate models are shown for the first
risk change from Level A to Level B, for the second risk change from Level B
to Level C, and for the pooled sample of the two risk changes. Table 12-5
defines the variables that are used in the models throughout this chapter,
In general, the models do not explain a large percentage of the variation
in the option price bids, and the explanatory power does not increase as more
variables are added to the simplest version of the basic model. Nevertheless,
there are several features of the models that merit additional discussion. The
relationship between option price and income is quite strong when each of the
two risk changes is estimated separately and when estimated using the pooled
sample. This is consistent with our experience in the focus group sessions,
and especially our videotaped interviews, in which respondents consistently
mentioned that their income (and their expenses) was the most important factor
they considered in forming their valuation responses.
The level of the conditional risk also has a significant influence on the
option price bids. As In the analysis of means in Chapter 11, the sign of the
variable is the opposite of our a priori expectations. The negative sign on
the conditional risk variable implies that the respondents with the lower risk
level (1/20 in our design) had higher option price bids, all other things being
equal. This inconsistency is explored more thoroughly in Chapter 13 in the
analysis using the restricted models.
The level of exposure risk does riot affect the option price bids in any
models estimated using samples composed of either of the risk decreases. How-
ever, in the simplest model estimated on the pooled sample, there is a positive
and significant relationship between increases in exposure risk and option
price, which is consistent with a priori expectations. Yet this is the only case
that shows any significant relationship. This lack of significance is also con-
sistent with the analysis of variance results presented in Chapter 11.
Additionally, the coefficient for the dummy variable for Versions 7 and 8
has a negative sign and is statistically significant for the initial risk change
(Level A to Level B). This suggests that the lower probability portion of
the design was associated with lower option price valuation responses, which
12-14
*
-------
TABIC 12-4. MODELS FOR OPTION PRICES FOR RISK REDUCTIONS; COMMON SAMPLE*
Mode!
variables
arid Level ft to Level B risk change Levei B to Level C.risk change Pooled risk changes
summary
statistics 1234S1Z3#512345
INTERCEPT
7.249
(1.458)
2S.284
(2.736)
30.845
(2.815)
29.361
(z.ese)
29.601
(2 592)
4,536
(1.006)
13,270
(1.709)
16.969
(1.840)
16.047
(1.686)
15.204
(1.585)
5.613
(1.627)
15.047
(2,480)
19,208
(2.661)
18.200
(2.445)
17.614
(2.348)
EXP
0.041
(1.396)
0.004
(0.104)
0 004
(0.099)
0.004
(0.113)
ft. 004
(0.107)
-0.006 *
(-0.117)
-0.041
(-0,745)
-0.042
(-0.751)
-0.040
(-0.714)
-0.043
(-0.764)
0 060
(2.648)
0.040
(1.603)
0.040
(-3.302)
0,041
CI.643)
0,041
(1.616)
CONO
-0.109
(-1.783)
-0.267
(-2 911)
-0.287
(-2.885)
-0.266
(-2.890)
-0.266
(-2.884)
-0.041
(-1.368)
-0.149
(-1.944)
-0.149
(-1.941)
-0.149
(-1.948)
-0.147
(-1 917)
-0.117
(-2.886)
-0.205
(-3.322)
-0.203
(-3.302)
-0,205
(-3.31?)
-0.203
(-3.302)
INCOME
0.521
(5.766)
0.509
(5.668)
0.494
(5.412)
0.432
(5.367)
0.489
(5.304)
0.356
(4.479)
0.353
(4.Ml)
0.3S3
(4,450)
0.3S0
(4.373)
0.345
(4.295)
0.435
(7.040)
0 429
(6.937)
0.421
(§,790)
§.419
(6,111)
0.415
(6,621)
VER/8
-20.343
(-2.310)
-19.933
(-2.280)
-19. Ml
(-2.258)
-19.942
(-2254)
-10,193
(-1.310)
-10.307
(-1 394)
-10.200
(-1.376)
-10.288
(-1.386)
-10.874
(-1.888)
-10,676
(-1 853)
-10.654
(1.848)
-If.§82
C-1.848)
AGE
-0.113
(-0.844)
-0.107
(-0.813)
-0.106
(-0.801)
-0.091
C-0.745)
-0.079
(0.631)
-0.07®
(-0.631)
•0.095
(-1 064)
-0.081
(-0.882)
-0.081
(-0 873)
NUMCHD17
0,649
(0.328)
0 666
(0.333)
0 686
(0.423)
0.713
(0.433)
0.762
(0,577)
0.785
(0.594)
inform
0.386
(0.244)
2.554
(0.743)
1.616
(0.591)
r!
0.12
0 13
0.14
0.14
0.14
0.09
0.10
0 10
0.10
0.10
.0.11
0 11
0.11
0.11
0.11
F
12.28
10.69
8.72
7.26
6.21
7.38
6,02
4.92
4.11
3.SO
13.85
15, is
12.91
10.80
9.30
n
282
m
282
282
321
233
233
233
233
233
516
516
516
516
SIB
*Th» numbers in parenthtsls below the estimated coefficient! are t-stalistics for th« null hypothesis of no association.
-------
TABLE 12-5. DEFINITION OF VARIABLES
Variable
Definition
COND-RISK
EXP
AGE
SEX
NUMCHD17
INCOME
OWN/RENT
EDUC
INFORM
CAUSE
PUBMEET
Acton,
Peabody,
Woburn,
Wakefield,
Stoneham,
Wei I s ley,
Norwood,
Franklin
DUMR
Conditional probability of death given exposure tha: was
postulated to respondent, multiplied by 1,000.
Exposure risk at the starting point for the risk change
(i.e., A for thu. first risk change, B for the second), multi-
plied by 1,000.
Age of the respondent in years.
Sex of the respondent, 1 = male, 0 = female.
Number of children in the household under 17,
Household income in thousands of dollars.
Qualitative variable = 1 if respondent owned his home.
Categorical variable for last grade of school completed.
Qualitative variable = 1 if individual recalled reading aoout
hazardous wastes in news articles and the information in-
volved his town.
Qualitative variable = 1 if individual had a particular
of death in mind and 0 otherwise.
cause
Qualitative variable — 1 if respondent has attended a public
meeting about hazardous wastes and 0 otherwise.
Qualitative variable = 1 if respondent is a resident of the
relevant town and 0 otherwise.
Qualitative variable = 1 if respondent has received ranking
question for risk decrease valuation.
12-16
-------
is consistent with our a priori expectations. However, this relationship does
not hold for the valuations of the second risk change, which suggests that
respondents may have reacted differently to the lower probability versions for
the second valuation. Equally important, the performance of this variable sug-
gests that attempts to model how individuals process risk information differently
may yield substantial dividends.
In general, models estimated with the pooled sample have more significant
coefficients at higher levels than do the separate models. Although due in
part to the larger sample sizes, this suggests that more thorough investigations
of the pooling issue seem warranted. (See Chapter 13 for some of our first
attempts at examining issues related to pooling.)
As noted earlier, our efforts to improve the specification using our simple
unrestricted model were not effective. Table 12-8 shows one model that in-
cludes some of the additional variables for adjustment avenues, including resi-
dential location and health status. Again, income is the most significant ex-
planatory variable in the equation. In addition, the qualitative variable for
Acton is also positive and statistically significant across the three samples.
This relationships is intuitively piausible. Acton residents with their greater
awareness of hazardous wastes, due both to more information and the drinking
water contamination, would more likely be willing to pay more to reduce haz-
ardous waste risks. Nevertheless, the qualitative variable for the Town of
Woburn, which also has experienced problems with hazardous wastes, is not
significant. However, this lack of significance may be attributable to the rel-
atively few interviews conducted in Woburn.
12,6,2 Option Price Results for Avoiding Risk Increases
We also estimated the simple unrestricted model on the sample of option
price amounts for avoiding an increase in hazardous waste risks. (See Chap-
ter 11 for the framing of the commodity for risk increases.) Table 12-7 pre-
sents the risk increase results for the same models presented in Table 12-4.
The questionnaire elicited a valuation for only one risk change in this risk
increase case.
The pattern of the results for these models is very similar to those pre-
sented earlier. However, the income variable is again a significant determinant
of the option price bids for avoiding risk increases. Additionally, the age
12-17
-------
TABLE 12-6, MODEL FOR OPTION PRICES FOR EXPOSURE
RISK REDUCTION--COMMON SAMPLE3
Mode) variables and
summary statistics
Level A to Level B
risk change
Level 8 to Level C
risk change
Pooled
risk changes
INTERCEPT
4.869
-5.767
0.124
(0.329)
(-0.411)
(0.012)
COND-RISK
-0.092
-0.073
-0.110
(-1.479)
(-1.389) •
(-2.762)
EXP
0.081
0.025
0.146
(1.362)
(0.207)
(3.267)
AGE
-0.102
-0.116
-0.092
(-0.664)
(-0.832)
(-0.883)
SEX
-3.407
-6.148
-4.763
(-0.788)
(-.696)
(-1.665)
NUMCHD17
-0.597
-0.676
-0.481
(-0.285)
(-0.397)
(-0.350)
INCOME
0.404
0.238
0.322
(3.501)
(2.523)
(4.289)
OWN/RENT
5.264
7.125
5.834
(1.060)
(1.668)
(1.750)
EDUC
0.509
0.975
0.660
(0.648)
(1.281)
(1.200)
INFORM
-8.487
-5.483
-7.158
(-1.784)
(-1.332)
(-2.236)
CAUSE
-1,162
-0.919
-1.5^0
(-0 282)
(-0.264)
(-0.555)
PUBMEET
2 486
8.702
5.595
(0 359)
(1.588)
(1.248)
Acton
17 826
10.200
14.I 18
(3 199)
(2.240)
(3.881)
Peabody
it ;:g9
23.928
19.550
(1 .011)
' 1 601)
(1.692)
(continued)
12-18
-------
TABLE 12-6 (continued)
Model variables and
summary statistics
Level A to Level B
risk change
Level B to Level C
risk change
Pooled
risk changes
Woburn
13.194
19.811
16.613
(0.857)
(1.512)
(1.613)
Wakefield
-21.951
-7.196
-16.641
(-1.179)
(-0.283)
(-1.105)
Stoneham
34.023
43.553
39.659
(1.018)
(1.713)
(1.867)
Wellsley
-31.109
-17.138
-25.808
(-1.599)
(-0.672)
(-1.723)
Norwood
25.380
31.308
27.600
(1.507)
(2.087)
(2.407)
Franklin
35.583
-16.113
7.059
(1.356)
(-0.877)
(0.462)
Poor health
0.193
1.387
0.536
(0.032)
(0.239)
(0.127)
R 2
0.20
0.20
0.19
|=i
3.23
2.61
5.59
n
240
230
470
aThe numbers in parenthesis below the estimated coefficients are t-statistics
for the null hypothesis of no association.
12-19
-------
TABLE 12-7. MODELS FOR OPTION PRICES FOR
AVOIDING RISK INCREASES3
Model
variables
and
summary
statistics
Level
X to Level
Y risk change
1
2
3
4
5
6
INTERCEPT
14.653
(1.156)
28.759
(2.04)
33.043
(2.279)
30.726
(2.120)
30.525
(2.089)
36.996
(2.522)
EXP
-0.011
(-0.465)
-0.096
(-0.042)
-0.101
(-0.426)
-0.103
(-0.438)
-0.106
(-0.446)
-0.032
(-0.137)
COND
-0.013
(-0.118)
-0.013
(-0.113)
-0.020
(-0.176)
-0.027
(-0.247)
-0.028
(-0.248)
-0.220
(-1.704)
INCOME
0.654
(6.547)
0.620
(6.162)
0.625
(6.213)
0.601
(5.948
0,600
(5.935)
0.601
(5.993)
VER78
-15.551
(-1.249)
-14.197
(-1.144)
-14.761
(-1.189)
-15.630
(-1.264)
-15.675
(-1.265)
-21.887
(-1.754)
AGE
-0.309
(-2.266)
-0.355
(-2.509)
-0.352
(-2.500)
-0.350
(-2.477)
-0.341
(-2.431)
NUMCHD17
-2.080
(-1.211)
-2.585
(-1.173
-2.606
(-1.177)
-2.084
(-0 946)
-INFORM
9.153
(2.073)
9.142
(2.068)
8.933
(2.037)
CAUSE
0.500
(0.114)
1 .689
(0.386)
DUMR
16.225
(2.866)
R2
0.10
0.11
0.11
0.12
0.12
0.14
F
12.29
10.95
9.38
8.71
7.61
7,79
n
444
444
444
444
444
444
aThe numbers in parenthesis below the estimated coefficients are t-statistics
for the null hypothesis of no association.
12-20
-------
variable is statistically significant in all models and has the negative relation-
ship that we would expect. Older individuals would benefit considerably less
from the reduction in a risk that will not be experienced until 30 years later.
Models 4 through 8 also show that the inclusion of an adjustment variable for
whether the respondent had recently acquired information about hazardous
wastes in their town had a positive and statistically significant effect on the
option price amounts. This also seems intuitively plausible.
The qualitative variable (in Model 6) for the respondents who received
the ranking version for the risk decrease questions is statistically significant
with a positive sign. This effect may be attributable to the differences in
the sequence of valuation questions. For example, in the previous section of
the questionnaire, the ranking respondents were asked to value a reduction
to zero for their household risks along with the intrinsic value question. Re-
spondents receiving the direct question version for risk decreases had been
asked to purchase two different levels of risk reductions as well as the risk
reductions for the ecosystem. The option prices for risk increases for ranking
respondents may have been influenced in some way by the certainty question.
If the certainty question elicited higher values, and If the respondents
"anchored" on these higher values in responding to the risk increase question,
then this may account for the differences.
Alternatively! the ranking respondents may have been influenced by the
dollar amounts on the ranking cards used in these risk decrease valuation ex-
ercises. If they anchored on these amounts, this also could have affected
their option price bid. Clearly, this is a question to be investigated further
because it may enable us to understand the process individuals used in devel-
oping their valuation responses.
12.7 IMPLICATIONS
This chapter has discussed our largely unsuccessful attempts to use a
simple model to examine the influence of differences in characteristics among
individuals on option price values. The principal finding of the empirical anal-
yses of a large set of models (only a few of which have been discussed here)
is that throughout all the models income is an important determinant of the
valuation responses. Also encouraging is the better performance of the models
estimated on the pooled sample of risk changes. Even without accounting for
12-21
-------
unequal variances among individuals, the models are generally better than ones
estimated on the separate risk change samples. This suggests one direction
for the further research that is presented in Chapter 13. ?
Finally, the attempts to include variables on health status and qualitative
measures of the potential availability of avenues for adjustment in the basic
model also were unsuccessful. The direction of further research involving
these important considerations will require further evaluation of related re-
search and examination of the survey results.
12-22
-------
CHAPTER 13
VALUATION ESTIMATES FOR RISK REDUCTIONS:
RESTRICTED MODELS
13.1 INTRODUCTION
The purpose of this chapter is to report the findings of further analyses
of the contingent valuation responses associated with individuals' valuations of
risk changes. Our primary focus is on the reported valuations for risk reduc-
tions; however, we also discuss some initial results on valuations for avoiding
risk increases. This analysis is intended to suggest potential avenues for fu-
ture research with the contingent valuation data.
According to the original design of our study, this report on Phase I
activities would have concluded the contingent valuation analyses with the re-
sults reported in Chapters 11 and 12. However, based on the inconclusive
results in Chapter 12, we felt that the second phase of the project could not
be adequately planned without some indication of the results of further analysis
of the contingent valuation responses. Thus, this chapter examines three
d mensions of this further analysis:
1. Respecification of the models for the contingent valuation re-
sponse to reflect restrictions implied by the conceptual analysis.
2. Pooling of each individual's responses to two exposure risk
change questions to determine the individual's comprehension
of those questions,
3. Evaluation of the procedures for determining the outlying re-
sponses to the contingent valuation questions.
13
.2 GUIDE TO THE CHAPTER
Section 13.3 of this chapter provides an overview of the issues associated
with the relationship between the conceptual analysis and the empirical modeling
of the valuation responses. Section 13.4 discusses four econometric qualifica-
tions that apply to the analyses presented in this chapter. Section 13.5
describes our restricted model and presents estimates for models to describe
13-1
-------
the marginal values of reductions in exposure risks. Section 13.6 performs a
test for the appropriateness of pooling samples of valuation responses. Sec-
tion 13.7 presents estimates for the pooled sample that account for differences
among individuals in their ability to comprehend the framing of the contingent
commodity. Section 13.8 presents results for models of the payments to avoid
risk increases. Section 13,3 evaluates procedures for determining influential
observations, along with presenting some results of tests for thick-tailed dis-
tributions, Finally, Section 13.10 suggests some implications based on the
results of this chapter,
13.3 OVERVIEW
Based on the conceptual analysis developed in Part I of this report, it is
reasonable to expect that the initial level and the size of the reduction in the
exposure risk as well as the specified level of the conditional probability of
death given exposure would all influence an individual's value of reductions in
hazardous waste risks. For a variety of reasons, important among them the
cost of the increased complexity in the questionnaire, our design was not suf-
ficiently detailed to allow all three of these aspects of the risk to be distin-
guished. Consequently, the proposed respecification of the model to be con-
sidered in this chapter interprets the valuation responses as providing the
information necessary for estimating the "arc" derivatives of the planned
expenditure functions described in Chapters 4 and 5. That is, this respecifi-
cation interprets the contingent valuation questions as requesting the equiva-
lent of a value for the derivative of the individual's planned expenditure func-
tion. This interpretation follows from the structure of the valuation questions.
They present the individual with a risk change and ask that person to make a
state-independent payment (i.e., an option price) for it.* This approach is
*The payment mechanism is explained as a change in prices and taxes.
It does not identify all the specific commodities whose prices would increase.
It does attempt to indicate that one could not choose to consume them and
avoid the problem. Thus, our objective was to give the impression of constant
increment in the effective cost of living and taxes to individuals regardless of
the state of nature. To the extent individuals feit that changes in the compo-
sition of their budget would be possible or other state dependent planned
expenditures could be made, then the valuations do not correspond to an option
price measure of the risk change. Rather they are a change in the planned
expenditure function associated with the perceived mechanisms for adjustment.
13-2
-------
equivalent to restricting the role of the exposure risk reduction as a determi-
nant of individuals' valuation responses.
The second dimension of further analysts evolves from the psychological
literature on behavior under uncertainty and our experience with individuals'
own interpretations and explanations for their behavior in these circumstances
during the focus group sessions. When confronted with decisions involving
uncertainty, individuals comprehend the questions differently and, given the
fixed time interval of an interview, may not respond as we would expect
a priori. Although this is not a new insight, it must be considered if we are
to represent these responses within a single model.
While our survey format gave each respondent the same amount of explan-
atory information, we can expect that some individuals* responses will yield
more reliable information due to their varying abilities to understand and to
answer the questions. This expectation is based on a simple, and somewhat
informal, model of the response process, which assumes that each individual's
valuation response has two components. One component is systematic and is
influenced by the information provided and the questions asked, as well as by
the standard socioeconomic variables. The second component is purely ran-
dom. To evaluate the reliability of responses, we assume that the greater the
individual's understanding of what is asked, the smaller, ceteris paribus, will
be the variance in the random error as a fraction of the variance in valuation
responses.* To gauge the relevance of this logic to the survey responses,
sufficient information is needed to estimate these error variances for each
respondent. With several assumptions, our research design permits some crude
estimates of these variances to be developed.
In the survey design, each individual was asked to provide valuation re-
sponses for two changes in the exposure risk—from an initial level (Level A)
to a level (B) that was one-half the initial exposure risk and then to a level
(C) that was 40 percent of this intermediate level. If these two responses
can be viewed as being determined by a single mode!, there is some basis for
*This framework is broadly consistent with the framework recently dis-
cussed by Hanemann [1988] as a basis for modeling the process by which indi-
viduals form their contingent valuation responses. See Smith [1985a] for a
discussion of the Hanemann framework.
13-3
-------
estimating the extent of conformity of the individual's responses to that model.
In this framework, an observed lack of conformity, or large residual variance,
is assumed to stem from the individual's incomplete understanding the framing
of the contingent commodity. Consequently, there is greater variance in the
random component of the valuation responses.
By pooling responses for these two valuation responses, then, an Aitken
generalized least-squares (GLS) estimator can be defined using the residuals
from first-round ordinary least-squares (OLS) estimates of the model to esti-
mate the variances in the errors attributed to each individual's responses.
While there are a number of qualifying assumptions underlying this approach,
it nonetheless responds to the suggestions of economists and psychologists who
have used experimental studies to investigate behavior under uncertainty, in
most cases these analysts have commented on the differing abilities of individ-
uals to process information associated with risk.
The third arid final aspect of the extensions discussed in this chapter
concerns the sample selection activities of researchers analyzing data from con-
tingent valuation experiments. Conventional practice has been to delete some
responses, in addition to the protest bids (or refusals),* as outliers or ineffec-
tive participants in the hypothetical market assumed to be represented by the
contingent valuation question. There are a variety of explanations for this
practice, (See Randall, Hoehn, and Tolley [1981]; Desvousges, Smith, and
Fisher f 19841; arid Mendelsohn [1984] for discussion of alternative approaches.)
All acknowledge that this prescreening of the data relies heavily on the ana-
lyst's judgments as to responses that are inconsistent with the contingent valu-
ation framework based on some norm. Given the somewhat arbitrary nature of
these judgments, St is important to consider this activity as a sample selection
process and to evaluate the rationale used in implementing the process as well
as its implications for the results. Moreover, rather than automatically deleting
the outliers, explicit consideration should be given to the tendencies for
skewness and thick-tailed distributions that have been observed as typical of
these studies. Consequently, in the final part of our discussion of con tin-
~See Chapter 11 for a definition and discussion of the characteristics of
the individuals who provided protest bids.
13-4
-------
gent valuation data prescreening we discuss the use of "tests" for thick-tailed
stributions on the valuation measures used in our models,
13.4 ECONOMETRIC QUALIFICATIONS TO THE USE OF
RESTRICTED MODELS
While this chapter focuses primarily on respecifying our contingent valua-
m response model, several technical qualifications are relevant to using these
3dels. Although we discuss these technical aspects individually, they should
it be regarded as independent considerations. They are:
1. Treatment of nonprotest zero values.
2. Selection of a functional form for the model.
3. Interaction effects between the problems posed by missing val-
ues for some variables (especially the attitudinal variables) and
the problem of model selection.
4. Model selection when respondents display differences in their
understanding of the contingent valuation questions.
13.4.1 Nonprotest Zeros
As we observed in Chapter 11, there are a number of zero bids that are
not protest bids among the valuations reported for both the first and the sec-
ond risk reduction questions. Tobin [1358] has shown that ordinary least
squares on such samples will yield biased estimates. In addition, omitting the
zero observations will not eliminate the bias,*
A Tobit maximum likelihood estimator offers one method for estimating the
model, in their standard form, the assumptions for describing the likelihood
function of the Tobit estimator are more acceptable when the bids are used to
estimate the derivative of the planned expenditure function than when they
•e used as an option price equation. ^ While Tobit is an attractive approach
*See Amemiya {1984J for a thorough review of the Tobit model and the
implications of alternative treatment of the limit observations.
^Using a Tobit model for the levels of the valuation responses would face
conflict. The basic framework assumes that the true bids, y, are generated
by a stochastic process, such as
v* = x-« *¦ e.
1 I A! I
but that we observe y,
y, = y* (continued)
13-5
-------
(and will be pursued further in subsequent research), it implicitly assumes
that the zero bids represent the same level of understanding of the contingent
valuation questions as the positive bids. That is, all observations are assumed
to come from the same framework with equal error variance. Thus, the prob-
lem posed by zero responses is treated as one of observing not the true but
the unknown values of the dependent variable for bids less than and/or equal
to zero. If there is reason to expect differences In these individuals® under-
standing of the questions relative to other respondents, it may not be a rea-
sonable assumption. Consequently, the first and the last problems are related.
In this chapter, we use the OLS estimates to gauge the likely importance
of each of these problems as well as to evaluate the potential' "payoff" (in
terms of increased understanding of the factors influencing an individual's mar-
ginal valuation of a risk reduction) before formulating a specific estimator to
take account of any one of the problems posed by the survey responses. How-
ever, it is important to consider the likely consequences of using OLS when
Tobit is appropriate. In simple cases, this is possible, for example, under
the assumption that the independent variables are normally distributed and
the error follows an independent normal distribution, Greene [1383] has shown
that a consistent estimate of the slope parameters can be derived from the OLS
(continued)
when y? > 0 (or equivalently some constant y in both cases), and
y. = 0 when y* < 0 ,
i ' t -
The two simplest versions of Tobit are distinguished by whether or not we
observe the values of X when y* < 0. With observation it is referred to as
the censored formulation; without observation the model is simply a truncated
regression model. What is at issue in our case is the interpretation of the
zero responses. The valuation model may only be defined from zero to positive
values. This would imply that :
X.a > 0 for all i
I
and that individuals would not report negative valuations even as a result of
random errors. Consequently, we might expect that a one-sided error would
more appropriately describe the process. Censoring or truncating a normally
distributed error is only one way of characterizing a one-sided error. There
do not appear to be compelling a priori reasons for preferring it to others.
These same considerations are not as relevant to the modeling of the marginal
valuation of a risk change where it might be reasonable to assume individuals
would have negative marginal valuations.
13-8
-------
estimates. The Greene approach simply scales the estimates by a constant that
depends on the number of zero values for the dependent variable in the sam-
ple. This correction can be used to gauge the asymptotic bias of the OLS
estimator. In our case, with 18 percent zero bids, the asymptotic bias of the
estimates of the slope parameters would be approximately 0.22 (or 22 percent
of the estimated value of the parameter). There are a number of reasons that
this should be considered an upper bound.* Even so, it does not seem to
pose an enormous problem in relation to a large and complex set of economic
issues associated with valuing risk changes,
13,4,2 Functional Form
The second problem to be considered in developing these models arises
from the conceptual analysis of the valuation of risk. For example, the de-
pendent variable in our respecified models is an estimate of the marginal value
of a risk change, the change in expenditures for a change in exposure
risk, — . It is not a constant in R . Nor is it likely to be linear in R. Thus,
our analysis considered a variety of functional forms for respecified models,
including linear and semi-log forms. It also considered models involving Box-
Cox transformations for both dependent variable and the exposure risk measure
to gauge the sensitivity of the results to the degree of nonlinearity incorporat-
ed in the specification. ^ Although this analysis does not exhaust the potential
approaches for evaluating the appropriate specification for the model, the semi-
log specification was accepted as a first step in modeling the nonlinearity in
the valuation function. However, using the semi-log specification required a
ae
transformation of legitimate zero bids, since — = 0 in these cases. A small
*This is the maximum proportion of zero responses. As the sample size
declines because of missing observations for some of the risk perception and
other demographic variables, some zero responses will be omitted.
^There are clearly problems with the Box-Cox procedure. The specifica-
tion of the dependent variable is not consistent with a continuous, normally
distributed random variable. Since the normality assumption provided the basis
for the definition of the likelihood function, it would be incorrect to consider
the Box-Cox estimates as maximum likelihood estimates. Nonetheless, this
transformation and iterative estimation of the transformation parameter has
proved to be a fairly effective means of detecting nonlinearities. For discuss-
ions on both Sides of this issue, see Amemiya and Powell [1981] and Spitzer
[1382].
13-7
-------
constant vafue (0.00001) was added to each estimate of the derivative to ensure
that the log of ) coutd be defined In those cases where the estimate would
be zero.
Clearly, there are problems with both of these practices. For example,
as noted above, the Box-Cox method is not a maximum likelihood approach to
model selection and is probably best interpreted as a diagnostic index of the
performance of alternative transformations of the dependent variable in a
model. Equally important, the displacement in the estimated derivatives is ad
hoc.* However, if the true specification is semi-log and the zero bids reflect
an inadequate or incomplete understanding of the contingent valuation ques-
tions, then this practice may actually be superior to a Tobit estimator.* Clear-
ly, there is no basis for evaluating: this conjecture using the empirical esti-
mates. Further progress in selecting a specification and an estimator will
require a model that incorporates the prospects for zero responses and should
be an important component of the research undertaken in the second phase of
this project.
13.4.3 Missing Observations
The treatment of missing observations poses an equally difficult issue.
There is a reasonably large body of literature on the treatment of missing
observations and, in particular cases, some information on the relative perform-
ance of these approachesHowever, in all cases, these analyses assume the
true specification for the model is known, in our case, this may be a particu-
~There has been remarkably little attention given to this problem in the
econometrics literature. Presumably the reason for the lack of interest in this
practical problem follows from the increased availability of maximum likelihood
estimators for a wide range of limited dependent variable problems. Nonethe-
less, it is not unambiguously ctear that these approaches will always be super-
ior to the use of OLS with adjustments. For some early discussion of this
problem, see Johnson and Rausser [1971a], Burt {1971], Johnson and Rausser
[ 1971b), and Hu [1972] .
*The performance of Tobit in the presence of heteroscedasticity depends
on the characterization of the model of the error structure, the extent of dif-
ference in error variances, and the extent of censoring in the sample. This
issue is discussed further in Section 13.7.
tSee Maddala [1377] for a reasonably good summary of this literature.
The most detailed summaries of specific results are in Aflfi and Elashoff [1966,
1987, 1969], ::
13-8
-------
lar problem because several of the risk perception and risk attitude variables
each have a large number of observations with missing values. These factors
are of considerable interest to our analysis, but often there is little a priori
basis for specifying a role for these variables in relation to others.
There are at least two ways to proceed in light of missing observations.
First, we could reduce the sample size to the largest number with complete
observations and use this set as the basis for comparisons of alternative model
specifications. Unfortunately, when all attitudinal and risk perception measures
are considered, this strategy could lead to the elimination of most of the sam-
ple. Second, we could estimate each model on the sample of complete observa-
tions for its variables. However, in this case, comparison across models re-
flects the effects of both the differences in specified determinants and the
sample size changes. Generally, we have attempted both approaches where
possible in order to evaluate the differences in conclusions implied by each.
In particular, we will illustrate some aspects of these differences in our dis-
cussion of the models used to describe the risk reductions individually (i.e.,
from Level A to Level B, and from Level B to Level C).
13.4.4 Pooled Samples
The use of samples that pool individuals' responses to the two risk change
questions allows a GLS estimator to be applied to the pooled sample. With
rather large differences in the estimated variances for the errors in responses
across individuals, there are potentially large efficiency gains to be made in
taking account of this heteroscedasticity. This can imply that models regarded
as inadequate based1 on their OLS estimates can appear substantially improved
with the GLS estimator. This divergence in performance implies that model
evaluation using an estimator that does not take account of this heteroscedasti-
city may be misleading.
In summary, our analysis does not resolve either this problem or any of
the three problems discussed above. It identifies them as areas needing fur-
ther research before a "final" set of models that describe the factors influ-
encing individuals' marginal valuations of risk reductions can be presented.
Thus, the results reported here suggest the potential importance of pursuing
these refinements and the relative influence of each of the problems on the
results to date.
I
j 13-9
-------
13.5 RESTRICTED MODELS
The objective of the contingent valuation questions requesting individuals'
bids for risk reductions was to estimate the value placed on increments in ex-
posure risk. However, it was recognized that where the process starts and
what the individual assumes the exposure will imply should influence the re-
sponses given. The conceptual analysis developed in Chapters 2 through 8
highlights both direct hypotheses for specific variables and more indirect ex-
pectations for others. The presence of a wide set of these a priori expecta-
tions implies that a multivariate statistical framework would seem to be essential
to the development of a full understanding of contingent valuation results.
13.5.1 The Model
While our conceptual analysis and valuation concepts have been based on
the expected utility model, the structure of the questionnaire and associated
research design have attempted to provide sufficient flexibility to recognize
the potential role of these findings. For example, throughout this report,
the discussions of individual decisionmaking under uncertainty recognize that
framing can affect decisions. Individuals may use "mental shortcuts," or what
Tversky arid Kahneman [1974] describe as heuristics, to derive answers. A
frequent example is the use of a reference point or anchor from which final
judgments might be made. The level of the initial probability of exposure (or
some perception of what it is in the real world) might be serving as an anchor
for individuals* responses. If it is reasonable to expect that individuals do
engage in what Kahneman and Tversky [1979] describe as an editing phase--
i.e., organizing options and gauging probabilities before a choice is made that
is ultimately based on the weighted utilities for all the outcomes at risk—then
variations in the set of probabilities presented to individuals along with the
specific inclusion of cases involving low-probability events may increase our
understanding of this process. Moreover, under this view, attempts to deter-
mine respondents' perceptions of the events at risk should aiso be important
to understanding these valuation responses.
The basic model used in our further analysis of the contingent valuation
results is described in Equation (13.1):
13-10
-------
fib ft = f ' <13-1)
where
AE. = the contingent valuation response, which, based on the form
[ of the contingent valuation question, is a constant, state-
I independent payment (i.e., an option price),
AR = the reduction in the likelihood of exposure to hazardous
; wastes,
i R = postulated initial level of the individual's (or a family member's)
j probability of being exposed to hazardous wastes sufficient to
imply a second-stage risk, q, of death,
X = a vector of variables describing the framing of the risk posed
to each individual.
Z = a vector of measures of the individual's socioeconomic charac-
teristics, measures of attitudes toward risk, effectiveness of
government, and information on the subject.
Our conceptual analysis suggests that, for an option price payment,
would increase with the level of the risk, R, and with income. The expecta-
tion for q would be the same as for R--a positive effect on with increases
in q. However, our characterization of the relationship between the exposure
and the event at risk is important to the equivalence of changes in R and q.
In Chapter 5, we postulate that the health effect that could result from expo-
sure to hazardous waste was assumed to be exclusively from that exposure.
When there was no exposure, there was no risk of incurring the health effect.
A change in this specification would alter the role of q in the planned expendi-
ture function. For example, a simple reformulation of the problem would spec-
ify a nonzero risk of the health effect even when there was no exposure.
This would replace Equation (5.1) from Chapter 5 with Equation (13.2):
EU = Rfq Vl (W|) + (1-q) V2 (W2)j + (1-R) [jiV^O + (1-n) V2 (W2)] (13.2)
Rearranging terms, Equation (5.1) can be written as follows:
EU = (Rq + (1-R)n) Vt (Wx) + (R(1-q) + (1 -R)(1 -n))V2 (W2) . (13.3)
Under this specification, the planned expenditure function would include
and the behavior of with respect to changes in R and q would not be
13-11
-------
equivalent. Indeed, this specification is similar to the form used for the ex-
pected utility in Freeman's [forthcoming bj definition of option price in the
presence of both supply and demand uncertainty. Not only does this change
complicate the analysis of our valuation concepts, but it implies that the valua-
tion of risk changes associated with hazardous wastes cannot be treated in iso-
lation from the other risks of death an individual faces.
Based on the experience with contingent valuation experiments to date,
and especially the attempts to measure the value of one public good while
recognizing the existence and potential for differences in the levels of others,
and based on the difficulties of communicating a single risk to individuals in
the focus groups, this refinement seems beyond our current survey abilities.
However, this does not mean it can be ignored. Rather, proxy variables
reflecting an individual's health status, actions associated with other activities
that affect the risk of death, and occupation (for fob-related risks) all can be
considered in attempts to understand the valuation responses.
Beyond these a priori expectations, our remaining hypotheses for vari-
ables in the model are largely informal. Given the nature of the time horizon
specified for the health effect to be fully realized (i.e., for the individual to
die), we would expect that the payment would decrease with the individual's
age,* We also expect that it would increase with the number of children at
home.
Since increases in risk aversion should increase the rate of change in —
with R, ^ we would expect that if a variable measuring an individual's desire
*The reason for this hypothesis stems from our explanation of the event
at risk. It was suggested to each individual that exposure led to a risk of
death in 30 years. Older individuals may well condition their response based
on their expected lifetime. Under this view, a 30-year-old individual at the
time of the proposed exposure would be more concerned about this risk than
one 70 years old, An alternative hypothesis, which was also supported in some
focus group sessions, is that older respondents based their bids on concern
for children and grandchildren. This might imply an interaction effect between
age and size of extended family, especially if those family members lived near
the hazardous waste facility.
t
This result can be seen with some manipulation of the second partial der-
ivative of the marginal valuation with respect to R when an option price is
specified as the payment mechanism. The slope of the marginal valuation func-
tion for an option price bid can be directly related to the Arrow-Pratt index
of absolute risk aversion for one of the state-dependent utility functions.
13-12
-------
to participate in gambles accurately measures risk aversion, then the estimated
impact of this variable would be negative for —~ . This qualification is impor-
tant. To the extent individuals treat different types of risk differently, then
we cannot expect that their responses favoring one form, of risk-taking behav-
ior will necessarily transfer to another. Unfortunately, this implies that we
cannot formulate a clearcut hypothesis on an index of risk-taking behavior
without first accepting a questionable (based on an increasing body of psycho-
logical research) assumption as a maintained hypothesis.
Several of these same issues are relevant to the "unrestricted" estimates
and models presented in Chapters 11 and 12 and may partially explain the dis-
appointing experience with the models reported in Chapter 12, It is clear that
prior information has a fairly limited role in determining the specific variables
to be considered and the a priori expectations for their effects on the valuation
responses. Consequently, analysis of these results requires searching a wide
range of specifications and progressive refinements in the model formulation
based on the results of that process. This is certainly not a new practice in
empirical research in the social sciences generally and economics in particular.
Nonetheless, it is a practice that has received increasing criticism. (See
Learner 11983) for a largely informal discussion of the deficiencies in such con-
ventional practices. ) At the same time, to dismiss the practice completely is
to reject the learning that can take place from such analyses of sample informa-
tion . This point was made by Theil [1361] 25 years ago. What is really at
issue is explaining the process used and the factors in that process that might
impinge on what are reported as the "final" results.
The overall sequence of steps in our analysis of the valuation responses
corresponds to the structuring of the three chapters that describe the results.
The initial analysis was confined to evaluation of the summary descriptive sta-
tistics and test results in Chapter 11, then multivariate regression analysis of
these responses using models that were linear in variables and parameters.
The results we now report are the beginning of the final stage of the process.
As we noted at the outset, they involve restricting the form of the models used
with the survey data and reexamining the conclusions derived ort the deter-
minants of the valuation responses.
13-13
-------
13.5.2 Estimates for Marginal Valuation of Exposure Risk Reduction
Table 13-1 reports a selection of the estimates for the semi-log models--
AE
i.e., the dependent variable is log (the marginal valuation oe risk
reductions as a function of a variety of variables. The variable definitions
are given in Table 13-2. Several general observations can be made with re-
spect to these models. First, separate equations were estimated for the first
and the second risk reductions proposed to each respondent. Second, in con-
trast to our discussion of the form, of the model at the outset of this section,
the conditional probability of death given exposure to hazardous waste was
not included in any of these models. It was never found to be a statistically
significant determinant of the marginal valuation. Table F-1 in Appendix F
repeats these models with this variable included.
Overall, these estimates identify several statistically significant determi-
nants of the marginal valuations. Four variables deserve particular attention.
Income is generally a significant influence on the marginal valuation. While
there are two cases for the risk, change from A to B where it would not be
judged to be significant using conventional criteria, the estimated parameter
for income is quite stable across all models including those for the second risk
change (i.e., from B to C).
The exposure risk is not a positive influence on the marginal valuation
as our theoretical analysis for the case of an option price predicted. Marginal
valuations appear to decline with increases in the exposure risk, but do so at
a decreasing rate. It should be noted that the deletion of the squared term
for the exposure risk does not change the negative estimate for this term.
This quadratic term was included in an attempt to investigate whether or not
these reductions in marginal valuations were continuous over the full range of
probabilities we considered. The positive effect of the quadratic term would
be consistent with a change in direction of the change in the marginal valuation
as the level of exposure risk increased.
This finding of a negative effect for the level of exposure risk was clearly
not expected a priori. While our conceptual analysis can provide an explana-
tion for this outcome, it requires that we assume individuals have opportunities
outside those posed in the contingent valuation question to make state-
dependent adjustments in response to the risk changes posed to them. This
13-14
-------
TABU 13-1. MODELS FOR MARGINAL VALUATION OF EXPOSURE RISK REDUCTIONS: COMMON SAMPLE'
mooci variants*
HoiM
•ntf-MnnMrir
StltlitJcS
1
2
3
4
1
f
3
4
Intercept
0.214
(0,544)
0.156
CO,391)
-0.128
(-0.2S9)
0 268
C-0,544)
0.095
(0.215)
0,080
10,178)
-0.187
C-0.333)
-0.306
(-0,536)
EXP
-5.024
(•2.913)
-0,021
(-2 884)
-0.023
(-2.799)
-0.022
(-2 671)
-0 04b
(-2 504)
-0 04S
(-2.511)
-0.044
(-2.443)
-0 .©«
(-2,330)
EXP*
0.054*10^
(1.S96)
0.054-10'3
(1.586)
O.OS2*10"3
(1.513)
0 049x10"3
(1.429)
0.177*10 3
(1 217)
0.182*10 3
(1-243)
0.180*10*'
CI.212)
0.172*10 3
(1 157)
NUMCHD17
--
—
-0.009
(-0.093)
--
—
--
-.003
(0.02S)
...
INCOME
0.011
<2.422}
0,010
(2.235)
0.009
(1.769)
0,009
(1.820)
0.011
(2.389)
0.011
(2.117)
0.011
(1.994)
0.012
(2.197)
VER78
1.611
(4.040)
1.628
<4,056)
1.642
(4.062)
t .670
(4.114)
1.840
(4.161)
I.B54
(4,159)
1.882
(4.176)
1.938
(4.250)
Acton
—
0,168
(0,888)
0.154
(0.790)
0.217
(1.088)
--
0.185
(0.87S)
0.168
(0.783)
0.184
(0.830)
Cambridge
--
—
0,158
(§.347)
--
—
—
-0.214
(-0.609)
Kingston
~
—
—
1 408
(1 143)
--
--
--
1.3®4
(1.138)
Stlem
—
—
0.702
(0.972)
--
--
0,376
(0.525)
Woburn
--
--
-•
1.020
(1.375)
--
--
—
0.490
(0 659)
Afle
--
--
0.005
(0.651)
0.004
(0.550)
--
0.006
(0.703)
0.006
(0 634)
REfF
--
—
0.131
(0.293)
0.030
CO.066)
--
--
0.622
(1,256)
0.588
(1.1S2)
RI5K-AT7
--
--
0.247
(1.297)
0,291
(1.508)
—
—
0,024
(0,112)
0.025
(0.115)
REL-RtSK
--
0.008
(0,575)
0.006
(0.441)
0 007
CO,831)
-0,035
(-0.495)
-§,089
(-1.093)
-0.078
(-0.944)
«*
0,833
0,635
0.641
0,849
0,674
0.677
0.683
0.688
f
75.05
43,930
29.942
23.482
70,32
46.744
27 942
21,586
r>
178
1/8
178
178
140
140
140
140
»»
1,448
1.456
1.489
1.460
1,371
1 380
1.397
1.403
numbers in parentheses below the esliroaled coefficients ara l-
statistics for the
null hypothesis of no association.
-------
TABLE 13-2.
DEFINITION OF VARIABLES
Variable name
Definition
EXP
MUMCHD1?
INCOME
VER78
Acton, Cambridge,
Kingston, Salem,
Woburn
Exposure risk at the starting point for the risk change
(i.e., A for the first risk change, B for the second)
multiplied by 1,000.
Number of children in the household under 17. |
Household income in thousands of dollars.
Qualitative variable that is unity for the low probability
design points (versions 7 and S of the contingent valua-
tion questionnaires) and zero otherwise.
Qualitative variables that equal unity if the respondent is
a resident of the relevant town, zero otherwise.
Age
GEFF
RISK-ATT
REL-RiSK
COND-R ISK
Age of the respondent in years,
A qualitative variable measuring individual's perception
of government effectiveness, equal unity if government
is considered not at all effective, zero otherwise.
An index of attitude toward risk based on individual's
responses to a hypothetical lottery with constant expected
value; a value of unity corresponds to an individual who
is perceived as liking risk, zero otherwise, ;
A measure of individual's ability,to perceive risk; ratio
of risk of death perceived by individual In comparison
to estimate of actual risk of death from accidents on the
job.
Conditional probability of death given exposure that! was
postulated to individual, multiplied by 1,000.
INFORM
GOVT
Qualitative variable = 1 if individual recalled read
about hazardous wastes in news articles and the in
mation involved his town.
Qualitative variable equal to 1 if individual rece
versions of the questionnaire with Section G specified
as the federal Government has decided to allow the risk
increase
ng
for-
ived
13-16
-------
explanation may be plausible. However, it is more difficult to interpret given
the positive estimated parameter for the VER78 qualitative variable (i.e.,
VER78 = 1 for the low-exposure probability cases and 0 otherwise). This vari-
able was included to investigate whether the low-probability design points
(i.e., design points D7 and D3) elicited a different type of response than the
higher probability cases (i.e., design points D1 through D6). The positive
statistically significant coefficient for this variable with all models indicates
that individuals appear to have a higher marginal valuation in these cases in
comparison to the higher probability cases (i.e., those with probabilities 10
times larger). This does seem to conform with arguments of psychologists,
most notably Kahneman and Tversky [1979], that individuals have difficulty
in dealing with low-probability events, and respond differently in these cases.
Nonetheless, these findings must be interpreted cautiously. As we observe in
Chapter 8, risk circles were used to convey the probabilities associated with
the risk reductions that were asked in each contingent valuation question. In
the case of the low-probability design points, it was not possible to display
darkened areas for all three probabilities. Figures 13-1 and 13-2 repeat the
risk cards used for these design points. As shown, the combined risk circles
are blank in all cases; respondents were told that the probabilities on them
were too small to display. Nonetheless, it may lead to differences in their
responses in comparison to cases where probabilities can be displayed that are
not easily explained within conventional models.
This difference in the materials used with the questionnaire for these low-
probability design points was investigated in several video-taped interviews as
part of the process of evaluating the final questionnaire. In those interviews,
it did not appear to influence the respondents' understanding of what was
asked. However, this estimated positive effect could be a reflection of the
difference in the materials used to explain the risk changes.
Kahneman and Tversky's [ 1979} analysis would explain this positive effect
through their probability weighting function, which is assumed to describe
how individuals translate probability information into their perceived likelihood
of certain events. Instead of the objective probabilities used in the expected
utility framework, the Kahneman-Tver sky framework uses these weights to
combine the values realized under different states of the world. Based on
13-17
-------
D-7
Card A-7
Risk of Exposure
(thirty-three
hundredths
ol 1 percent!
Possible
Pathways
Risk of Death
if Exposed'
Heredity
(sod Health
Combined Risk;
Exposure and Death
Personal
Risk
D-7 Card B-7
Risk of Death CoroWrted Risk:
Hisk of Exposure if Exposed Exposuw awl Death
Risk
D-7
Card C-7
Risk of Exposure
rOS Stole
Pathways
Risk of Death
if Exposed
Heredity
and Health
Combined Risk;
Exposure and Death
Figure 13-1, Examples of low probability risk cards: design point 7,
13-18
-------
D-8
Card A-8
Risk of Exposure
Possible
Pathways
Risk of Death
if Exposed
Htwdlty
and Health
Combined Risk:
Exposure and Death
Risk
D-8
Card B-8
Risk of Exposure
Possible
Pathways
Risk of Dwrth
If Exposed
Heredity
and Health
Combined Risk:
Exposure and Death
D-8
Card C-8
Risk, of Exposure
Pttssibte
Pathways
Risk of Death
'• txpossd
and Health
Combined Risk:
Exposure end Death
Risk
Figure 13-2, Examples of low probability risk cards: design point 8,
*•
13-19
-------
their experimental evidence, Kahneman and Tversky have suggested that low
probabilities are generally overweighted and high ones underweighted.
An alternative explanation, consistent with the expected utility model,
would suggest that individuals have differential confidence in probabilities that
are presented to them. That is, each is implicitly regarded as an estimate
with a corresponding density function. Low-probability events are rare.
Therefore, it would be entirely reasonable to expect that they are less pre-
cisely estimated. This would imply greater second-order uncertainty with the
low-probability cases and the positive coefficient for VER78 as a reflection of
individuals' responses to this perceived uncertainty. At this stage of our con-
ceptual analysis both of these explanations are observationally equivalent.
Further conceptual and empirical research will be necessary to explain these
differential responses to the low-probability events.
The remaining variables in these models do not appear to exert significant
effects on the marginal valuations. Our age and number of children (NUM-
CHD17) variables have estimated parameters that are inconsistent with our
a priori expectations but are not significantly different from zero. The risk
attitude (RISK-ATT) and risk perception measures (REL-RISK) are not sig-
nificantly different from zero. A qualitative variable for whether the respond-
ents have little confidence in government effectiveness (GEFF) does not appear
to be a significant determinant of the marginal valuations. This finding was
somewhat surprising because these attitudes were found in the focus groups
to be closely associated with the individual's willingness to support regulations
that would reduce exposure risks from hazardous wastes. In effect, those
individuals with confidence seemed more willing to believe that government
could "deliver" the risk reductions specified.
Finally, some of the models include qualitative variables for the towns in
our survey area that had disposal sites with hazardous wastes as of 1982 (see
Chapter 15 for a specific identification). As we observe in Chapter 10, there
has been a great deal of information on hazardous wastes in this region. In-
deed, Acton residents in particular have had a decade of experience with con-
tamination episodes (see Table 10-4, for example). However, none of these
town variables appeared to influence these marginal valuations.
13-20
-------
The missing observations problem identified earlier can be easily illustrated
with these results. AH of the models reported in Table 13-1 for the first risk
reduction (i.e., A to B) were estimated using the same sample by reducing it
to the set consisting of complete information on all of the variables of interest.
Had the process been treated differently, the resulting estimates would have
been quite different. Consider, for example, the simplest model for A to B
in Table 13-1 (Mode! 1). Reestimating this model with the sample containing
full information for this variable produces the results in the first column of
Table 13-3. There is a 58-percent increase in the sample size and substantial
changes in the estimated parameters. The most notable of these is the change
in the magnitude and statistical significance of the qualitative variable, VER78,
intended to identify the responses corresponding to the low-probability sce-
narios. The same differences arise for the second risk change from B to C
and are given in the second column of the table.
It is also interesting to note that these results reinforce the inclusion of
the linear and quadratic terms in the exposure risk. Clearly, if this simple
specification is regarded as the final form of the model for the marginal valu-
ation of risk reductions, the results based on the larger sample would be pre-
ferred. However, the decision is not as clear for the problem of comparing
results across models. As we observed earlier, there is the further problem of
differential understanding of what has been asked and the potential for heter-
oscedasticity. Indeed, this problem appears to dominate the missing observa-
tion issue in terms of its importance for model selection. Thus, as a practical
matter, we shall argue that model comparisons should be based on the pooled,
GLS estimates of each model. Only these estimates take account of the poten-
tial for differential understanding across respondents and reflect the full avail-
able sample information. Of course, these comparisons will need to recognize
the potential effects of differences in the sample sizes for the apparent per-
formance of each specification.
13-6 POOLED RESPONSES OF RISK CHANGE VALUES
The estimates reported in Tables 13-1 and 13-3 are not without value.
They can be used to gauge whether pooling responses to the two risk reduc-
tion questions is warranted based on the estimated parameters for the variables
of greatest interest to this analysis. That is, since we have argued that the
13-21
-------
TABLE 13-3. MODELS FOR MARGINAL VALUATION OF EXPOSURE
RISK REDUCTION: SPECIFIC SAMPLE®
Model
Model variables
Level A
Level B
Pooled |
and summary
to Level B
to Level C
risk j
statistics
risk change
risk change
change!
s
Intercept
-3.124
-4,030
-4.090
(-2.800)
(-2.907)
(-5.549)
EXP
-0.050
-0.033
-0.032
(-2.218)
(-0.560)
(-2.148
)
EXP2
0.197xi0"3
0.168X10"3
0.143X10
-3
(2.114)
(0.350)
(2.156
)
INCOME
0.069
0.044
0.057
(6.026)
(2.857)
(6.134)
VER78
0.074
1.278
1.131
(0.067)
(0.908)
(1.531
)
R2
0.151
0.074
0.104
F
12.355
4.544
14.872
n
282
233
516
'
s2
18.326
23.337
20.952
The numbers in parentheses below the estimated coefficients are t-statisties
for the null hypothesis of no association.
13-22
-------
selection of a final model must be "made within a framework that takes account
of the potential for differences in the understanding of the contingent valuation
questions across individuals, we cannot select a specification based on the in-
dividual responses to each question. To do so requires that we first test
whether the model for responses to the first question appears different from
the model for responses to the second question. This type of test would in-
volve a Chow test. However/ it requires that the model specification be known
in advance.*
For this preliminary analysis, we have used the Tiao-Goldberger [1962]
test for stability in individual coefficients of a larger model across alternative
estimates of that model (i.e., one for each sample). In this case, it provides
the basis for comparing the models associated with the two risk changes.
The specific test statistic is given in Equation (13.4):
where
L
number of distinct models
b.
OLS estimate of a parameter from the jth sample
b
P.
weighted average of the OLS estimate defined in (13.5) beiow
T -1
diagonal element of (X X). for the relevant parameter
55R, = sum of squared residuals for jth sample
T. = number of observations in jth sample
K
number of parameters in each model .
Equation (13.5) defines the weighted average of the OLS estimate;
L
(13.5)
*lt also assumes homoscedasticity in the errors.
13-23
-------
Under the null hypothesis of equality of parameters, TG follows an F distribu-
L
tion with degrees of freedom (L-1) and I (T, - K). This test was applied to
j=1 J
all pairs of parameters from comparable specifications for the models used with
the two risk changes. The estimated parameters in Table 13-1 in Models 1
through 4 for the risk change from A to B and those for the corresponding
models for the change from B to C do not lead to rejection of the null hypoth-
esis of equality. Moreover, this conclusion does not appear to be affected by
allowing the sample size to be different for each model. While we did not. con-
sider all of the models in Table 13-1, those that were considered, including
the models in Table 13-2, lead to the same conclusions. Consequently, on the
basts of the available information, pooling the responses across the two risk
changes should provide a reasonable basis for estimating the degree of re-
spondent comprehension of the contingent valuation questions.
13.7 RESTRICTED MODELS: POOLED SAMPLE AND GLS ESTIMATES
Table 13-4 presents the estimated marginal valuation equations based on
pooling the responses to the two risk change questions. Each model was esti-
mated with the largest number of complete observations. All of the variables
described for the individual analysis of each risk change have been considered,
along with some additional variables that were not found to be statistically sig-
nificant in those initial specifications. They were reconsidered because we
have argued that the selection of a final specification requires that the heter-
oscedasticity induced by differences in the understanding of the contingent
valuation questions be taken into account with model specification. The models
in Table 13-4 represent the first step in the development of the GLS esti-
mates. They provide the basis for estimating the residuals associated with
each respondent's valuation responses. For those individuals answering both
questions there are two estimates of the residuals. These are used to estimate
the residual variance as follows:
2
13-24
-------
"ABLE 13-4, ORDINARY LEAST-SQUARES ESTIMATES OF MARGINAL VALUATION MODELS; POOLED SAMPLE3
Model variables
Model
dnd 5um*rt#ry
statistics
1
2
3
4
S
S
7
Intercept
-4.390
-4.20?
-2,883
-2.271
-2.320
¦2.438
-2.311
(-5.549)
(5 706)
(-3.625)
(-2.184)
<-2.183}
(-2.340)
(-1.738)
INCOME
0.056
0.052
0 027
0.026
0.025
0.026
0.082
(8,134)
(5.470)
(2.509)
C2.277)
(2,210)
. (2.240)
< ) ,9Wl
E
XP
-0.032
-0,032
-0.021
-0.01?
-0.017
-0,016
•0.015
(-2,148)
(-2.118)
(-1.328)
(-'.063)
(-1.054)
(-0,966)
(-0.886;
EXP8
0.143«10"3
0 141*10'3
O.087X10*3
0,075*10~3
0 075»10*3
0.071X10"3
0.069»1Q*3
(2.186)
(2.131)
<1.229)
CI,052)
(1.044)
(0.984)
(0,948)
vj€ft78
1.131
1.153
1.978
2.144
2 148
2.138
2,036
a.ssn
(1.565)
(2.525)
(2.887)
(2.SS7)
(2.654)
(1.851)
Acton
0 923
0.588
J). 583
0.787
1.214
(2 039)
(1.25B)
(1.2*5)
(t.5381
£2.1721
R
EL-R1SK
-w
-0.024
-0.028
•0.029
-0.025
-0.027
C -0.742}
(-0.861)
( -0.029)
(-0.761)
(-0.821»
Age
»-
..
-0.026
-0,029
-0.032
-0.026
(-1.54$)
(-1,554)
(-1.681)
(-1.328 >
ft
ISK-ATT
..
..
0,117
0.184
0,209
CO,260)
(0.381 }
(0.4S6)
G
EPF
..
..
1.895
1,885
1.702
1,502
CI.800)
(1,786)
f 1.588)
1 1 369)
Cambridge
0.901
1.235
(0.339)
(1 1231
S
aiem
..
--
..
..
--
1,783
t 827
11.0305
< 1 . 109?
W
bburn
—
..
..
1.548
2,005
(0,932)
i 1,1881
Kingston
--
..
..
..
3.778
4,6=2
(1.254)
(1,519t
NUMCHD17
..
..
..
0.097
0 103
0. 129
(0.435)
(0.456)
k0,5S5>
ffjlFORM
--
..
-0.827
1-1,5341
CONO-RiSK
..
..
..
-0,003
t-0,312'
R?
0.104
0.111
0.090
0.107
0.107
0.117
0,123
F
14.872
12.80
7,«
4.874
4,382
3.67
3, 156
n
SI 6
516
381
375
375
37S
375
5*
20.952
20.823
17,52?
17.SIS
17.589
17,549
17.560
afhe numb*ri Below the coefficients are t-raiios for the null hypothesis of no association.
13-25
-------
where
k = the subscript to identify each respondent
ejk = the estimated residua! for respondent k to risk change
l,£jk = the average value of the residual for respondent k.
Two estimates of these residuals were considered in forming the GLS esti-
mates for each model. The first was based on the OLS residuals and thp sec-
ond on a scaled residual that, under the assumption of homoscedastic errors,
would have a scalar covariance matrix.* Since there are not substantia! differ-
ences in the resulting GLS estimates between the two estimated covariahces,
only those based an the OLS residuals are reported in the following discus?-on.
Before turning to the GLS results, aspects of these pooled OLS results
deserve attention. Income remains a positive and statistically significant deter-
minant of the marginal valuation. However, the estimated effect for income is
substantially larger than those reported in Table 13-1 with either risk change
using the more limited subsets of the data. It is more consistent with the re-
sults observed with the larger samples as reported in Table 13-3. Indeed,
*The OLS residuals will not have constant variance. If e. designates the
OLS residual, the variance in e. is given as
Var (e.) = u£ (1 hj) ,
where
hj = x. (XTX)"1x, T
x. = ith row of matrix X
i
X - TXK matrix of regressors used in model to derive the model
used to form the OLS residual.
The scaled residuals are the studentized residuals:
e.
~ _ i
S| s(i) VT^hTj
where s(i) is the standard deviation of the residuals omitting the ith observa-
tion. If the error structure has constant variance, the studentized residuals
will also have constant variance. See Belsley, Kuh, and Welsch [1980] for
further discussion.
13-26
-------
the pooled results for the simplest model are repeated in that table for com-
parison purposes. The same pattern also characterizes the results with the
exposure probability variables for the models associated with responses to the
first risk change question. The estimated parameters have the same sign pat-
tern but the absolute magnitude of the measured effects is larger for the model
estimated with larger samples. The estimated parameter for the qualitative
variable associated with the low-probability design points is also somewhat un-
stable across the alternative samples. However, in this case, the estimated
parameters are larger with the smaller sample. Again, this problem seems to
be confined to the models for the valuation responses to the first risk change
question.
Tabte 13-5 reports the GLS estimates for the same models. Clearly there
are rather substantial differences in the ability of the model to "explain" the
determinants of these estimated marginal valuations. The absolute magnitude
of the income effect tends to rise with the number of variables included in
the model, but in all cases falls within the range of estimates derived from
the analysis of the subsamples with OLS. The absolute magnitude of the expo-
sure probability effects declines with the inclusion of additional variables.
The signs conform to the earlier results.
(n contrast to those earlier findings, a farge number of factors appear,
based on the GLS results, to influence the marginal valuations. A few vari-
ables will be discussed in particular. One notable change is the negative and
statistical significant effect for the conditional probability. Had we relied ex-
clusively on the OLS results, the models would have remained very simple
(i.e., similar to Models 1 through 3 in Table 13-5). With the OLS results,
this variable was never a statistically significant determinant of the marginal
valuation. These estimates contradict our a priori expectations, indicating
that higher levels of "susceptibility," as might be implied by a higher condi-
tional probability, reduce an individual's marginal valuation.*
*They are, however, consistent with the results with the estimated means
far the valuations of risk changes for the design points with comparable expo-
sure risk changes and differing conditional risks. See Chapter 11 and Figures
11-7 and 11-8.
13-27
-------
TABLE 13-5, GENERALIZED LEAST-SQUARES ESTIMATES OF MARGINAL VALUATION MODELS*
Model variable
and summary
statistics
$
Model
1
2
3
A
5
6
mttrctp!
-0.018
(-0.074)
-0.165
(-0.573)
•0.282
(-1.003)
-0.665
(-1.344)
-0.781
(-2.286)
-2 187
{-7.484}
+ 2 336
(-lu 38C
INCOME
0.017
(6,065)
0.012
(4.167)
0.019
(5.615)
0.039
(12.418)
o osa
(12.229)
0.049
(18.284)
0.051
<20 ;bi
EXP
-0.051
(-11.222 5
-0.043
{-7.631}
-0.044
(-9.116)
-0.036
(-1.304)
'0.037
(-8.732)
-0.0J5
(-6.880)
ji;
e-u
EXP1
O.2O1"0'3
(10.060)
0,1?5xie"3
(6.4515
0.164»10~3
(8,237)
0.127*10"3
(7,172)
0.13O*10"3
(7.441)
0 590*10 ^
(£.819)
0.075' 'I/'
(18 31 V»
VER78
1,578
(3 900)
1.959
;3.587)
1.623
(5.1801
1.456
15 642)
1 .445
(5.737)
1.760
<7.351)
i *" Sfc'J
Ac ton
--
-0.487
(-3,701)
..
0.218
(1.660)
0.118
(0.917)
0.301
{2.204)
u b04
i:« jy»
REL-RISK
--
—
0.118
(3,32*)
-0,044
(-0.700)
-0.078
(-1.27®)
-0.049
(-0.754)
j 49
i53
Age
--
—
--
-0.008
(-1,302)
-0.010
(-1,812)
-0.003
(-0.498)
C 'Sib
>.Z 3?S
RISK-ATT
~
—
--
--
0.554
(4.479)
0.496
(4,058)
-0 i136
(-0 3fl7
GEFF
—
—
--
1.003
(2.161)
1,014
(2.243)
0,796
(1.723)
• "8
12 >84
Cambridge
--
—
—
—
—
-0. 229
(-0.890)
-0 440
(-1 480
Salem
--
--
--
--
1,263
(1.827)
1 .49
: ! •!<&
Woburr
--
—
• —
--
„
1 .277
(2.422)
0 '-Si
i t
Kingston
--
—
--
-
2.008
(2.121)
3 jbfc
s 3 b 5i:
MUMCH017
--
—
' -
-0.235
(-3.793)
-0.179
(-2.893)
..
-'J '4.1
i-J '64
INFORM
--
—
—
--
—
--
l - "
COND-RI5K
--
--
—
--
--
--
-0 Hi
t "4
RJ
0 334
0.254
0.555
0,693
0,708
0.787
o.ste?
F
75.33
33.06
83.297
87.306
84,183
97.601
15 3*2 '3 7
PI
48?
mi
363
357
357
357
357
®The rumt>«r» in parentheses below th» •ttirruitcd coefficients »re the ratios of the P4r»m«t*r estimate 10 the estimated
standard error and follow normal distribution »sympiotic»ily .
13-28
-------
There are a variety of potential explanations of these results, including
incomplete understanding of the role of the conditional probability in the proc-
ess? a form of cognitive dissonance such that individuals assigned higher con-
ditional probabilities were unwilling to accept them as relevant to their specific
circumstances; or the ability to take actions to mitigate the effects of these
risks which Implies state-dependent adjustments. Clearly, there may also be
others. At this stage, we cannot discriminate between them. However, it
should also be acknowledged that the negative sign for the conditional risk
measure appears to be relatively sensitive to model specification. Table F-2
in Appendix F reports a selected set of the simpler models estimated with gen-
eralized least squares but expanded to include this variable.
The estimated effects for age and the number of children less than 17
years old in our most detailed model (Model 7 in Table 13-5) are also contrary
to our a priori expectations, although, in both cases, there are plausible argu-
ments to explain these measured effects. For example, the larger the number
of children under 17, the greater the demand on household income and hence
the lower the ability to pay. This effect couid easily be expected to affect
both the level of feasible payments and the marginal increment to those pay-
ments with a further reduction in risk.
The qualitative variable associated with the individual's lack of confidence
in government's effectiveness has a positive and statistically significant effect
on the marginal valuation contrary to what we would have anticipated based
on the focus group sessions. Of course, it should be acknowledged that the
coding of this variable focused on the lowest extreme in perceived Ineffective-
ness on a scale from 1 to 10. Further analysis will be necessary of the impli-
cations of how this and other attitude variables are used in understanding the
determinants of the valuation responses.
One more encouraging result concerns information-related variables.
Where the variables measuring familiarity with hazardous waste—the information
variable and the qualitative variables for residents in towns with hazardous
waste facilities — are statistically significant, they generally exhibit a plausible
association with the marginal valuation. Neither the risk perception nor the
risk attitude variables were significant determinants of the marginal valuation
in the most detailed model. While they did appear to be statistically significant
13-29
-------
determinants for other specifications, the signs of the measured parameters
changed with the model and were opposite to those in the most detailed model.
It is difficult to evaluate the implications of this instability with our ava iable
information,
However, in both cases, improvement in the information used in these
measures is possible. At present the variables are fairly crude attempts to
reflect these influences. Other potential specifications for the risk attitude
information are possible and need to be explored. For example, the individ-
ual's perceived risk of a fatal accident on the job is compared with an approx-
imate estimate of the actual risk. This latter variable can clearly be improved
upon. Moreover, greater attention should be given to individuals reactions to
differences between their risk perceptions and risk information. These reac-
tions should be important to individuals' valuations of risk reductions. Over-
all, these findings are clearly supportive of the need for and potential payoff
to further research with these restricted models. It seems reasonable to
expect that this further analysis will need to incorporate explicitly adjustment
for heteroscedasticity with the estimation of the model. Thus, application of
a standard Tobit estimator would lead to inconsistent estimates. Indeed, there
may be little basis for preferring this approach over GLS. Based on a much
simpler model, Arabmazar and Schmidt [1931], for example, found that the
severity of the impact of heteroscedasticity depended on the nature of the dif-
ferences in error variances and the extent of censoring in the sample. Vari-
ance differences less than a factor of two and/or censoring of more than half
of the sample are the thresholds that they suggest. As we observed earlier,
our sample clearly falls within the range for censoring with 18 percent of the
sample for the first risk change and a much smaller percentage of the pooled
sample exhibiting a zero bid.*
The fraction declines with the pooled samples because of the question-
naire design. If individuals responded zero to the first risk change, they
were not asked the second risk change as a change from level B to C but were
asked instead for the complete change from A to C. These responses are
analyzed differently in our summary statistics. Therefore, we add two obser-
vations for all positive responses and only one for the zero responses, reduc-
ing substantially the share of the sample that is zero -esponse.
13-30
-------
At present the GLS sample is composed primarily of nonzero responses
because of the question sequence. Only zero responses to the second risk
change were included. A zero response to the first risk change altered the
questioning format. Further analysis will be necessary to determine whether
these observations can be included in the pooled sample. Regardless of the
treatment of the zero observations, heteroscedasticity seems to be a very
important problem. The differences in the estimated sample variances are pro-
nounced, greatly exceeding a factor of two. Nonetheless, with two responses
per individual, it should be possible to develop a maximum likelihood estimator
that accounts for the zero responses and the heteroscedasticity.
13.8 MODELING THE PAYMENT TO AVOID A RISK INCREASE
This section presents some initial results from an application of the model
for describing the estimated marginal valuations of risk reductions to the re-
quest for payments to avoid risk increases. These increases were described
using the same endpoints for the level of exposure probabilities as for the risk
reductions. However, as described in Chapter 11, they were discussed in
separate sections of the questionnaire. Individuals were informed that these
were completely different situations.
The design of the questionnaire and the survey implies that ail respond-
ents--! .e. , the full sample — were asked a question indicating that a medium-
sized company was located in their town, 3 miles from their home. This com-
pany was described as producing hazardous wastes and disposing of them in a
landfill on the site. Individuals were shown the risk card corresponding to
the lowest level of risk in their design (card C for the contingent valuation
questionnaires and card C for the contingent ranking questionnaires). A
change in the volume of the wastes placed at the site was to be allowed. This
change was described as providing the prospect of increasing risk to the high-
est level for the particular design point (card A). Individuals were then asked
what they would be willing to pay to avoid the increase in risk. These re-
sponses are the valuations considered in the models presented in this section.
Two further points should be noted, when the risk change was explained to
respondents, it was described as being permitted by "the government" or
"your town council." Recall that we discussed the implications of this distinc-
13-31
-------
tion for the mean bids in Chapter 11. (The complete text of the question is
available in Chapter 11.)
We did not attempt a detailed analysis of restricted models for the valua-
tion responses for avoiding risk increases. Rather, we considered two issues;
(1) did the semi-log specification with the exposure risk measure perform as
well as with the responses to realized exposure risk decreases; and (2) were
the estimated parameters in these two models comparable? To investigate these
issues, we considered only the respondents to the contingent valuation ques-
tions. In further research we plan to consider the other component of the
full sample.
Table 13-8 reports a sample of the estimated models. They'conform to
the simplest of the specifications used wtih the models based on the risk re-
duction questions and include the Acton town variable only because of the
amount of activity related to hazardous wastes contamination in Acton. They
also include a variable, GOVT, that identifies the individuals whose valuation
question explained the increase in risk as due to "the government" (i.e,
GOVT=1 in these cases) rather than "your town council" (i.e., GOVT=0 in
these cases).
Income, the qualitative variable for the low-probability design points,
and the qualitative variable for residents of Acton are statistically significant
in all of the models considered. The four models in Table 13-6 illustrate the
general nature of the findings. While exposure risk has a negative effect on
the marginal valuation that declines in absolute magnitude with the level of
the exposure risk as in the case of bids for risk reduction, neither exposure
risk variable is significantly different from zero.
It is interesting to note that several of the estimated parameters in these
models are approximately the same order of magnitude as the parameters esti-
mated for the risk reduction models in Table 13-3. The only notable excep-
tions are the coefficients for the exposure risk squared (EXP2) and the quali-
tative variable for the low-probability design points.
Under the assumption of independence, the Ttao-Goldberger [1962] test
statistic was calculated to gauge whether each pair of these estimated coeffi-
cients would lead to a rejection of the nuli hypothesis of equality for the pop-
ulation parameters. Based on these findings the null hypothesis could not be
rejected.
13-32
-------
TABLE 13-6. MARGINAL VALUATION TO AVOID RISK INCREASES
Mode
>def variables
and summary *—
statistics 1
Model
intercept -5.439 -5.584 -5.527 -5.641
(-4.969) (-5,127) (-4.907) (-5.036)
INCOME 0.076 0.069 0.076 0.070
(6.823) (6.081) (6.820) . (6.072)
EXP -0.106 -0.102 -0.107 -0.102
(-0.960) (-0.929) (-0.959) (-0.929)
EXP2 0.002 0.003 0.002 0.003
(1.036 (1.009) (1.032) (1.006)
ER78 2.123 2.141 2.114 2.135
(1.962) (1.991) (1.949) (1.981)
Acton -- 1.234 -- 1.227
(2.180) (2.160)
GOVT — -- 0.171 0.111
(0.343) (0.223)
R2 0.192 0.206 0.193 0.206
,
F 16.31 14.18 13.03 11.78
n; 278 278 278 278
s*; 17.184 16.951 17.239 17.010
13-33
-------
Clearly, there are a number of additional issues that must be addressed
with these valuation responses. The research to this point has not attempted
to refine the models used for these valuation responses. It has not exploited
the full sample or considered the full set of potential determinants of the mar-
ginal valuations for avoiding risk increases. Finally, as in the case of the
models for the risk decrease bids, a Tobit estimator would be more appropriate
for these models. However, in this case there is only one valuation response.
Consequently, if this model is assumed to be different from the relationships
describing the valuation responses for risk reductions, there is not sufficient
information to take account of heteroscedasticity. Of course, based on the
tests with the simple models considered to date, it would be desirable to con-
sider further testing to determine whether responses for risk avoidance seem
consistent with a pooling of all three responses.
13.9 INFLUENTIAL OBSERVATIONS, THE ROLE OF JUDGMENT
IN SAMPLE SELECTION AND THICK-TAILED DISTRIBUTIONS
All of the statistical analyses of the valuation responses in this report
have used the full sample excluding only the protest bids. The protest bids
were eliminated because a separate set of questions was used to determine if a
zero bid was intended as a "legitimate" zero bid — i .e., it was all the individual
could afford or it was what he felt the "commodity" was worth. (See Chapter
11 for more details.) In this section we consider past practices used in eval-
uating contingent valuation responses to judge the individuals" acceptance of
the contingent valuation questions and the corresponding definition of some
observations as outliers, the treatment of outliers implied by the GLS estima-
tion of our models for the estimated marginal valuations, and the results of
several tests for the extent to which the contingent valuation bids appear to
come from thick-tailed distributions. This last issue was included in our dis-
cussion of outliers because it clearly relates to the plausibility of some of the
indexes used to judge these outlying observations, since symmetric distribu-
tions have been assumed to characterize the bids. Equally important, given
the sensitivity of Tobit and related maximum likelihood estimators to the distri-
butional assumptions used in characterizing the model's error (see Goldberger
[1980] and Amemiya [19841), the character of this distribution affects the
potential for using Tobit (with an adjustment for heteroscedasticity) as an
13-34
-------
alternative to a sample selection process that eliminates some observations as
outliers,
13.9.1 Past Practices in Screening Contingent Valuation
; Responses for Outliers
It has been common practice to reduce the samples derived from contin-
gent valuation studies by deleting what are judged to be outlying observations.
The difficulty with this process arises in establishing some basis for defining
these outlying observations. There are two general approaches in the litera-
ture: (1) a purely statistical criteria that establishes a fixed bound for "legit-
imate" responses (e.g., the valuation responses would be judged as legitimate
if they lie within k standard deviations of the mean, where k is a fixed con-
stant), and (2) a criterion that combines statistical and economic considerations
by focusing on the effect each observation has on the percentage change in
an estimated parameter that is hypothesized, based on an economic model of
the valuation process, to be an important determinant of the responses.
However, both approaches are ad hoc. They rely on analyst judgment.
The first relies on an implicit assumption that the distributions of the valua-
tion responses are symmetric and have finite variances. Presumably, the
rationale for a fixed bound follows from an assumption that responses outside
the interval realized 99 percent of the time by nearly all finite variance sym-
metric distributions cannot have been from individuals who understood or were
willing to take seriously the contingent valuation questions.*
There are several problems with this approach. One of the most crucial
is its focus on the valuation responses. By treating all bids as capable of
being described by a constant mean distribution, the scheme ignores the role
of other economic variables that we would expect to influence these bids. In-
come, for example, would be expected to influence the bids, leading to differ-
ing mean bids from households with differing incomes. Depending on the par-
ticular application, there are likely to be other variables as well. With the
present study, measures of the character of the risk and the risk change
would be expected on a priori grounds to influence the bids. This method
^Examples of this approach include Randall, Ives, and Eastman (19741,
Brcokshire, Ives, and Sehulze [ 137*6]# and Rowe, D'Arge, and Brookshire
[1980].
13-35
-------
has no systematic means for taking these influences into account. However,
for a large value of k, the multiple of the standard deviation defining the
interval for legitimate bids, it is possible to indirectly reflect the role of these
other influences,
A value for k is easily established for finite variance symmetric distribu-
tions. Consider, for example, the normal distribution. If the threshold for
legitimate responses is a 99-percent confidence interval, then k would be
three. By increasing k beyond three (many of the past contingent valuation
studies have used ten) the analyst implicitly allows for variation In the mean
at the center of the interval. This variation could be the result of income or
other variables that described the respondent or commodity that was being
valued. Of course, the difficulty with this adjustment is that its effectiveness
will depend on the influence of specific economic variables on these mean bids
and the extent of variation in these variables in any particular sample. This
would imply that it is not possible to define in advance a fixed criterion for
the size of the interval as a constant multiple for the standard deviation about
the mean.
The second approach proposes the use of regression diagnostics (see
Belsley, Kuh, and Welsch [1980]) to gauge the influence of each observation
by its percentage impact on the parameters of variables that can be specified,
in advance, as potentially important determinants of the valuation responses.
I n the first application by Desvousges, Smith, and McGivrtey [1983], this was
the estimated parameter for household income in a model for the option price
bids for water quality improvements. A 30-percent change in the estimated
parameter with the deletion of one observation was the criterion for identifying
the outlying observations.
While this procedure is also ad hoc and to some extent arbitrary, it has
some advantages over the first approach. First, it recognizes the potential
for differing mean bids for individuals with differing incomes or other socio-
economic characteristics. Second, it expresses the objective of the screening
process as one of identifying observations that do not appear to be consistent
with or to have accepted the contingent market. This is accomplished by
focusing on the observation's influence on an estimated parameter of an eco-
nomic variable that would influence the bids if the respondents were providing
13-36
-------
their actual valuations of the commodity of interest. Finally, it evaluates influ-
ential observations within a framework that treats the bids as the sum of sys-
tematic and random components. The greater the size of the estimated residu-
al'5 implied adjustment to the parameter, the larger will be the regression
diagnostic. Large changes in the estimated parameters for important economic
variables are treated as indications that these observations may not be drawn
from the same model. In effect, they may not have accepted the terms of the
contingent market on the same basis as the other respondents. Of course,
defining a large change is purely a judgment. However, the first application
of this approach by Desvousges, Smith, and McGivney [1983] did acknowledge
the need to examine the features of the observations judged to be influential
in order to determine if there were similarities in their economic or demographic
characteristics.
Nonetheless, these advantages do not change the fact that the procedure
is arbitrary. It uses an index of influence to determine observations that are
judged to be inconsistent with the contingent valuation framework and cleariy
requires analyst judgment. Mendelsohn [ 1384J has recently criticized the ap-
proach on the grounds that an observation's influence on income is "only of
passing interest" to the objectives of most contingent valuation studies. He
argued that the central objective of the screening should be to remove obser-
vations that are incorrectly affecting the mean valuation. To meet this objec-
tive, he proposed using a set of variables associated with potential biases in
the contingent valuation responses (e.g., qualitative variables for the inter-
viewer, question format, and starting point, etc.) and relating them to the
contingent valuation bids. Such models could then be used to predict what he
referred to as the bias component of the contingent valuation responses. Of
course, as he acknowledged, economic variables could also be associated with
incentives to provide biased responses. Hence, it is entirely possible that one
could not specify any model that would isolate biases (as distinct from legiti-
mate economic influences) in contingent valuation responses.
In addition, his specific criticism: of the Desvousges, Smith, and McGivney
[1383] analysis apparently overlooked that all of the variables that Mendelsohn's
bias analysis called for were included in the option price equation used for
the regression diagnostics. Thus, the calculations of Desvousges, Smith, and
13-37
-------
McGivney for indexes of influence were providing a set of gauges of the error
in the valuation responses after taking account of specific sources of error,
such as the interviewer effects and form of the question. Mendelsohn's other
criticisms ignore the fact that the mean responses can be expected to change
across respondents because of differences in their economic circumstances.
Judgments on outliers that ignore this feature of the process face the same
problems we outlined for the statistical approach,
13.9.2 Judging Influential Observations for the Present Study
The regression diagnostic framework requires that we be able to specify
a mode! to explain the character of the valuation responses. in the present
application, the role of important attributes of the contingent valuation commod-
ity and of economic determinants of the responses was not clearcut on a priori
grounds. Use of one model as a standard to gauge the influence of observa-
tions relies on the acceptability of the general form of that model even if the
statistical fit to the particular equation is not necessarily good. Since we have
not completed the process of model selection, it would be premature to evaluate
the sample responses for outliers. Our most detailed model does not include
variables for the inteviewers and will require further analysis, both from the
perspective of considering new potential determinants of the marginal valuations
and reformulations of some of the existing variables to better understand what
they may be representing. This process is especially important where the esti-
mated parameters are found to be sensitive to the specification of other deter-
minants of the marginal valuations.
However, our findings implicitly take account of the observed differentials
in respondents' understanding or acceptance of the contingent valuation ques-
tions. This is accomplished in the weighting of observations according to the
estimated variances attributed to each respondent based on their bids (and
the associated estimates of the marginal valuations) for the two risk changes.
By pooling responses across questions, we imposed additional structure on the
analysis of individuals who may not understand or accept the contingent valu-
ation experiment. It is possible to judge two of their responses in relationship
to the model used to describe the marginal valuations of risk changes.
This approach is consistent with the regression diagnostic format used in
earlier studies, but does not eliminate observations. Rather, it gives them
13-38
-------
small weight in the estimation of the marginal valuation models. Indeed, Learner
[1984] recently used a GLS framework to characterize regression approaches
that discard or reweight outlying observations. By defining the variance-
covariance matrix as the sum of the conventional least squares weight (i.e.,
the identify matrix) plus a matrix that reflects the reweighting implied by
informal approaches to treating outliers, he established that the corrected or
reweighted estimates are adjustments to the OLS estimates of the parameters
based on the judgmental criteria, the least squares residual for each specific
observation, and values of the independent variables, Moreover, this approach
can be shown to be closely related to the Belsley, Kuh, and Welsch [1980]
regression diagnostics for the selection of given criteria for influential observa-
tions, Of course, it differs from the regression diagnostic/outlier approach
in its treatment of the summary statistics describing contingent valuation
responses. That is, in Learner's case, these weights are used only in the
regression models for the bids. They are not taken into account in calculating
the estimated mean valuations reported in Chapter 11.
Nonetheless, this approach offers an alternative basis for judging outlying
observations. Further research will be needed to judge the plausibility of
using regression diagnostics based only on one of the risk changes with the
estimated variances that reflect the weight of individual observations. In addi-
tion , it will be important to evaluate the sensitivity of both approaches to the
specification of the model used to describe the estimated marginal valuations
of risk changes. Finally, this further research must consider the economic
and demographic features of the respondents whose valuations are judged to
be outliers. *
13.9.3 Some Preliminary Tests for Thick-Tailed Distributions
A number of contingent valuation studies have found some evidence of
bids coming from either skewed or thick-tailed distributions. ^ Distinguishing
*lf the detection of outliers can be done in a more systematic way, it
should be possible to treat the problems it poses as analogous to the selectivity
issues addressed in many recent econometric analyses of survey data.
^This was one of the observations made by the review panel for the state-
of-the-arts assessment of contingent valuation methods (see Cummings et al.
[1:984] for more details.
13-39
-------
the two features of distributions from sample information can be quite difficult.
Both imply more weight in the tails of the distribution than one would expect
for a symmetric distribution such as the normal, Skewness arises when one
tail has more weight than another. By contrast, thick-tailed distributions can
be symmetric but have a flatter density at the center than the normal.
The objective of this section is to report the results of several tests for
the degree of thickness in the tails of the underlying distribution of valuations
for risk reductions based on the sample responses. Our-analysis has been
conducted using the valuation responses and the transformed measure of the
marginal valuation used in our restricted regression models of the determinants
of the individual bids (i.e., log ). Based on Monte Carlo studies comparing
uK
the power of five tests of normality against alternative distributions with either
lighter or heavier tails, Smith [1975] found the kurtosis (K), U (Uthoff
[1970]), and V statistics were more powerful in detecting heavy-tailed distri-
butions. The three statistics are defined as follows;
n
2 (X. - X)4
i=1 '
n
r n "]
X
I (X. - X)2
L i-i 1 J
n
[x (X, - *).] '
u =
n
I
i-1
n
X. - xm
i m
(13.7)
n
V =
1r
2
i I
l X. " Xm
n
(13.8)
13-40
-------
where
X. = ith value of the relevant response (i.e., valuation or trans-
formed estimate of marginal valuation - log (— ))
X = sample mean
X -• sample median
m ^
R = sample range
n = sample size.
Table 13-7 presents the estimates for each statistic using the actual bids in
response to the first and second risk reductions and for the case of avoiding
risk increases. Using the empirical critical values reported in Smith [1375]
for a 5-percent significance level and a sample size of 100 observations as ap-
proximations for the relevant critical values, ali three statistics would reject
normality in favor of a thick-tailed (or potentially skewed) alternative distribu-
t on for all three valuation responses.* These results should not be particular-
ly surprising given that the valuation responses are truncated at zero and
there is considerable heterogeneity in the character of the risk change that
was presented to sample respondents. By contrast, when these tests are re-
peated using the transformed-measures of the marginal valuations, the conclu-
sions are not as clearcut. These results are reported in Table 13-8. With
the marginal valuations for risk reductions, two of the three statistics (i.e.,
vj and K) would not reject the null hypothesis of normality at the 5-percent
level. Only the U statistic would call for a rejection. In the case of the valu-
ation responses for avoiding risk increases, two of the three statistics (i.e.,
U and K) would call for rejection of the null hypothesis while V would not.
*The specific values for the estimated critical values by significance level
and sample size are:
=0.05
=0.10
n=20
n=50
n=l00
n=20
n=50
n = 100
k
4.035
3.812
3.670
3.651
3.543
3.375
U
1 .382
1.330
1 .304
1.346
1 .314
1 .291
V
3.101
3.446
3.720
2.936
3.273
3.578
See Smith [1975], p. 665.
13-41
-------
TABLE 13-7. TESTS FOR THICK-TAILED DISTRIBUTIONS
WITH LEVELS OF VALUATION BIDS
Test statistic
Risk change
U
V
K
First risk reduc-
tion (A to 8)
1.8?
7.11
15.90
Second risk reduc-
tion (B to C)
2.14
10.32
34.94!
Avoiding risk
increase (C to A)a
1 .79
10,12
30.5?j
aThese results relate to the sample of all respondents. For the subsample of
individuals receiving the contingent valuation questionnaires, the results]
were; U = 1.93, V = 8.65, and K = 20.98.
TABLE 13-8. TESTS FOR THICK-TAILED DISTRIBUTIONS WITH
TRANSFORMED ESTIMATES OF MARGINAL VALUATIONS
Test statistic
Risk change U V K
First risk reduc- 1,58 2,48 3.34
tion (A to B)
Second risk reduc- 1.43 2.26 2.57
tion (B to C)
Avoiding risk 1.73 2.93 4.41 ,
increase (C to A)a
a AE 1
'These results are based on using tog (-75) as the transformed measure of the
ZiK
marginal valuation of each risk change.
13-42
-------
There are several important qualifications to these findings. First, these
tests are approximate and should be interpreted only as indicative of the pres-
ence of thick-tailed distributions. They are not definitive. Measuring skew-
ness and the degree of flatness of a density near the center (or equivalently
thickness in tails) is difficult. The two phenomena are easily confused by
statistics intended to indicate their presence. Only a complete set of moments
will determine a random' variable's distribution exactly.
Second, as we noted earlier, substantial differences in the valuation re-
sponses in relation to the sample mean could be expected solely as a result of
differences in respondents' income and the character of the risks asked of in-
dividuals. To avoid this inherent heterogeneity in what individuals responded
to, it would have been necessary to calculate the indexes for each design
point. White this would not control for individuals' characteristics, it would
eliminate the problems posed by heterogeneity in the risk changes posed to
individuals. Unfortunately, the sample sizes become quite small and the power
of these tests correspondingly diminishes.*
Finally, we could have used the residuals from a model that relates the
valuation responses to what has been asked and to the features of the indiv-
idual, Since our research has not identified a final model, the results with
the transformed marginal valuation represent an intermediate position. That
is, the estimation of controls for the size of the risk change. The logarith-
UkK
mic transformation also serves to reduce the skewness in the distribution of
bids.
Overall, these results seem to suggest that evidence of skewness and/or
thick tails in the distribution of contingent valuation bids may well indicate
the need to pay more specific attention to the determinants of valuation re-
sponses and to the specification of the functional form for the bid (in our case,
option price) models. Based on these preliminary results, a thick-tailed dis-
tribution does not seem to be a likely characterization for the risk reduction
responses. The results for valuations to avoid risk increases is more dispar-
ate. It may well be that these responses are more correctly modeled as arising
from a non-normal distribution.
*See the Monte Carlo results in Smith (1975] as one example.
13-43
-------
Further research into the characteristics of the residuals from formal eco-
nomic models for all responses will be necessary to complete a final evaluation
of this issue.
13.10 IMPLICATIONS
The primary objective of this chapter has been to explore the implications
of using restricted models to summarize the valuation responses for the expo-
sure risk reduction questions on the questionnaire. These results were not
intended to be final. Rather, our objective was to consider whether there
would be a payoff to more restrictive modeling of the contingent valuation re-
sponses in terms of the increased understanding of the determinants of individ-
ual valuation of risk changes.
These restricted models were based on estimates of the marginal valuation
of the risk changes. After examination of a number of specifications, they
were developed using a semi-log form. Based on these preliminary findings,
the answer to the question of potential payoff to further research appears to
be clearly in the affirmative. Moreover7 in the process of developing these
estimates it has been possible to consider several additional issues.
One of the most important of these issues involves the pooling of valuation
responses across the two risk reduction questions. Based on the estimates
for all of the models applied to the valuation responses for risk changes indi-
vidually, pooling appears to be acceptable. This implies that the two responses
can be used to gauge the variance in the errors associated with each respond-
ent. These variance estimates have several important uses. They permit the
construction of a GLS estimator for the valuation models and may well provide
a superior method for identifying respondents who reject the contingent valua-
tion framework.
Our analysis also considered the potential relevance of the mode! to the
valuation responses for avoiding risk increases. Here our preliminary work
has been more limited. The initial results with the application of these models
was less successful. While they appear to indicate that income and Che risk
measures play the same role in determining the marginal valuations (i.e., the
null hypotheses of equality of their respective parameters could not be reject-
ed), this finding must be interpreted cautiously. Further analysis with the
13-44
-------
full sample of these responses arid more extensive model specifications is clearly
warranted,
Another dimension of our analysis involved reconsideration of the proce-
dures used for analyst intervention in deleting outlying contingent valuation
responses, before analysis of the mean valuations or modeling and testing of
the valuation responses In relation to other variables,. After reviewing both
approaches to this screening, including recent criticisms of the approach based
regression diagnostics, we used Learner's 11984} recent work to argue that
GLS estimator and the diagnostic approach are closely related. Hence, the
LS weights may well be the most appropriate basis for judging respondents
ho have not understood or accepted the contingent valuation questions.
Finally, our preliminary analysis of the extent to which the valuation re-
sponses appear to have been generated by thick-tailed distributions suggests
that this problem cannot be considered independent of the modeling of the be-
havioral function that explains the determinants of the valuation responses.
Further research with this information will require refining the econo-
metric methods used to analyze the responses, explicitly taking account of the
implications of zero bids for the selecting of probabilistic models of the data
generating process as well as refining the character of the variables intended
to measure the individual's attitude toward information on and perception of
risks.
13-45
-------
CHAPTER 14
THE USE OF CONTINGENT RANKING MODELS TO VALUE
EXPOSURE RISK REDUCTIONS: PRELIMINARY RESULTS
14.1 INTRODUCTION
1 ne purpose of this chapter is to describe the design and results from the
contingent ranking component of our survey. The contingent ranking was used
a$ an alternative means for eliciting individuals' preferences for reductions in
the risk of exposure to hazardous wastes. This approach was first proposed in
the economics literature by Beggs, Cardell, and Hausman [1981] as a method for
measuring the demand for new commodities. Subsequently, Rae [1981a,b] used
it as an alternative to contingent valuation for measuring the benefits associated
with improving environmental amenities.*
Since its introduction to the literature, the approach has attracted con-
siderable attention. However, to date, specific applications using contingent
ranking for benefit estimation have been limited. Desvousges, Smith, and
McGivney [1983] applied the method as part of their comparative evaluation of
methods for estimating water quality benefits. More recently, the application of
the method to valuing visibility improvements in Cincinnati have raised ques-
tions concerning both technical issues in the modeling and the estimation of
contingent ranking utility functions and the sensitivity of the benefit estimates
to the model selected to describe household rankings. Accordingly, there is a
need for further research on both the performance of the contingent ranking
method and, equally important, the consistency between independent applica-
tions of the contingent ranking and contingent valuation methodologies to value
a specific amenity change.
~There has been some controversy over the plausibility of the results
derived from contingent ranking studies. Indeed, based on the sensitivity of
the findings of the most recent effort to use the contingent ranking approach to
model specification, the Electric Power Research Institute (EPRI} has sponsored
a reevaluation of the contingent valuation and contingent ranking approaches
for estimating the values of changes in environmental amenities.
14-1
-------
Most contingent valuation studies have been structured to describe a
hypothetical market to the individual in a way that places that individual in an
active rote in the market—as a bidder for a specific outcome. That is, the
valuation questions have requested bids from individuals for stated changes in a
carefully defined commodity. In effect, the individual is confronted with the
propsect of being able to "purchase" the change (if it is a utility-enhancing
increment) or pay to avoid it (if it will decrease utility).
In a few cases, notably Bishop and Heberlein [1979]* and Bishop et al.
[1984], the contingent valuation experiment has been structured so that the
market outcome—a commodity (or change in a commodity)--and a price are pre-
sented to the individual; then he is requested to accept or reject the stated
outcome. Contingent ranking can be viewed as an expansion on this second
approach to describing the market. That is, several specific outcomes involving
payments and changes in the commodity (or outcome) of interest are presented
to the individual as possibilities. Rather than requesting a single choice or a
yes-no decision on each, the interviewer asks the respondent to rank the
choices from most to least preferred.* These types of responses may wefl be
easier for individuals to answer and, as a consequence, may lead to more
accurate information than attempts to directly elicit individuals* valuations.t
To date there has not been a direct comparison of the contingent valuation
and contingent ranking methods using independent sample information.+ As
*All of the applications of the contingent ranking method have not allowed
individuals to suggest that subsets of the options being ranked would tie in
their preference ordering. Ties cannot be accommodated in the Beggs, Carded,
and Hausman [1931] formulation of the decision process. It is, however, possi-
ble to amend the framework to permit ties (see Cox [1972]). This is a poten-
tially important addition to the method, because it appears that individuals
undertake the rankings by first identifying the extreme alternatives—the most
preferred and least preferred cases--and move to the choices that are most
difficult to order in the middle. At present the method requires individuals to
order all. This may introduce greater error in the center of the ranking
relative to the extreme choices. ;
tSee Hanemann [1984, 1985] for further discussion of the modeling of
individuals' responses to valuation questions.
fThere was comparative information on the contingent valuation and con-
tingent ranking approaches reported in the EPRI -sponsored Cincinnati study.
However, this material is not in the public domain. Therefore, it cannot be
reviewed at this time.
14-2
-------
we noted earlier, in a study designed to estimate the- option price for water
quality improvements, Desvousges, Smith, and McGivney [1983] used both the
contingent valuation and the contingent ranking approaches to elicit values for
water quality changes. However, all individuals were asked both the contingent
valuation and the contingent ranking questions. Consequently, we might expect
that the responses to the first type of question asked would affect the
responses given to the second. Indeed, the authors note that several
individuals who refused to provide values for the water quality changes in a
contingent valuation format (which was administered first in their interviews)
also recognized the contingent ranking questions as eliciting the same type of
information and refused them as well. While there was a high degree of
consistency in the two sets of estimates reported in their study, the lack of
independence in their administration makes this finding difficult to interpret. It
may simply indicate that individuals attempted to be consistent in their re-
sponses .
Our survey design for the present study was structured to provide inde-
pendent contingent valuation and contingent ranking estimates. By separating
the sample into two groups, one group (approximately 60 percent of the sample)
receiving the contingent valuation questionnaires and the second receiving the
contingent ranking format, it was possible to develop independent estimates
from each approach for individuals' valuation of exposure risk reductions.
This chapter reports the preliminary results from an analysis of these
contingent ranking responses and a simple comparative assessment of the valu-
ation estimates implied by these unrestricted contingent ranking models in rela-
tion to the contingent valuation responses. It is described as preliminary for
several reasons. First, the models used to describe individuals' rankings of
risk-payment combinations do not attempt to impose any theoretical restrictions
on the functions estimated for the indirect utility functions. Second, our com-
parison of the valuation responses and estimates from the contingent ranking
approach are designed as illustrative comparisons using the contingent ranking
models to estimate the bids that are implied for the contingent valuation re-
spondents. Our findings suggest that the relationship depends in important
respects both on the benefit concept used and, especially, on the contingent
ranking model selected for the comparison.
14-3
-------
There are two aspects of these findings that should be noted. First,
estimates of the mean (by town) changes in payment required to hold utility
constant in the presence of a reduction in exposure risk, when positive, are
generally larger than the contingent valuation responses for comparable risk
changes. Second, in an evaluation of two of the estimated contingent ranking
utility functions for two specified exposure risk reductions and four subsets
of the contingent valuation sample, only two cases fail to reject the nuil
hypothesis that would suggest contingent valuation bids and the corresponding
predicted contingent ranking responses for the same individuals cluster about
a 45° line. Of course, these results are based on fairly preliminary models
for the random utility function that do not reflect the restrictions one might
wish to impose based on theory. Nonetheless, they do permit standardization
for individual respondents' characteristics through the process used to con-
struct the estimates of the payment changes which hold utility constant.
Aside from this relative comparison of the contingent valuation and con-
tingent ranking findings for consistency, the results do confirm the earlier
Desvousges, Smith, and McGivney [1983] finding that the contingent ranking
framework was easily understood by respondents. This finding was anticipated
prior to the survey. The experience with the focus group sessions used to
develop the questionnaire and in the pretest of the survey instrument bath
indicated that the ranking tasks were more easily accomplished by the individ-
uals involved.
Section 14.2 begins the chapter with a brief review of the random utility
model and the issues that arise in applying it under uncertainty. This section
also provides a description of the maximum likelihood estimator based on a log it
formulation for the random utility model. In the third section the form of the
ran king questions is described together with the experimental design for the
survey using the contingent ranking questionnaires. Section 14.4 presents
the results. Beginning with some informal information on the patterns of rank-
ings that emerged from the survey, this section describes the estimated models
used to in erpret the rankings and discusses their sensitivity to specification
and version of the questionnaire used in the survey. Section 14.5 outlines the
approach used to estimate the value of risk reductions from these models and
provides some comparative information on the relationship between these asti-
14-4
-------
mates and the contingent valuation estimates for comparable risk changes. Sec-
tion 14.6 summarizes the findings of the analysis and discusses their implica-
tions for further research.
14.2 THE RANDOM UTILITY MODEL
The random utility model has had the greatest application in modeling con-
sumer behavior with respect to discrete choices. These choices are assumed to
involve some degree of indivisibility, so that conventional marginal analysis in
describing the incentives to consumer choice is not directly relevant. The
individual is described as comparing a set of specific alternatives and selecting
the one that yields the greatest total utility. In this framework the analyst is
assumed to observe a set of individuals and their choices, but does not have full
information on individual preferences. Behavioral observations are a set of
trials, each one representing different individuals making choices. With
assumptions concerning the distribution of types of individuals and information
on the characteristics of the specific individuals who made particular selections,
it is possible to describe the conditional probability of the choice of the com-
modities of interest.
To develop this framework in more specific terms, let Equation (14.1)
describe individual i's utility function for commodity j« Individual i's charac-
teristics are described by a vector X., and the attributes of the commodity by a
vector Cj:
M(C., X.) = v(C., X.) + e(Cj, Xj) , (14.1)
where
(j(.) is the total utility provided to an individual with X. features
from a commodity with C. characteristics. 1
It has a deterministic component, v(.), and a stochastic component, e(.)• To
describe the conditional probability a commodity with attributes will be se-
lected by an individual with characteristics X., we must specify that the prob-
ability m(C^, X.) will exceed all the possible alternatives as in Equation (14,2):
M(Ck, X.) > p(Cn, Xj) for all n ^ k (14.2) •
or, substituting
14-5
-------
v(C. , X.) + e(C , X.) > v(C , X.) + e(C , X.)
k i k i n i n t
4.3)
Equation (14.3) can be rewritten as
v(CX ) - v(C , X ) > e(C , X ) - e(C X ) . (14.4)
k i ni n i k i i
By specifying a distribution for the stochastic component of utility, e( . ), and
specific set of determinants for v(. ), Equation (14.4) can be transformed into
a specific probability statement. For example, assuming that the e's follow
independent, identically distributed Weibull distributions, then the probability
expression for Equation (14.4) is given as follows;
exp (v(C X.))
Prob [jj. > |i for all n t k] = —— —— —- (14.5)
k n
I exp(v(Cn, X.))
n=1
where
T = all feasible alternatives including the kth.
All of this framework has been developed with little direct specification
of how the v(.) functions relate to the conventional theory of consumer choice.
While in many applications the connection has remained loose, it can also be
argued that for conventional choice problems under certainty, that v( . ) is sim-
ply an indirect utility function with the prices of commodities among the C.
i l
and the individual's income among the X..
The form selected for v(.) has generally been linear in parameters.
Nonetheless, as McFadden [1981] observed, it is possible to impose the theoret-
ical properties of an indirect utility function on v( . ) by appropriately defining
the roles for C. and X. in the nonstochastic component of the utility function.
Once this framework is used to model decisions under uncertainty, the same
observation can hold. For example, we might consider the two state case where
C^. and C,,. correspond to the vectors of commodity specific variables in each
state and n., the probability of the first state, v(. ) would then be described
J
as follows:
vfC , Cp., n , X ) = 7i V (C , X ) + (1-tt )V (C X ) , (1tt,6)
1j 2j j l J I J J i \ I l\ \
14-6
-------
#bere
V.(.) = state dependent, nonstochastic utility function.
in this form we are assuming that e(. ) arises because of stochastic influences
that are not associated with the uncertainty in the indivdiual's choice among
the commodities in differing states of the world. e( .) represents uncertainty
that is independent of the choice process among commodities. For the empiri-
cal analysis described in this chapter we have treated the exposure risk similar
to any other commodity (or characteristic). Thus, we have not attempted to
pecify forms for the V.(.) functions and derive v(.) as a restricted function.
o illustrate the distinction, consider the two alternative approaches to speci-
fying v( . ). In both cases suppose for simplicity the individual faces only two
choices, one involves a stochastic outcome the second does not. They might
be described as follows:
Choice 1 Choice 2
Type 1 charact? ur rex of Cn C2i
1
ype 2 characteristic 1-% of C12 0
With Choice 2, then the n2 implied by the general definition given in Equation
(14.6) is unity. We can assume C1X, C12, and C2i can be either vectors of
attributes or scalar quantities. The argument holds for either case. The un-
restricted model implied by a linear specification for the nonstochastic compo-
nent of utility might be Equation (14.7a) for choice one and Equation (14.7b)
for two:
v(Ci, X.) = OfqH^ + ofiC 11 + $212 * cfg-X. f (14. 7a)
where
Cj_ = (C|| C 1 /
where
c2 = [c21 0] .
v(C^, ,X.) — Sq ^ Cg] ^ Cfg Xj / (14.7b)
14-7
-------
In this case, the two values of the function differ only as a result of the sub-
stitution of the relevant values for each variable. It should also be noted that
this example is deliberately simple. In practice we could not estimate the par-
ameters for the X. variables unless these variables were assumed to enter in an
interactive format with the characteristics of the commodities being ranked.
This is easily appreciated by substituting Equations (14,7a) and (14.7b) into
(14.4). In a I inear-in-parameters framework, the differences in the indepen-
dent variables contributing to individual utility are what motivate the assign-
ment of a relative standing. Without these interactions, individual character-
istics cancel in determining an individual's ranking of the commodities.
Past research has addressed this issue in different ways. With a sufficient
number of commodities being ranked separate functions can be estimated for
each individual. This is one approach proposed by Beggs, Cardeil, and
Hausman {1981] and corresponds to what Rae designated the "individual" model.
By contrast, one can consider defining interaction variables between the
characteristics of the commodities being ranked and the attributes of the
individuals doing the rankings. This is the approach used by Desvousges,
Smith, and McGivney [1983] because there were not a sufficient number of
choices being ranked. It amounts to a specification of how the parameters for
the elements in C are expected to change with changes in individuals' attri-
butes. To some extent it was treated arbitrarily by Desvousges, Smith, and
McGivney. That is, each specification for the attributes of individuals thought
to be important determinants of a ranking was interacted with each of the
characteristics of the commodities being ranked and the findings compared.
This simple example does serve to illustrate that it may be possible to improve
on this practice using more information from our conceptual model of the deci-
sion process.
To consider how we might alter the specification of the random utility-
functions to more specifically reflect the conceptual framework/ assume the
state-dependent functions are linear in type one and two characterstics, but
give each function different weights depending on the state of the world, as in
Equations (14.8a) and (14.8b) below. Then we have Equations (14.9a) and
(14.3b) as the "deterministic" utility functions corresponding to the two
choices;
14-8
-------
ViCCn, Xj) = a0 Cn + ax Xf (14,8a)
V2(C12, X.) = b0 C12 + bi X, (14.8b)
i i
v(Ci, X.) = a0 nxCn + bo (1-nx)C12 + at rtjX. + bx (l-n^X. (14.9a)
v(C2/ X.) = a0 C21 + 3l X. . (14.9b)
;he important distinction between the (14,7a) (14.7b) pair and the (14.9a),
14.9b) pair is the role assigned to the probability in the two formulations.
It is also important to recognize that if we are willing to assume that certain
Variables have the same effect on each of the state dependent utility functions
then the nature of the role of the probabilities as interaction variables changes
with the events at risk. This possibility offers another set of potentially in-
teresting testable hypotheses.
Clearly, these are simple examples. However, they do serve to illustrate
that explicit consideration of the source of the state dependency, the form of
the deterministic component of |j(.), as well as of the appropriate interaction
between elements in C. and X. (such as would be implied by homogeneity of
degree zero in income and prices), each offer the potential for introducing
testable restrictions on v(.). Indeed, without explicit consideration of some
of these restrictions it will not be possible to argue that we have established
a consistent theoretical basis for the valuation estimates derived from the esti-
mated deterministic component of the utility function.
Estimation of the random utility model with ranked data and an assumption
of independent VVeibuil distributions for the errors can be accomplished using
a maximum likelihood estimator. Beggs, Cardell, and Hausman [1381] derive
the likelihood function for this case as follows:
N T
L = 71 H
i=1 j=1
exp (v(C., X.))
"—J !
I exp(v(Cs» X,))
s=j
(14.10)
14-9
-------
where !
T = number of alternatives ranked
N = number of individuals.
After specifying a functional relationship for v(.) and taking the logarithm of
Equation (14.10), we have a globally concave, tog-likelihood function. This
function can be numerically optimized to derive the maximum likelihood esti-
mates, Our estimator used a modified Davidon, Fletcher, Powell [1963} (DFP)
algorithm with numerical derivatives.*
14.3 STRUCTURE OF THE CONTINGENT RANKING QUESTIONS
AND EXPERIMENTAL DESIGN
The objective of the design of the questions used in the contingent rank-
ing questionnaires was to maintain, as nearly as possible, complete consistency
with the information described in the contingent valuation component of the
survey. Consequently, the description of the problem as one associated with
a medium-sized company disposing of hazardous wastes (on site) in a landfill,
the explanation of exposure risk, and the identification of the conditional prob-
ability of death given exposure parallel those given for the contingent valuation
questionnaires. (See Chapter 11 for a more complete description of this materi-
al.) Respondents were given experience with the use of the risk circle cards
by asking them to consider the distance they felt that their home would have
to be moved from its present location to experience a specific reduction in the
risk of exposure.t This practice conforms to the procedure used with the con-
tingent valuation questions asking for respondent valuations. Thus, in terms
of the definition of the context of the task requested, the description of the
risks involved, and the ways in which individuals pay for risk reductions, the
contingent ranking and contingent valuation questionnaires were virtually iden-
tical ,
~The estimates were prepared using the GQOPT program developed by
Richard Quandt at the Econometrics Section at Princeton University. More
details on the program and convergence criteria used are available on request
from the authors. <
fSee Chapter 15 for a more detailed discussion of this question and the
results from it.
14-10
-------
As in the contingent valuation component of the survey, the contingent
ranking instruments elicit information on individuals' perceptions, attitudes,
nd valuations for different types of risks in two ways: (1) by asking several
ifferent questions to the same individual and (2) by asking different questions
different individuals. The first aspect of this process we have referred to
s relating to the structure of the questionnaire and the second to the experi-
ental design for the survey. With respect to the structure of the question-
naire received by respondents with the contingent ranking* format, nearly all
Other dimensions of the questionnaire and tasks requested are identical between
the contingent valuation and contingent ranking approaches. Thus, for exam-
ple, use of a contingent ranking format to elicit valuation information associated
with changes in the risk of exposure to hazardous wastes did not affect the
questions concerning the distances an individual would select from hazardous
facilities or the process used to investigate responses to job-related risks.
It did have effects on the questions designed to elicit the existence values
for risks to the ecosystem as our discussion in Chapter 11 indicated. The
reason is straightforward; the individual was assumed at the outset of this
question to have bid for a risk change for his household. Therefore, based
on the outcome of this process, he would have a postulated base exposure
probability. We wanted to ask for an additional value for additional motives.
Consequently, a contingent valuation question had to be introduced in the con-
tingent ranking questionnaires before the existence value question. Since this
was placed after the ranking tasks were completed it would not have influenced
the rankings derived. Of course, it is possible that the responses to these
contingent valuation questions were affected by the information provided in
e rankings. This is one of the reasons these valuation responses were
eated separately in Chapter 11.*
There were several other changes that were required by the use of the
nking format. The most important of these involves the logic of the task
*lt is possible that the suggested payments used in the contingent ranking
design points served as anchors for the valuation responses reported in these
intingent valuation questions. We have a rather limited basis for investigating
this issue because there are only two payment vectors. Nonetheless, it can
je considered as part of a more detailed evaluation of bids to realize a level
nth no exposure risk, since this was the way these questions were posed.
14-11
-------
requested of each respondent. With the contingent valuation questions each
individual was presented two risk cards and asked his valuation of the reduc-
tion in the exposure risk associated with the changes from one card to the
next. Consequently, the individual must compare the two cards to determine
the "commodity" (i.e., the risk change) that is to be valued. By contrast,
in the contingent ranking form, the individual was given four cards, each con-
tains a risk level and a payment level. The payment level is explained to be
required to realize the risk level. Figure 14-1 displays the four cards used
for one of the design points (R-1) comparing the contingent ranking component
of the survey.
A second difference in the ranking questionnaire arises from the approach
used to evaluate how each respondent would react to the cause of death stem-
ming from exposure to hazardous wastes. Recall that, in the contingent valua-
tion questionnaires, each respondent was asked if they wished to change the
reported valuation of the risk reduction when the event that could result from
exposure was assumed to be one of two possibilities. In the ranking question-
naires these same effects were retained, but the question posed involved a
revision to the earlier ranking with the proposed modification in health effect.
Since the experimental design used to alter the conditions presented to
respondents affects the wording used to describe the ranking task requested
of survey respondents, the features of this design must be reviewed first be-
fore proceeding to an overview, of the questions used to elicit individuals' pref-
erence ordering. Approximately 40 percent of the survey respondents were
given ranking questions.* Since the overall size of the sample for the ranking
component of the survey was smaller than that devoted to the contingent valu-
ation approach, the design selected was correspondingly simplified. It is a
full-factorial design investigating the effect of the vector of exposure probabil-
ities paired with a vector of proposed payments. Two variations in the vector
of exposure probabilities and two in the payment vectors are considered.
Thus, there were four different combinations of these two features of the con-
ditions that were to be ranked, figure 14-2 repeats one panel from Figure
*As we discussed in Chapter 9, all of the questionnaires were randomly
ordered so that the effects of the questionnaire and interviewer on responses
would not be confounded.
14-12
-------
HI
CatdAI
Payment required: #0 (W month
In higher prices and tauss
B-1
Card B-1
Paywsnt required; MO par montfi M240 par
ta higher pricvs and Im«i
4*
OJ
B-1
Card C-1
Paymant r«p*edr #§§ per month ($660 per yawl
to higher prices m '
B-1
Card 04
Payment required: $105 per month
I $1,260 per yew) In higher prices and taxes
H«k of Expotut*
RnkofOaMh
II EKpossd
CoiMndiH;
fciptisura and Death
Comfaintd flbk:
Possibte
Palhways
Pathwayi
Figure 14-1, Modified risk cards for contingent
ranking questons-design point R-1.
-------
Contingent Ranking Question Format {R)
Levels of risk
Amount of monthly payment
Risk of
exposure
Risk of death,
if exposed
Vector A
Vector B
Vector
$0 $20 $55 $150
$-20 $5 $40 $80
1/10
I
1/20
1/50
1/1U
R1
R2
1/100
1/20
II
1/30
1/60
1/10
R3
R4
1/150
Figure 14-2. Outline of the design for the contingent ranking component of the survey.
-------
7-4 in Chapter 7 defining the specific combinations associated with each design
point.
Two aspects of these selections were important motivations for the design.
The vector of exposure probabilities was selected to span approximately the
same range as was used in the full-factorial component of the contingent valu-
ation design (i.e., design points D1 through D6 in figure 7-4). The distinc-
tion in the exposure risk vectors arises from differences in the level of start-
ing and ending risks and differences in the rate of change over the four cases
ranked. Exposure vector I postulates a reduction by a factor of ten in the
exposure risk over the four cards ranked, while vector II has a reduction by
a factor of seven and a half over the four. The rate at which these changes
take place also varies between the two. In the first vector the rate is rela-
tively constant—a reduction of a factor of two from Card A to Card B, then
by two and a half from B to C, and then by two, again, from C to D, The
second vector has an increasing percentage reduction from a factor of one and
a half for A to B to a factor of two for B to C and then to a factor of two
and a half for C to D,
The second motivation concerns the implicit assignment of property rights
implied by the specification of the payment vectors. The first, payment vec-
tor, a, starts at an initial condition, comparable to the A risk card in designs
D1 and D2 for the contingent valuation component of the survey. That is,
the individual has the same exposure risk as in the initial state for designs
D'l and D2 and makes no payment to receive this condition. Regulations are
described as requiring more stringent containment technologies and thereby
progressively reducing the exposure risk.
The payments along the top of the figure are the monthly additional costs
associated with the new risk level. Thus, to calculate the implied payment
increment for the risk reduction from Card B to Card C the individual must
subtract the levels presented on each card. In the case of design points R1
and R3, for example, this increment would be $35.00 more per month for a
reduction from 1/20 to 1/50 in the case of R1 and 1/30 to 1/60 for R2.
To provide an alternative implied set of property rights within the choices
ranked, we introduced the possibility of a negative payment. In other words
it was suggested that a reduction in taxes or the overall price level was possi-
14-15
-------
ble as a result of a relaxation in the regulations governing the disposal of haz-
ardous wastes. This is the case described with payment vector b.
In the process of using drafts of the questionnaire in focus groups and
the pretest activities we found that individuals had difficulty in dealing with
the negative payment as the first card presented. Moreover, it was difficult
to control the reference positron starting from this possibility. Consequently,
we changed the ordering of the cards in presenting design points R2 and R4.
The individual was introduced to the ranking task using Card B and then pre-
sented with Card A as a reflection of the possibility of receiving reduced
taxes, but greater risk. Following this explanation of the possibility for in-
creasing risk and reducing current taxes, the remaining cards with higher
payments and lower risks were introduced. This process proved to be the
most convenient method to control the context of the scenario described to the
individuals and limited (in the pretest) the problems with respondents' under-
standing of the negative payments. However, as a result of this change fn
ordering the wording of the ranking task description is somewhat d;fferent
for each of the two pairs of design points. The full text for the question with
payment vector a (i.e., design points R1 and R3) is given as follows:
Interviewer hands respondent Cards B and C with dollar amounts.*
Now, think about these cards and about paying in higher prices and
taxes. As you can see on the cards,the risk of exposure decreases
from 1 chance in 20 on Card B, to 1 chance in 50 on Card C. The
decrease means your combined risk of exposure and death gets smal-
ler. The amounts you would pay in higher product prices and taxes
increase while the risk of exposure decreases.
Using a hypothetical situation, I'm going to ask you some questions
about paying for different levels of exposure risk for you (and your
household members).
This is the hypothetical situation. A medium-size company that pro-
duces electronic parts is located 3 miles from your home. This com-,
pany generates .2,000 gallons of hazardous wastes each day and dis-
poses of them, using established industry-wide pratices, in a landfill;
right at the plant site. If you're exposed to a large enough amount:
of these wastes for a long enough period, there's a chance you will
tThe dollar amounts and probabilities mentioned in the text change) with
the design point for the questionnaire.
14-16
-------
die in 30 years. Under these circumstances, and if you didn't pay
any more in higher product prices and taxes, your (and your house-
hold members') risk of exposure to these wastes would be at the
level on Card A. This is a risk you could potentially face for all
these years until the health effect is known.
Now, suppose the government added regulations requiring the com-
pany to install special liners that would seal the landfill and monitor-
ing systems that would detect leaks. These regulations would reduce
the chances that the landfill would leak and your risk of exposure
would be at the level on Card B. This would require a monthly pay-
ment of $20 in higher product prices and taxes.
Suppose the government added more regulations requiring the com-
pany to remove the most toxic materials from; the wastes before
they're put into the lined and monitored landfill. This regulation
would require a monthly payment of $55 in higher product prices
and taxes, and your risk of exposure would be at the level on Card
C .
Suppose additional regulations would require the company to use
more expensive ways to make its products. There would be a re-
duction in fomci of the most toxic wastes generated. These regula-
tions would r squire a monthly payment of $105 in higher product
prices and taxes, and your risk of exposure would be at the level
on Card D.
Look over the hypothetical situation on Card 7 [a card used by the
interviewer to remind the respondent of the elements in the problem.
It is given in figure 14-3] once more. Now, thinking about your
monthly income and what you spend it on in your budget, rank these
cards. Place on top of the pile the card with the payment and risk
combination you prefer the most and the card with the combination
you like least on the bottom.
A substantial portion of the text does not change for the design points (R2
and R4) involving negative payments. The portion which changes begins in
the fourth paragraph of the above discussion. The text for design points R2
and R4 is given as:
This is the hypothetical situation. A medium-size company that pro-
duces electronic parts is located 3 miles from your home. This com-
pany generates 2,000 gal Ions of hazardous waste each day,and dis-
poses of them, using established industry-wide practices, in a land-
fill right at the plant site. If you're exposed to a large enough
amount of these wastes for a long enough period, there's a chance
you will die in 30 years. Under these circumstances, your (and
your household members') risk of exposure is a risk you could
! potentially face for all these years until the health effect Is known.
14-17
-------
Exposure Risk Circumstances .
Electronic parts company
Located 3 miles from your home
Generates 2,000 gallons of hazardous waste
each day
Company disposes of the wastes in a landfill at
company site
If you are exposed, there is a chance you will
die in 30 years
Figure 14-3, Description of hypothetical situation.
-------
The government coufd introduce regulations which require the com-
pany to install special liners that will seal the landfill and monitoring
systems that will detect leaks. These regulations would reduce the
chances that the landfill could ieak and your (and your household
members') risk of exposure would be at the level on Card B. This
would require a monthly payment of $5 in higher product prices and
taxes.
If the government decides not to introduce regulations requiring
special liners and monitoring systems, this couid lead to a govern-
ment. cost savings, and the company would not raise its product
prices as it would do with the regulations, tf these regulations are
not added, taxes could be reduced $20 per month. The risk of
exposure for you (and your household members) would be at the
level on Card A.
Alternatively, the government could add more regulations than
described for Card B. These would require the company to remove
the most toxic materials from the wastes before they are put into
| the lined and monitored landfill. Your risk of exposure would be at
I the level on Card C, and these regulations would require a monthly
1 payment of $40 in higher product prices and taxes.
The balance of the explanation was not changed from that used with R1 and
R3.
One final aspect of the contingent ranking design should be noted. It
was not possible to consider the effects of changes in the conditional probabil-
ity of death given exposure to hazardous wastes, so this probabifity was held
constant at one-tenth.
j
14.4 EMPIRICAL FINDINGS
As we indicated in Chapter 9 the actual performance of the ranking ques-
tionnaires compares quite favorably with what was expected based on the sam-
pling plan and experimental design. A total of 227 complete interviews with
complete rankings were obtained, approximately 37 percent of the sampie of
completed interviews (including both contingent valuation and contingent rank-
ing).
Our preliminary analysis of these data has focused on three tasks; (1)
examining the responses to develop some insights as to the types of rankings
provided and what these patterns might indicate, at a rather general level,
about individual preferences; (2) testing the effects of the design variables
on the rankings derived under the assumption that respondents can be treated
as homogeneous; and (3) preliminary estimates of the random utility functions
14-19
-------
using the ranked logit framework and a largely unrestricted specification for
the utility function assumed to describe the determinants of an ordering of
exposure risk-payment combinations. We have not, at this stage of the re-
search, attempted to analyze the revisions in the initial rankings provided in
response to the proposed health effects resulting from exposure. In what fol-
lows we will summarize each dimension of the results of the analyses conducted
with these initial rankings.
14,4.1 An Overview of the Nature of the Rankings Provided
Table 14-1 provides a summary of the frequency each of the cards de-
scribing a risk-payment combination was ranked first. In order to interpret
the table some background on the labeling conventions used in this table and
in the others which follow is necessary. The card Iabels--A, B, C, D—corres-
pond to the paring of similarly positioned elements in the payment and expo-
sure risk vectors. For example, referring back to Figure 14-2, Card A in
design R1 would involve a pairing of the first payment, 0, with the first expo-
sure risk 1/10. Card B relates to the second elements; C to the third and D
to the fourth. This convention is maintained for al! four design points..
Equally important, 'we have used the term "Version" as synonymous with design
point. Thus Version 1 corresponds to the set of respondents receiving the
questionnaires associated with design point, R1.
Table 14-1 provides some interesting general information. First, in the
cases where negative payments are used (Versions 2 and 4) the lower initial
exposure risk associated with Version 4 does not increase its frequency of be-
ing selected as the first choice. Indeed, the cases involving the negative pay-
ments have the lowest frequency of selection as the most preferred alternative.
If we assume individuals are homogeneous with no differences in constraints
affecting their behavior (whether actual or in the context of a hypothetical
situation such as posed here), then we would label the higher rates of ranking
Card A first with Version 1 in comparison to 2 or Version 3 in comparison to
4 as irrational. In the case of 1 (3) they are receiving the same exposure
risk as 2 (4), but the latter costs less. Indeed, it leads to a reduction in
expenses and this may well be the source of the problem. Respondents simply
may not believe this case would happen. As a consequence, they may well be-
14-20
-------
TABLE 14-1. FREQUENCY OF CARD CHOSEN FIRST BY VERSION
Version
Card chosen
first3'b
Total0
A
B
C
O
; 1
13
13
23
8
57
(5.73)
(5.73)
(10.13)
(3.52)
(25.11)
' 2
4
21
18
16
59
.
(1.78)
(9-25)
(7.93)
(7.05)
(25.99)
' 3
13
15
15
7
56
;
(8.37)
(6.81)
(6.61)
(3.08)
(24.67)
\4
5
16
16
18
58
(2,.20)
(7.05)
(7.05)
(7.33)
(24,23.)
Totala
41
65
72
49
227
;
(18,08)
(28.63)
(31.92)
(21.59)
(100.00)
aParentheses denote percentage of overall total.
includes only those respondents who had ranked all four cards.
c l<
Row and column percentages may not add to 100 due to rounding,
14-21
-------
tieve selection of this case would involve realizing the higher exposure risk
but no corresponding reduction in taxes.
There are also examples consistent with a priori expectations in the table.
For example, compare the first place rankings given to Card D with Versions
1 and 2 as welt as 3 and 4. D in Version 1 yields the same risk level but
higher payments than D in Version 2. Thus, we would expect with homogene-
ous individuals, D to be ranked first more frequently in 2 and 4 in comparison
to 1 and 3, respectively. This is precisely the outcome .realized. Another
confirmation can be found with Card B in Versions 1 and 2. However, the
remainder of the cases would appear to contradict a priori expectations based
on simple assumptions, although none is as glaring an example as we observed
for the case of Card A.
In Table 14-2, we present the frequency of all the potential types of
rankings of all cards observed in our sample. Of the 24 possibilities, only 15
were observed. The frequencies for each by version are displayed in this
table. Clearly, version does seem to affect the frequency with which these
rankings are observed. Perhaps the most dramatic difference arises with the
change in the frequency observed for the ranking ABCD and DCBA with de-
signs involving negative payments versus those that do not,
14.4,2 Some Simple Tests for the Effects of Exposure and Payment Vectors
The first step in our more systematic analysis of these rankings is analo-
gous to the tests involving samples means with the contingent valuation re-
sponses in Chapter 11. That is, we treated individuals as homogeneous so
that the only factors that might affect their respective rankings of the combin-
ations of payments and exposure vectors would be a change in either of these
variables. Tables 14-3 and 14-4 present the results of chi square tests for
hypotheses based on this simple view of the role of individual characteristics
on the rankings.* In the first of these tables we report the results of four
*The chi -square statistic is defined as follows:
(Ot. - E.. )2
i 1
JJL
f
I « Eij
where
O.. = observed frequency in the ijth cell
E.j = expected frequency in the ijth cell.
14-22
-------
TABLE 14-2. RANKING PERMUTATIONS CHOSEN, BY VERSION
Version^
Rank
permutation3
R1
R2
R3
R4
Total
Frequency
Percent
Frequency
Percent
Frequency
Percent
Frequency
Percent
Frequency
Percent
ABC 0
13
22.81
4
§.78
17
30.36
4 *
_
38
18.74
ACBD
0
0.00
0
0.00
1
1.79
1
i
id
2
0.88
ADBC
0
0.00'
0
0.00
1
1.79
0
0
00
1
0.44
BACD
4
7,02
13
22.03
4
7.14
9
16-
36
30
13.22
BADC
1
1.75
0
0.00
0
0.00
0
0.
00
1
0.44
BCAD
0
0.00
3
5.08
5
8.93
2
3.
84
10
4.41
BCDA
8
M. 04
5
a. 48
6
10.71
5
9.
09
24
10.57
CABD
1
1.75
2
3.39
1
1.79
3
5.
4S
7
3.08
CBAD
5
8.7?
3
5.08
4
7,14
2
3.
83
14
S.17
CBDA
7
12.28
7
11.68
7
12,50
1
1.
82
22
9.83
CDAB
0
0.00
1
1.69
0
0.00
1
1.
82
2
0.88
CDBA
10
17.54
5
8.48
3
5.36
9
16.
36
27
11.89
DBC A
0
0.00
1
1.69
0
0.00
0
0.
00
1
0,44
DC AB
0
0.00
3
5.08
0
0.00
7
12.
72
10
4.41
DCBA
8
14.04
12
20.34
7
12.50
11
20.
00
3a
16.74
T otal
57
100.00
59
100.00
56
100.00
55
100.
00
227
100.00
aor ihe 24 possible permutations, only 15 ware actually chosen.
^Percentages may not add to 100 due to rounding.
-------
TABLE 14-3. TESTS CONCERNING THE 1NDEPENDE NCh OF VERSION ADMINISTERED
AND CARD CHOSEN FIRS I
Null hypothesis
X2
statistic
Degrees
of
freedom
Critical
value of
X2
at the
0.05 level
Reject
the null
hypothesis
at a
0,05 level
Version administered is independent of card
chosen first (across all versions and first
choices)
25.82
9
16.32
Yes
Payment vector given an exposure vector;
Exposure vector 1/10, 1/20, 1/50, 1/100
9.88
3
7.81
Yes
Exposure vector 1/20, 1/50, 1/60, 1/150
3.65
3
7.81
Yes
Exposure vector given a payment vector;
Payment vector 0, 20, 55, 105
2.90
3
7.81
No
Payment vector -20, 5, 40, 80
0.89
3
7.81
No
-------
TftBCE W-C TESTS CONCERNING THE UNDERLYING DISTRIBUTION OF CARD
CHOSEN FIRST, BY VERSION
Null hypothesis
X2
statistic
Degrees
of
freedom
Critical
value of
X2
at the
0.05 level
Reject
the null
hypothesis
at a
0.05 level
Card chosen first follows a uniform probability
density function for;
Version 1
10.82
3
7.81
Yes
Version 2
10.61
3
7.81
Yes
Version 3
5.20
3
7.81
No
Version 4
7.62
3
7,81
No
-------
tests. First, we consider whether the combination of payment and exposure
risk (or risk-payment card) ranked first is independent of the version consid-
ered, Clearly, the results call for a rejection of this null hypothesis. The next
four tests refine this analysis to consider the attributes of a design point that
would be expected to influence the selection of a risk-payment card firsts. The
first two tests consider the effects of the payment vector holding exposure risk
constant, and the last two reverse the process. As the decisions suggest, the
payment vector appears to have an effect on the card ranked first for each
exposure risk vector, while reversing their roles and reanalyzing the results
suggests that we cannot reject the null hypothesis of independence for the card
ranked first and the exposure risk vector.
Table 14-4 uses the chi-square test as a goodness of fit test. In this case
we maintain that each risk-payment card is equally likely to be ranked first.
Thus we would expect the frequencies to be uniformly distributed across the
four cards. The first four rows of this table compare the observed frequencies
of each card being ranked first with what would be implied by the uniform
assumption by version. In Versions 1 and 2 we reject the null hypothesis of a
uniform distribution, while in Versions 3 arid 4 we do not at the 0,05 level. Of
course, it should be acknowledged that this is a close decision for the case of
Version 4.
14.4.3 Preliminary Estimates of the Random Utility Models
The estimation of models in terms of the attributes of the choices being
ranked (i.e., the exposure risk-payment combinations) and the socioeconomic
characteristics of respondents can be interpreted as a direct extension to the
simple chi-square tests for independence of one component of the ranking (i.e.,
cards designated first and the attributes of the combinations ranked) reported
earlier. In those cases we effectively assumed that the features of each re-
spondent did not effect the conditional probability of selecting one alternative as
first and tested whether the payment and exposure vectors did. Now we
include additional information in the form of the specific assumptions for the
functional form for the random utility models that were hypothesized at the
outset of this chapter as providing the basis for individuals' rankings. These
models include variables describing the features of individuals and the attri-
butes of what they have been asked to rank. I
14-26
-------
Given the pronounced differences in the rankings by version we have
developed these models in stages. The first stage results are based on a stra-
tegy that estimates a separate model using the responses to each version (i.e.,
design point). The specification of these models is maintained constant across
all of the subsets of the sample and includes the payment and exposure risk
measures, as well as the individual's age, sex, household income, years of
education, and years of residence in the town. These variables related to
each respondent are treated in two different ways. In one-set of models, re-
ported in Table 14-5, they are each interacted with the payment measure and
in the second set, reported in Table 14-6, they are each interacted with the
exposure risk measure.
The results from these simple models are remarkably good. The payment
and exposure risks are usually statistically significant determinants of the in-
dex. utility relevant to ordering risk-payment alternatives.* Both have a nega-
tive effect on the utility index as we would have expected a priori. The only
exception to this sign pattern for the estimated parameters occurs in the case
of the estimates based on the Version 2 sub-sample with the model involving
individual characteristics interacted with exposure, and this parameter would
not be judged to be significantly different from zero.
Several of the remaining variables would also be judged as statistically
significant determinants of the preference indexes in some models. Moreover,
there does appear to be a pattern across the versions and models considered
in this simple case. Income is generally a positive influence when its parame-
ter would be judged to be statistically significant in models involving interac-
tions with payment, while the reverse sign arose for the significant parameter
estimates in models with exposure interactions. Since these interaction varia-
bles modify the effect attributed to the attributes of the payment-risk cards
being ranked, this would seem a plausible result. It suggests in the first
case that, ceteris paribus, higher income leads payment increases to have a
smaller negative effect on ranking a combination first. It is important to
*These results must be cautiously interpreted because the tests are based
on the asymptotic distributions for each test statistic. Nonetheless, our judg-
ments can usually be based on generous margins over the conventional critical
values used in testing the null hypotheses for parameter estimates.
!
:
14-27
:
-------
TABLE 14-5. BASIC MODEL FOR THE RANDOM UTILITY MODEL WITH THE BANKED LOG1T ESTIMATOR,
BY VERSION — INTERACTION WITH PAYMENT MEASURE
Versions
_3L
R2
R3
R4
Alternative Interactive native Alternative Alter rmlv.
Independent variables specific Payment Exposui cific Payment Exposure specific Payment exposure specific Payment Exposure
Interactive
I nteractive
interactive
Alternative specific
Payment
b
Exposure
Individual specific
Age (X)
S«xC
-------
BY VERSION •"•INTERACTION WITH EXPOSURE RISK MEASURES
Vf»lons*
SI
JSL.
R3
m
Interactive
Interact! ve
Alternative — """*11"' Alternative —— Alternative Interactive Alternative
Independent variables specific Payment exposure specific Payment Exposure specific Payment Exposure specific
Interactive
Payment Exposure
Alternative sp«Ctfie
Payment
£xpoiureb
Individual tpeclfic
Age {X}
5«C (X)
Income*1 (X)
education (X)
Years in lawn (X }
Initial likelihood value
Number of iterations
Final likelihood value
Number of obsei vatioris
-0.03
(-5,64)
-0.0?
(-0.78)
-m.j
i?
-118.2
49
6.9*10
{2-11)
2.2x10'
(2.84)
-5.2*10
(-2.71)
-2-SkIO
(-1-55)
•5.V10
(-2.40)
-0.04
(-S.5Z)
0.01
(0.26)
-2
-4
-3
-4
-152.5
15
-124,9
48
-4,7*10
(-0.01)
-8.1*10
("1 M)
4.2*10
(0.2?)
-3.9»10
(-2.91)
3.8*1#""
(O.Stt)
-0.06
(-6,10)
-0.20
(-4.313
6
-3
-5
-3
-1*9.6
16
-116-9
47
2.3x10
(3.89)
-3
-2.6x10
(-1 -87)
-2
-1.5*10
(-2.57)
-3
1. 7*10
(0.65)
-3
-9.3*10
(-1 JO)
-4
-o.os
(-4.19)
-0.16
(-2.53)
-155.72
14
-136.19
m
Numbtri in pifwilheses. denote the ratio of the Ml estimate of llw coefficient to th* asymptotic standard deviation,
btxp€Hsur« risk is scaled by 1,000 in these estimates ( Exposure = actual probability * 1,000).
c Denotes a binary variable equal to one if male, zero otherwise.
{J
Household income In thousand* of dollars.
2 6*10
(0 36)
3.7*10
(2 63)
-4
-2
-6.0*10
(-186)
• 4
9,2*10
(0 29)
3.4*10
(0.03)
-5
-------
recognize that this interpretation not only holds other characteristics constant
but the level of exposure risk. In the models where income was interacted
with exposure risk, the parameters suggest higher income makes individuals
less tolerant of exposure risk increases (for a given payment level}. Hence
the modeis would predict that these individuals would be willing to pay more
for risk reductions in these cases.
These same sign reversals arise for all the other variables included in
the basic model, when their estimated parameters would be judged as statistic-
ally significant. Age has a negative effect when interacted with payment and
positive when interacted with exposure. Sex has the same pattern, while edu-
cation and years in town resembles income with a positive effect when inter-
acted with payment and negative with exposure risk.
Table 14-7 reports the estimates of these basic models (I.e., the same
independent variables with the interactions with payment and exposure risk).
The results parallel those obtained with the model applied to the subsets of
the sample by design point. The payment and exposure risk measures have
parameter estimates that would be judged to be statistically significant. Both
have negative influences on the utility index. Income and age are aiso statis-
tically significant determinants, with the sign pattern for the estimated parame-
ters agreeing in each case with what was found with the models estimated for
each version, provided we considered only the parameters judged to be statis-
tically significant at approximately the 5 percent level.
Thus, on the basis of these findings, it seems clear that respondents'
characteristics do matter. The nature of their effects on the utility index may
help to explain some of the contradictory results found by examining the cards
chosen first in isolation. That is, the collection of all ranks taken together
do appear to be "explained" reasonably well by a very simple specification for
the random utility function.
Of course, one might reasonably ask just how good is this explanation?
One informal means of gauging this conclusion is to consider the within sample
performance of the model in predicting the exposure risk-payment combinations
that were selected by each respondent as their first choice. To illustrate what
is involved, consider an example. Suppose we used each specification of the
basic model (i.e., that using interactions with payment and with exposure risk
14-30
-------
TABLE 14-7, BASIC MODEL FOR THE RANDOM UTILITY MODEL WITH THE
RANKED LOG IT ESTIMATOR, USING THE FULL SAMPLE
independent
I variables
Alternative
specific
Model 1
interaction
with individual
specific variables Alwrn,tjv.
Payment Exposure specific
Model 2
I nteraction
with individual
specific variables
Payment Exposure
Alternative specific
Payment
Exposure
Individual specific
Ag® (x)
Sexb (x)
c , ,
ncome (x)
Education
-------
and the full sample) to predict the average survey respondents' conditional
probability of selecting each of two of the exposure risk-payment cards first.
Table 14-8 reports these estimated probabilities for Cards A and D with each
version for each of the two specifications of the basic model. These differ-
ences in estimated conditional probabilities illustrate the effect of design point
and model on predictions for a single hypothetical individual. To gauge the
effect for actual individuals we repeated the process using the respondents in
our sample. Below the estimated equations in Table 14-7 we-report the results
of this exercise. It was undertaken for subsets of the sample corresponding
to each design point (or version) using the random utility functions estimated
from the full sample. These chi-square statistics test the null hypothesis of
conformity between observed and expected frequencies (based on treating the
first ranked card as the one with the largest of the conditional probabilities
predicted from the estimated random utility models) for each card as the first
choice of the respondents receiving each of the four versions.
Only in the case of Version 4 (the negative payment case with the lowest
exposure risks), would the model be judged to be inconsistent with practice.
Of course, this is simply an index of performance, not a test of the model.
The predicted probabilities and corresponding designations for the first cards
are a within-sample prediction. * Nonetheless, it does pinpoint the same com-
bination that was a part of the problems observed in our general interpretation
of the rankings.
Refinement of these models can proceed in a variety of directions. Since
our analysis under this phase of the research was intended to be preliminary
we have sought to consider model refinements for only three reasons;
1. Does refinement to these models, either through inclusion of
other variables describing an individual's attitudes toward risk
(or hazardous wastes), information, health, or family status
change the overall conclusions based on these basic models?
Moreover, would any of these models be judged unambiguous!>
superior to the basic model?
¦"It converts the predictions into categorical exact variables and will be
sensitive to the procedure used to designate each outcome. Nonetheless, it is
one simple way to gauge the conformity of the model with subsets of the
sample.
14-32
-------
j TABLE 14-8, ESTIMATED
CONDITIONAL
PROBABILITY OF A
! PAYMENT-EXPOSURE RISK
COMBINATION
RANKING AS FIRST
Payment-
Basic model3
i exposure risk
Payment
Exposure risk
Version combination
irte -action
interaction
1 A
0.054
0.130
D
0,240
0.145
2 A
0.057
0.139
D
0.277
0.178
3 A
0.227
0.325
1 D
0.155
0.090
! 4 A
0.237
0.344
D
0.179
0.107
'The basic model refers to the random utility model estimated using the fulf
sample as given in Table 14-7.
'The A and D correspond to the risk cards used to describe the payment-
exposure risk combination. A specifies the combination of the first elements
in each vector and the combination of the last elements.
14-33
-------
2. Does the inclusion of these additional variables improve the con-
sistency of the estimated models for each subset (I.e., accord-
ing to version)? _
3- Do the results for a selected set of refinements in the economic
variables, changing their respective roles to provide somewhat
more consistency with what would be expected from economic
theory improve the results? And, therefore, do these modifica-
tions help to identify directions for further research? ;
Tables 14-9 through 14-13 begin the process of addressing these ques-
tions, Rather than review the specific results in detail, we will attempt to
summarize the general responses to these three questions that seem to be im-
plied by the findings to date.
First, while several of the knowledge, attitude, and risk preference meas-
ures would be judged to be significant determinants in some models for partic-
ular subsets of the sample, overall patterns are difficult to discern. The in-
clusion of these variables does not appear to affect the importance and sign
patterns observed for the variables in the basic models presented earlier, In
drawing this conclusion, it is important to note that the income and payment
variables have been entered in a different format in these models. That is,
-1
the term income is followed by (t) , This is intended to suggest that the
ratio of payment to income has entered the model.
Finally the results do not provide clear patterns as to how to proceed in
model development. It does appear that the theoretical arguments sketched
outlined may in this chapter help to improve upon the task associated with
model selection. That is, it should be possible to use state dependent specifi-
cations for the 'deterministic component of the utility model to develop a set of
interaction variables that would follow from theory. Moreover, by considering
the variables that might be expected to lead to state dependency in prefer-
ences, it should be possible to reduce the set of potential models and to formu-
late specific expectations for the variables that are included in them.
14.5 COMPARING CONTINGENT VALUATION AND CONTINGENT
RANKING ESTIMATES OF THE VALUE OF RISK REDUCTIONS
The objective of this section is to report a preliminary comparison of a
set of valuation estimates implied by the estimated utility functions derived
14-34
-------
I ABU 14-9. SELECTiD HtMlt-15 FOR I HE RANDOM UMLIIV MOOfi WITH
THE RANKtLl IOC.I1 1ST IMATOR BY VEKi>lON
I rid ep en (ten 1
variables
Alternative
specific
R1
InlWMCDdft
wild Individual
stmcttic variables
Payment Exposure
Versions
M.
HI
Alternative
specific
!ru»raccffff>
with indivhiu.it
specific variables
wilh individual
AlwrrwllV. *»"»*"«» MtemMvm
Payment t*poiur« specific Paymenl Exposure specific
_ R<
interaction
with individual
specific variables
Payment Exposure
OJ
cjn
-0.10
(-3.67)
-0.04
(-4.85)
Allirrutiva >p«:nl>t
Payment
E xpoture
Individual specific^
Age < x )
SexC (x>
£* *1
I rtcome (?)
Education (x)
Risk8 , zero olhrrtvtse
-------
TABLE 14-10. SELECTED RESULTS FOR THE RANDOM UTILITY MODEL WITH
THE RANKED IOGIT ESTIMATOR BY VERSION
Versions
Independent
variables
Alternative
specific
Bl—
Interaction
with individual
specific variables
Payment Exposure
HZ
Alternative
specific
interaction
with Individual
specific variables
Payment exposure
Alternative
specific
R3
interaction
with individual
specific variables
Payment Exposure
Alternative
specific
M
interaction
with individual
specific variables
Payment Exposure
Alternative specific
Pa yment
Exposure
individual specific*3
Age (x)
S«xc (x)
Income11 (i) 1
Education (x)
Risk® (x)
Number of children
less than 18 (x)
Home ownership1" (x)
-0.10
(-3.S7)
0 04
(-4.99)
-2.9x10
(-1.10)
-1.9x10
(-2.32}
1.1x10
(1.53)
4.7*10
(2.92)
5.2*10
(116)
2.4*10
C3.2S)
•1.5*10
(-1.72)
-2
-0 06
(-247)
-0.05
(-5 81)
-2,97*10
(-131)
6.2X10
(1 02)
-5.4*10
{-123),
2.8*10
(2.41)
-1.0x10
(-2 86)
-2.9*10
(-0.41)
-3
-2
-3
-2
3
4.7*10
(0.72)
-3
-0 01
(-0,45)
-0,13
(-7.22)
5 7*10
(2 80)
4.0x10
(0.76)
-2 7x10
(-6.66)
-4
-3
-1
•8.6*10
(-8 *9)
-4
2.7*10
CO.09)
-2.2*10
(0.34)
1.4*10
(11.06)
-2
-0.01
(0.30)
-0.14
(-4 87)
-6 6*10
(-0.03)
-1.8*10
(•2.86)
-2.4*10
(-2.49)
-1.9*10'
(-1.30)
*6
-2
-1
-J
-1.8*10
(-0.09)
S.2«10
(0.75)
5.1x10
(0.37)
,"3
-3
Summary statistics
Initial likelihood value
Number of iterations
Final likelihood value
Number of observations
-152.55
13
-110.54
48
•155.72
12
-125.83
49
-146.19
13
-113 35
46
-162.08
12
-137 95
51
Parentheses below the parameter estimate* denote asymptotic t-ratios for the null hypothesis of no association.
b Parentheses to I he right of each individual specific variable denote the form of the interaction with payment or exposure,
cPenates a binary variable ecjual to one if the respondent was a male, zero otherwise,
d
Household Income in thousands of dollars.
iey^-u«Bl«b4«-equ»l-te-Me-M:-MM-hiMlfvMhiel-»Urted~'»~fwef«f,«nc«-fiir-ri3tty -stto*rtton»7-zero-
-------
TABLE 14-11. SELECTED RESULTS FOR THE RANDOM UTILITY MOOit WITH
THE RANKED LOGIT ESTIMATOR BY VERSION
Independent
variables
Alternative
specific
R1
Interaction
with individual
sp«ific van.,bl»_ A|((rna|iv.
Payment Exposure specific
Ri?
R3
"Tftwrsetttwv ""
with individual
Alternative -»ClfiC V""b^
Payment Exposure specific Payment Exposure
Interaction
with individual
specific variables
Alternative
specific
m
Interact.or.
with individual
specific variables
Payment Exposure
03
-J
Alternative tpecific
Payment
Exposure
_i£5cifKb
Age (*)
-0.07
(-2.96)
-0.04
<-3.87)
(x>
a
Sex
Income^ { 4 )
Education tx)
Years lived In town (x)
Concerned about
hazardous waste6 Ik)
Hazardous waste pollu-
tion is harmful Cx)
Bisk of dying from
hazardous waste9 (x)
Summary statistics
Initial likelihood value -152.55
Number of iteration* 15
Final likelihood value -112.64
Number of observations 48
-2.9"10
(-1 18)
-2.2*10"
(-3.36)
-4.9*10"
(-0.6?)
1.8*1#"
(2.49)
s.a*io"
(2.80)
-4
-3.4x10
(-0.33)
-I.5*10
(1 69)
-3
-2
-9.1*10
(-0.22)
-3
-0 07
(-3.03)
-0.04
(-4.04)
-155.72
13
-128.14
~9
-S
,*3
».?X10
(0.33)
6.1*10
(1 02)
~2.Z*10"2
(-0.52)
2,5*10"3
(2.13)
-8.2*«"4
(-1.SB)
-7.0x10
("0 32)
-7 5*10
(-0.89)
7.7*10'
0.«>
-3
-3
-0.01
("0 59)
-0.09
(*3.69)
-146.19
IB
-107 17
46
-9.2*10
(-4.28)
5.0x10'
(0.89)
-2.5x10"
(-5.97)
-3.5*10"
(-0.35)
4.8*10*
(1.92)
-3
-4.2*15
(-239)
-3.5*10"
{-0,69)
-2.3x10
(-1.63)
-2
-?
-0.02
(-0.61)
-0.11
(-3.47)
• 162 08
IS
-130,1?
51
2.7*10
(1.03)
-2 1x10
(-3.31)
-2,7*10*
(-2.8!)
-1.4*10
(-0.97)
-6.5*10
(-0,16)
-4.2*10
(-0.02)
-3.4X10"
(-2.24)
-2.5x10*
(-5.96)
*P»re*Uheie» below the parameioi ••mimales denote asymptotic l-ralios for the null hypothec* of no association.
bf»arenths*es la the right of each Individual specific, variable denote the form of the interaction with payment or exposure.
Denotes a binary variable equal to one if the respondent was a male, iero otherwise.
^Household Income in thousands ol dollar*
^Denotes a binary variable equal to one if She respondent stated a concern about hazardous waste, zero otherwise. This variable is based on responses to
question K4.
Oenotes a binary variable equal lo one if the respondent considered (wardens waste pollution as harmful.
^Respondents evaluation of Hie risk of dying from hazardous waste given {he ris.k ladder.
-------
TABLE 14-12. SELECTED RESULTS FOR THE KAMOOM UTILITY' MODEL WITH
THE RANKED LOGIT ESTIMATOR BY VERSION
HI
HZ
S3
R zero.
-1.5*10
(-0.51)
-2,6x»0*
(-3,79)
-7.7*10"
(-2.64)
6.1x10
CI.91)
-2.6*10"
(-0.26)
-4.1x10
{'-2. S3)
-3.SM0
£-2.63)
-5.2*10"
(-0.86)
-3
-2
-2
T his variable is based on
-------
TABLE 14-13. SELECTED RESULTS FOR THE RANDOM UTILITY MODEL WITH
THE RANKED LGGiT ESTIMATOR, FULL SAMPLE
Modet-ipecific sample
1
Model 2
independent
variables
Alternati ve
speci fic
I nteraction
with individual
specific variables
Payment
Exposure
Alternative
specific
I nteraction
with individual
specific variables
Payment
Exposure
Alternative specific
payment
Exposure
Individual specific
Age
-0.02
(-1.66)
-0.04
(-9.66)
-0.01
(-1.56)
-0.03
(-5.02)
Sex
Income (+)
Education (x)
<3iskd (x)
Years in town (x)
Reason for rank*
-1.3x10"
i-Z.74)
2. 3* TO"
<0.42)
Se
e footnotes at end of table.
(continued)
14-39
-------
TABLE 14-13 {continued)
I nd«pend«nt
variables
Alternative
trJfie
Model-specific sample
jsaaau.
interaction
with individual
¦specific v>ri«bi»f
Payment
Exposure
Alternative
specific
Model 2
I nt»r*cticr
with ojai
specific a serious probium, i4rf
otherwise j
hD* notes a binary variable equal to ont if the respondent claimed there wm 3 relatively large risk of ay ing rtigm
hazardous waste pollution, zero otherwise.
'This statistic «siimat«» th-ts significance of the div«rg«nc# b*tw««n actual una predicts c*ra chos«n first «i|h
smaller x2 values denoting a small divergence. Chi-squsre v»iu#s presented by version have 3 degrees of
freedom »no 4 critical value of 6.25 at the 10-percent level of iiflnificance.
14-40
-------
from the contingent ranking models in comparison with the contingent valuation
resutis. It should be acknowledged that since we have not imposed specific
theoretical restrictions on the estimated utility functions, we do not have an
explicit theoretical interpretation for the valuation estimates that are derived.
This problem was first identified in Desvousges, Smith, and McGivney [1383],
It arises because theory would suggest restrictions on the nature of the rela-
tionship between income and the proposed payment for an exposure risk level,
Of course, the exact nature of these restrictions depends on how these pay-
ments are defined. Without these restrictions, different valuation estimates can
be derived depending on whether they are defined as changes in the payments
to maintain constant utility in the presence of a change in exposure risk, or
changes in income to offset the change in exposure risk.
For the present purposes, we will confine our attention to calculating the
changes in the proposed payments that would be required to maintain a constant
level for the utility index when the exposure risk changes. With the simple
models, the payment changes can be defined for each of the two specifications of
the basic model. Consider, for example, the specification involving interactions
between the payment and the independent variables describing the
characteristics of the individual. A general statement of this model is given in
Equation (14.11);
M
v = a0P + ajR + P • I b.X. , (14.11)
j=1 ' 1
where
P = payment
R = exposure risk
X. = jth characteristic of the individual.
Totaily differentiating Equation (14.11) with respect to the arguments that are
assumed to change, we have Equation (14.12);
M
dv = a0dP + ajdR HP I b.X, . (14.12)
j=1 1 J
Holding total utility constant implies the dv=0. Thus, solving Equation (14.12)
for dP, we have:
14-41
-------
dP = ' (|l 4,13)
(a0 + I b.X.)
ink e
J=1 ' J"
The same basic logic can be applied to the case of a model where the interac-
tions are with the exposure risk variable, as defined in Equation (14.14):
M
V = C0P + C! R + R I d.X. . (
1=1 1 J
14,14}
Repeating the same process, the payment increment, dP, in this case would
be given as follows;
M
-(Ci + I d.X.JdR
i;s1 J J
dP = I * (14.15)
co
To implement this approach to estimating the valuation of a risk change,
we must select one or more of the estimated random utility models. Ms we
acknowledged earlier, our refined models add little to the simple basic model
we used to start the process of trying to interpret the role of respondent
characteristics in explaining the observed rankings. Nonetheless, this conclu-
sion is at this stage purely a judgment based on rather casual comparison of
the estimates. Moreover, it is also clear that we can improve on the infor-
mation used in forming this judgment. For example, we can consider using
adaptations to the Hausman [1978] specification error test (see also Hausman
and McFadden [1384]} to evaluate our models. These are direct avenues for
future research in improving our preliminary estimates. However, it should
be acknowledged that these tests will not provide information on the overall
importance of the selection of a model for the valuation estimates. That is, if
our objective in estimating these models is exclusively one of estimating the
representative individual's valuation of a reduction in the risk of exposure to
hazardous waste, and if Equations (14.13) and (14.15) are accepted as the
appropriate valuation measures, then our interest lies in the sensitivity of
these results to the model used. It may well be, for example, that we cannot
14-42
-------
judge a particular model as "best", but that the selection from a set of close
competitors does not matter for the implied benefit estimates.
1 We can provide a simple illustration of the problem by comparing the im-
plied valuation estimates for all versions of the basic mode) in some specific
scenarios. For example, consider the task of predicting the valuation of the
risk changes posed in the contingent valuation questions by the respondents
receiving the ranking questionnaire. This is one potential valuation scenario.
Table 14-14 reports the average valuations derived from these calculations by
town for each of the three exposure risk reductions within the range of proba-
bilities posed in the contingent ranking questionnaires. All ten of the possible
versions of the basic model have been used (i.e., 5 based on sample * 2 based
on interaction variables used).
While this is a detailed table, a few general patterns do emerge. Models
estimated from subsets of the sample routinely have more negative predictions
for what are improvements in the respondents' circumstances. There are also
a >/ery wide range of estimates for these payment increments over the various
models and towns. Of the set of models reported, it appears that the basic
model estimated using the full sample offers the most consistent results. There
are no negative predictions and there is also a more modest set of estimates
for the valuation increments. Based on these findings alone, it seems clear
that the selection of a model for valuation estimates from this set can make a
substantial difference in the results derived. Moreover, even on the basis of
a rather limited set of models and the maintenance of parallel specifications in
models used, the problem of devising a set of criteria for refining model selec-
tion does appear quite relevant to this application.
Since developing a proposed resolution to this issue was beyond the scope
of the Phase I research activities, we have developed two illustrative compari-
sons of the contingent valuation and contingent ranking valuation estimates.
The first of these relies on mean valuation responses from the contingent valu-
ation component of the sample and the mean estimated valuation response using
each of the estimated utility models with the respondents to the contingent
ranking questionnaires. Table 14-15 reports these results for each of the
three exposure risk changes falling within the range of risks used in the rank-
ings.
14-43
-------
TABLE 14-14. AVERAGE VALUATION OF EXPOSURE BISK CHANGE BY TOWN8
'.'JO t-j
- r
W0101/20
Version 1
Version 2
Version 3
Vtnioi) 4
Version 5
tfumlon 1
Version 2
V#rato«3
Ifcralon «
Vinl
on#
Town
Income
P
e
P
E
P
E
P
E
P
E
P
E
P
E
P
E
P
E
P
E
Ab.T!g(ofi
2SB
X4
262
m
115
358
NV
NV
51 I
327
288
793
785
NV
344
1073
NV
NV
1546
982
863
Acton
456
61.8
286
MV
208
431
NV
NV
481
482
355
1MB
854
NV
625
1292
NV
NV
144,2
144.8
1064
Arlington
300
IS 1
176
NV
171
41j3
NV
NV
367
286
2S4
482
539
NV
514
1223
NV
NV
1162
S88
762
Belmont
508
NV
350
NV
14?
451
NV
NV
SS7
93 S
366
NV
ICS 1
MV
441
1353
NV
NV
1670
2816
IW4
Bywify
256
464
20?
NV
•5
328
NV
NV
446
368
256
1381
621
NV
584
984
NV
NV
1338
1096
*68
Biainwee
455
231
22B
NV
185
413
NV
NV
486
398
288
694
684
NV
434
1240
NV
NV
148?
1194
863
Brooklme
442
NV
212
NV
18.1
304
NV
NV
51 7
431
210
NV
636
NV
542
312
NV
NV
1550
1472
631
Cambridge
IBS
298
24 0
NV
110
388
NV
NV
461
370
12 2
634
721
NV
508
1164
NV
NV
1382
1111
887
Cdnk?o
29?
201
223
NV
82
441
NV
NV
478
377
24 3
625
669
NV
245
1324
NV
NV
1434
Hi?
730
Cattela
675
229
306
NV
24.0
728
153
NV
471
680
42 7
685
SI 7
NV
721
2183
460
NV
1412
2040
1282
Chelsea
27b
308
234
NV
117
375
NV
NV
41.1
324
299
919
703
NV
Sit
1125
NV
NV
1234
971
896
Cehassei
?75
372
306
NV
20s
34.4
NV
NV
507
431
364
1115
913
NV
81.4
1033
NV
NV
m t
129 i?
1092
113
154
137
13
SB
323
NV
6109
45 2
251
18 2
463
410
38
293
870
NV
1832,7
1357
753
54S
Framinglon
36S
141
255
NV
17.4
401
NV
NV
45,4
347
303
444
783
NV
52,2
1204
NV
NV
1361
104 2
909
Franklin
475
588
34 2
NV
169
433
NV
NV
55 5
565
389
1503
102 7
NV
506
1299
NV
NV
1664
1696
1166
Hill
58
178
126
NV
1*3
2BM
NV
109.6
42,0
266
17.5
52.7
37.9
MV
42a
865
NV
3289
<259
799
524
Holbfook
275
21 2
264
nv
KB
41.1
NV
NV
5? 8
42.3
344
637
793
Nv'
438
eaa
NV
NV
1584
1270
103,1
Kingston
75
sa
NV
NV
11.8
24.6
NV
NV
323
163
1 1
292
NV
NV
349
73.9
NV
NV
969
409
32
Lexington
STB
104
34 s
NV
206
42.2
NV
NV
496
48®
374
312
1036
NV
61,7
1267
m
NV
W88
146,7
1123
Lynn
175
18 4
9 7
NV
153
353
NV
MV
410
32.2
1B7
651
292
NV
459
1058
NV
NV
mo
965
562
Maiden
388
IB
269
NV
15.9
36.9
NV
NV
439
50.2
S3
1205
807
NV
476
1108
MV
MV
1497
15016
87.9
Marbteheari
t?S
NV
36 7
NV
12 7
24.7
NV
NV
441
830
24 0
NV
1100
NV
MO
*1
MV
NV
ma
69,1
719
Marshtwrtd
fib
144
11.6
NV
163
32.4
NV
NV
4ao
261
EOS
432
34.8
NV
489
97.8
NV
NV
1291
782
614
MwJtord
292
28.)
231
NV
17.1
36.7
NV
NV
49*}
358
284
84 2
68,4
NV
513
1072
MV
NV
147.0
W7.3
851
M«lre$e
425
586
312
NV
20,2
405
MV
NV
488
47.6
349
1666
#35
NV
805
121.4
MV
NV
1460
1424
1046
Millts
350
148
186
NV
14.0
420
MV
NV
474
318
268
443
568
NV
421
125.9
NV
MV
1420
95.3
804
Milton
Mam*
450
926
363
NV
232
57.7
58
NV
487
759
456
8780
1090
NV
898
1730
17,4
NV
1460
2277
136.4
Net,>tJ)k«n
175
152
191
NV
110
61.2
51
NV
390
401
338
456
574
NV
508
1836
15.3
NV
1171
1203
1013
Newton
505
82 t
330
NV
SB
39.7
MV
NV
«1
458
354
1562
M0
NV
59.4
1191
NV
NV
147,3
IT/3
1063
NOfweK
475
312
1??
NV
170
47.2
NV
NV
509
47.7
369
938
967
NV
509
1416
NV
NV
1528
1431
1108
Norwood
325
289
830
NV
17.7
41.1
NV
NV
461
44 0
309
808
681
NV
532
1233
NV
NV
1383
1343
928
Pejbody
45 B
299
26.4
NV
180
at
MV
NV
49fl
365
294
898
79 1
NV
478
114 4
NV
NV
1494
1094
862
OuBtcy
195
254
228
NV
m.s
38.6
NV
NV
447
339
293
783
683
NV
486
1159
NV
NV
1341
1016
878
Rev«
75
168
149
NV
125
24.2
NV
NV
454
?>3
148
504
447
NV
374
72.7
NV
NV
1,362
639
444
Satem
225
210
221
NV
12.1
2fi4
NV
NV
4&0
253
21 7
aai
662
NV
361
852
NV
NV
1488
760
650
Saugus
325
ao»
254
NV
tat
3B.2
NV
NV
527
351
303
eas
752
NV
41.7
1085
NV
MV
15ft 1
1054
808
Sctluats
225
191
175
NV
282
313
NV
NV
352
283
261
572
524
NV
786
saa
NV
NV
1057
850
784
Shertxxn
825
NV
470
NV
110
475
48
NV
604
658
445
NV
1409
NV
534
142 5
145
NV
1812
1973
1336
Somervtite
208
NV
235
NW
197
39.1
NV
NV
409
356
323
NV
764
NV
58.1
1173
NV
NV
122,6
W68
868
Stoneham
zra
175
165
NV
185
436
NV
NV
386
358
262
824
494
NV
555
1308
NV
NV
1157
1075
788
Slooghton
so
182
IS 2
NV
15.7
aai
NV
NV
425
211
177
486
455
NV
471
644
NV
NV
1274
§95
530
Swampscott
475
442
351
NV
139
405
02
NV
558
491
385
1326
1054
NV
416
1396
04
NV
1674
1474
1153
"foptfm
-------
Version 2
; TABLE 14-16, AVERAGE VALUATION ESTIMATES FOR RISK REDUCTIONS USING
i CONTINGENT VALUATION AND CONTINGENT RANKING ESTIMATES®
^ Valuation Change in.axpoaura n»k ___
framework 1/5 to 1/10 n 1/10 to 1/20 n 1/30 to 1/60 n
I I Contingent valuation
Conditional risk 1/10 14,19 42 13,85 47 19,19 48
(19.89) (20.56) (25,82)
Conditional risk 1/20 28,20 46 31,02 46 19,73 45
<4?,39) (48,34) (42.95)
rj. Contingent ranking0
Version 1
P Nc 185,42 206 N 77.71 206 N 25.90 206
(928.81) (464,41) (154,80)
E « 152.15 R 78,§8 H 25.38
(67,28) (33.64) (11.2.1)
R -115.17 206 ft .58 206 B -13,19 206
(316.83) .42) (52,81)
E ft 107.48 R 53.74 N 17,91
(33.93) (16.97) (5.85)
Version 3
P R 240.98 20® R 120.49 206 R 40,16 206
(64.09) (32.04) (10.68)
E R -48,48 R -24,24 ft -8,08
CIS.64) (38.32) (12.77)
N -455.76 206 N -227.89 206 N -75.96 206
(2.882.36) {1,441.18) (4«0.39)
R 212.36 R 1*1.18 R 47.06
(44.71) (22.35) (7.45)
R 250,18 206 R 12S.09 206 S 41.70 206
(136.06) (68.03) (22,68)
R 184.88 206 R 92.44 206 R 30,81 206
(65.42) (32,71) (10,90)
sNym6ers in parenth«sas baiow th« means ara tr>« attimatatf sample standard dtvunoni, n is the sample
»'*• involved in esch case.
p refers to tha model using the paymont interaction, and E the exposure risk InMracttori.
*l designates tha failure to reject the null hypothesis of equality of m*an* and R indicates rejection.
The t-statistic us#d for these calculation* a»
Version 4
Version 5
sp
sp
1 /(n'.-Dst* * (n^-Ds,2 /ft, ~ n, \
y "i * °2 - 2 \"i • j
14-45
-------
There are two ways of interpreting these results. First, since the means
were developed independently, we might suggest the application of a test for
the null hypothesis of equality of means for the relevant exposure risk reduc-
tions using each of the potential contingent ranking models. This is not the
correct interpretation of the contingent ranking valuations. Each estimate in-
volves a nonlinear combination of random variables (see Equations (14.13) and
(14.15)). There will be a different mean and variance for each individual by
the assumptions of the model. Thus, to use these calculated values is equiva-
lent to ignoring this information. Nonetheless, it is a practice that closely
parallels the types of comparisons in valuation estimates reported in earlier
studies. Consequenlty, we have reported it to provide an approximate gauge
of the relationship between the average estimates derived from the two meth-
ods. When the conventional t test for difference in means is applied as an
index of disparity in the two estimates, there are only seven cases where the
null hypothesis of equality would not be rejected at the 5 percent level. These
cases are identified in Table 14-15 with an N prior to sample mean (R desig-
nates rejection of the null hypothesis). In each case the test involved a com-
parison of the mean of the calculated contingent ranking valuations with the
relevant contingent valuation mean. Since the contingent ranking design points
held the conditional risk constant at 1/10/ this implies the tests involve only
the first row of the table.
It is difficult to isolate features that distinguish the cases where we would
be led to a judgment that the null hypothesis could not be rejected in relation-
ship to the remainder where it is. Clearly, the differences can arise from
substantial differences in the underlying distributions for these random vari-
ables, making the use of the test statistic as a crude index of disparity a poor
discriminator. Equaily important, differences in the composition of the individ-
uals in each contingent valuation subsample versus respondents receiving the
contingent ranking questionnaires who were used for these calculations could
be another factor in these pronounced differences.
One approach to avoid these criticisms is to assume we can only focus on
the mean estimates from each method and apply instead the Cummings, Brook-
shire, and Schulze [1984] reference accuracy criteria. Here we can find sev-
eral cases with overlapping intervals implied by taking ±50 percent of their
14-46
-------
respective estimated values. The cases of consistency as judged by this
approach are confined to those with the smallest risks (i.e., the change from
1/30 to 1/60),
A second approach for developing a comparison of the contingent valuation
and contingent ranking estimates does so at the individual level. To illustrate
how it would proceed we have selected two contingent ranking based models--
the basic specification applied to the full sample using the payment (P) and
the exposure risk (E) formulations. With these estimated models it is possible
to calculate for the contingent valuation respondents a valuation for the risk
changes they are asked. Moreover, these estimates can be developed taking
account of each individual's characteristics (to the extent they are included
as determinants of the contingent ranking utility functions). Table 14-18
reports a summary of these results for two smaller risk changes that overlap
the contingent ranking design risks. This table reports regression models
following Theil's [1961] proposed approach for evaluating the predictive per-
formance of economic models. In this case the contingent valuation for these
specified risk changes have been regressed on the estimated payment change
to maintain constant total utility in the presence of that risk reduction that
was calculated for each individual using each of two of the random utility mod-
els estimated with the contingent ranking responses. Each of the four contin-
gent valuation design points involving these two risk changes is considered
separately, because we would expect that the level of the conditional risk,
which cannot be reflected in the contingent ranking models, would affect the
consistency between the two approaches to estimating the valuation..
These results confirm the findings using the Cummings et al. reference
accuracy criteria in that the contingent valuation and contingent ranking esti-
mates appear more consistent at the lower risks (i.e., the case of a change
from 1/30 to 1/60). The correspondence does not appear to be affected by
the tevel of the conditional risk. Indeed, the results are approximately com-
parable for the two levels. Finally, the model using interactions with exposure
risk is more compatible than that involving interactions with payments.
Overall, while these models are preliminary and the comparisons limited
tq the two basic models estimated using the full sample of contingent ranking
responses, it does appear that contingent valuation and contingent ranking
can provide comparable valuation estimates. Of course, the selection of the
14-47
-------
TABLE 14-16. COMPARISON OF CONTINGENT VALUATION AND CONTINGENT RANKING VALUATION
ESTIMATES USING INDIVIDUAL RESPONSES3
1/10
to 1/20
1/30 to
1/60
Independent
variables
D1
D2
D5
D6
P
E
P
E
P
E
P
E
Intercept
-4.24
(-0.87)
-5.15
(-0.79)
27.97
(2.29)
-8.56
(-0.38)
6.60
(0.53)
2.26
(0.20)
-42.60
(-3.18)
-32.85
(-1.28)
CR payment
0.15
(5.11)
(-29.31)°
0,25
(3.73)
(-11.19)
0.06
(1.08)
(-17.09)
0.48
(2.167)
(-2.34)
0.33
(1.06)
(-2,17)
0.56
(1.59)
(-1.25)
1.45
(5.31)
(1.83)
1.83
(2.47)
(1.12)
R2
0.42
0.28
0.03
0.13
0.63
0.06
0.58
0.13
F
26.13
13.87
1.17
4.70
1.12
2.51
34.97
8.12
n
37
37
34
34
38
38
38
28
3
The numbers in parentheses immediately below the estimated coefficients are t statistics for the null hy-
pothesis of no association.
Id
The second set of numbers in parentheses test the null hypothesis of unity for the slope parameter.
-------
model used to organize the contingent ranking responses and range for the
risk change are the key variables in this conclusion. The consistency exists
for only one model over a limited range of the risk change. This may reffect
inadequacies in our preliminary models for the contingent ranking results or
ranges where the two may not be expected to perform consistently.
14.6 SUMMARY
This chapter has summarized the preliminary results from the analysis of
the contingent ranking component of our survey. These results seem promis-
ing . Simple versions of the random utility model appeared to do reasonably
well in "explaining" the rankings provided by the survey respondents. The
role of the basic influences (i.e., the payment, exposure risk, and income
variables) on the utility indexes seems quite stable. Nonetheless, the initial
efforts to enhance the model specification suggest that there may be scope for
incorporating information on risk perceptions, information, and the nature of
the individual's circumstances in the description of the determinants of these
rankings. However, the development and a selection from among candidate
models appears difficult without a fairly specific theoretical structure to guide
the specification.
Finally, to consider the issues involved in developing a comparison of
contingent valuation and contingent ranking estimates, we developed two simple
comparative appraisals using the basic model. These suggested some consist-
ency in the two valuation estimates. However, this conclusion was found to
be sensitive to both the contingent ranking model used arid the range selected
fojr the exposure risks.
14-49
-------
CHAPTER 15
A COMPARISON OF CONTINGENT VALUATION AND HEDONIC
PROPERTY VALUE MODELS FOR RISK AVOIDANCE
15.1 INTRODUCTION
This chapter presents a comparative evaluation of the contingent valuation
and hedonic property value approaches for valuing risk avoidance. This evalu-
ation focuses specifically on measures of risk-avoidance actions taken by house-
holds in response to the location of industrial facilities that have hazardous
waste landfills. Risk avoidance differs from risk reductions in the sense that
the risks of exposure to hazardous wastes from the facilities are not reduced;
only the household can control the size of these exposure risks. It can reduce
these risks by some action it takes to avoid them. Thus, our evaluation
focuses on household location decisions as an action for reducing the risks of
exposure to hazardous wastes.
Our evaluation differs from most previous comparative evaluations of the
approaches for estimating benefits associated with environmental resources,
It does not compare values or benefits. Instead, it uses an entirely different
standard to gauge the relative or "reference" accuracy of contingent valuation
and hedonic property value models. Specifically, we compare the "predicted"
distance that an "average" household, in each of 54 survey-area towns, would
choose to locate from the facilities that have hazardous waste landfills. To
develop these predictions we have proposed a combined framework—one that
uses both the hedonic property value model and a "demand for distance" model
developed from the contingent valuation survey together. Each framework pro-
vides an element in the information required to develop these predictions.
Thus, we would expect that if they are mutually consistent descriptions of
the decision process, their combined predictions would be consistent with the
performance of other economic models for location decisions. This argument
implies that the mutual consistency of the models can be judged by how these
predictions compare with the average of the actual distances households have
15-1
-------
selected from these types of facilities. Therefore, this approach offers a new
standard for gauging reference (or relative) accuracy. Previous comparisons
compared two estimates, with the actual, or "true," values unknown.
Despite the alternative perspective provided by this comparison, however,
it does have disadvantages. Specifically because we use the two approaches
jointly to make the predictions, we cannot attribute any inaccuracies exclusively
to one approach or the other. In addition, for our specific comparative evalu-
ation, it should be acknowledged that it is very preliminary. It is based on
initial models for both approaches that are more exploratory than final. I n-
deed, we view this chapter as structuring an agenda for further research
rather than as a report on the final results of an exhaustive comparison. *
15.2 GUIDE TO THE CHAPTER
Section 15.3 of this chapter summarizes past comparative evaluations of
contingent valuation and indirect methods for estimating the benefits associated
with changes in some aspect of environmental quality. Section 15.4 highlights
some of the conceptual issues associated with comparing hedonic models intend-
ed to reflect households' responses to risk and contingent valuation surveys.
Section 15.5 details the elements in the questionnaire and adjustments made to
the structure of the experimental design so that comparative information could
be elicited. Section 15.6 summarizes the data and features of the hedonic
property value model used in our comparison. Although our comparative eval-
uation was based on an initial version of the model developed by Harrison
[1983], our review includes a discussion of the relationship between this vari-
ant of the model relative to a revised formulation recently proposed by Harrison
and Stock [1984] . Section 15.7 describes the ability of the survey respond-
ents, to use distance as a mechanism to obtain risk reductions. Section 15.8
reports the demand-for-distance results derived from the survey responses
*Our recent findings with refined versions of the generalized travel cost
model (Smith, Desvousges, and McGivney [1983]) used to value water quality
improvements (see Smith, Desvousges, and Fisher [1984]) suggest that this
caution is indeed warranted. I n the presence of incomplete plausibility and
sensitivity analyses of the specific models derived from each method, the com-
parati ve analysis can easily reflect specification or other modeling errors with
each method's candidate. They need not be the result of inconsistency in the
underlying methodologies' "true" results.
15-2
-------
and how these are merged with information from the Harrison hedonic model to
predict the distance selected by the "average" household for each town in the
survey area. Section 15.9 reports a comparison of these findings with the
average distances selected by households in each town based on the reported
sales in the Harrison data set. Finally, Section 15.10 discusses the limitations
of the analysis and the potential implications of these findings for further re-
search.
15.3 THE ROLE OF JUDGMENT IN COMPARATIVE STUDIES .
This section discusses the role of judgment in past efforts to compare
alternative benefits estimation approaches. Our purpose here is to provide an
additional perspective on these comparative efforts and to indicate how the
comparison in this chapter differs from them. (See Cummings, Brookshire,
and Schulze [1984] for a detailed overview and evaluation of the majority of
past efforts to compare contingent valuation and indirect methods' estimates of
individuals' valuations of environmental resources. )
Cummings, Brookshire, and Schulze [1984] concluded their recent evalua-
tion of comparative studies by observing that, with the exception of one esti-
mate in Desvousges, Smith, and McGivney [1983), all comparisons yielded esti-
mates that were within their reference accuracy bounds of ±50 percent.* That
Is, ±50 percent intervals defined using the contingent valuation estimates gen-
erally overlapped those defined using the indirect method's estimate by ±50
percent. With one other exception, these studies all sought to compare esti-
mates of the representative individual's valuation of some environmentally rela-
ted good or service--!. e. , services of recreation sites, hunting permits, water
quality levels at water-based recreation sites, air quality levels, and earth-
quake hazard information. (See Table 6-12 in Cummings, Brookshire, and
Schulze [1984] for a summary. ) The only exception involves a comparison of
the estimated elasticity of substitution between wages and the services o|' the
social infrastructure in New Mexico communities. In this case, a hedonic wage
model and a contingent valuation survey's estimates of this elasticity were com-
*This one estimate in Desvousges, Smith, and McGivney [1983] was for
the loss of the area and was Identified in the study as likely to be subject to
error because of the generalized travel cost model's treatment of substitute
sites.
15-3
i
-------
pared and found to provide consistent results. Cummings, Brookshire, and
Schulze [1984] also noted that, while the true values individuals place on com-
modities can never be known, all of the empirical evidence available at the time
of their assessment indicated that contingent valuation estimates are indistin-
guishable from those available from indirect methods in order of magnitude
terms. Moreover, in most cases (and especially where the Cummings, Brook-
shire, and Schulze reference operating conditions are satisfied), the contingent
valuation estimates are within ±50 percent of the estimates derived using other
methods. Consequently, the authors closed their summary with a positive
evaluation of the accomplishments of contingent valuation and of the prospects
for further advances.
One element discussed by Cummings, Brookshire, and Schulze [1984] but
not specifically considered in evaluating the results of the various comparisons
is the role of the analyst's judgment In the construction of each method's ben-
efit estimates. For example, subsequent analysis of the Desvousges, Smith,
and McGivney [1983] comparative results has highlighted the important role
such judgments can play in shaping each method's results and, in turn, the
conclusions that are derived from comparative assessments. Although this
point has generally been recognized as an important component of most con-
tingent valuation research, it has not been specifically made for the indirect
methods' results.*
The Smith, Desvousges, and Fisher [1984] reevaluation of the comparison
reported in Desvousges, Smith, and McGivney [1983] indicates that an analyst's
judgment can have a direct and important effect on the numerical results re-
ported for benefit comparisons involving the valuation of water quality changes
with the travel cost model. Table 15-1 presents revised estimates of the recre-
ation benefits from water quality improvements for both travel cost and contin-
gent valuation approaches, using the average of the two iterative bidding ques-
tion formats to be comparable with the estimates presented in Cummings,
Brookshire, and Schulze [13841, The revised travel cost estimates resulted
*This is simply a question of relative emphasis. Cummings, Brookshire,
and Schulze [1984] do clearly discuss the problems with selecting specifications
for the econometric models used in indirect methods and cite recent evidence
by Coursey and Nyquist [ 1983J on the sensitivity of demand functions to the
assumptions made concerning the stochastic error as one example.
15-4
-------
TABLE 15-1. RESULTS OF THE REVISED DESVOUSGES,
SMITH, AND MCGIVNEY COMPARATIVE STUDY
Water quality change
Approach
Loss of area
Boatable to
game fishing
Boa tab ^ to
swimmab'e
Contingent
valuation8
^ . . ,b
Original
travel cost
Revised0
travel cost
Contingent
valuation ±50%
Revised
travel cost
+50%
20.14
82.65
3.53
10.07 to 30.21
1.77 to 5.30
11.48
7.01
7.16
5.74 to 17.22
3.58 to 10.74
28.00
14.71^
28.861
14.00 to 42.00
14.43 to
43.
29
These estimates are slightly different from those reported in Cummings,
Brookshire, and Schulze [1984]. They reflect the different number of obser-
vations associated with the two iterative bidding estimates and campi te a
grand mean taking these differences into account.
See Desvousges, Smith, and McGivney [1983] for more details.
'These travel cost estimates are based on a simple travel cost model estimated
using the responses of survey respondents who used siLes along the Monon-
gahela River (the study area). See Smith, Desvousges, and Fisher f '984]
for discussion of the specific model.
15-5
-------
from efforts to refine the earlier generalized travel cost mode). However, they
are from a simple travel cost model and not the generalized model or its ante-
cedents . This model was selected after extensive evaluation of the plausibility
of a number of alternative models. Applying the reference accuracy criteria
of Cummings, Brookshire, and Schulze [1984], two of the three ±50 percent
intervals for the revised travel cost estimates for water quality changes overlap
the corresponding intervals for the "average" contingent valuation estimates.
The third estimate remains inconsistent, yet is now much lower than the con-
tingent valuation estimate. When the first model's estimates were used in this
comparison, the travel cost estimate for this case (i.e., a water quality deteri-
oration leading to the loss of the recreational use of the site) was clearly lar-
ger than the subsequent estimates. Thus, this reevaluation illustrates the
important role the analysts' judgments can play in the benefit estimates derived
from each method as well as in any comparative evaluations of different methods
based on such estimates.
The effect of the analysts' judgments is especially important to the inter-
pretation of the findings reported in this chapter. As we rioted at the outset
of the chapter, our objective is to compare the average distances from indus-
trial facilities with on site hazardous waste landfills selected by households with
the distances that are predicted as the "average" household's selections based
on a "demand for distance" model. This demand model was derived from the
responses in our contingent valuation survey. To calculate these predicted
distances, we used an early version of Harrison's analysis of the housing sales
in suburban Boston to estimate the implicit price of distance from an industrial
site with hazardous wastes. In a subsequent ¦analysis (see Harrison and Stock
[1984]), completed too fate to be included in further comparative analyses, a
new specification was reported for a hedonic property value model based on
the same sales data. This new analysis uses a substantially different measure
of the hazardous-waste-related attribute that households are assumed to be
selecting in their site location decisions. Given the differences in the two
models (described further in Section 15.4), it is reasonable to expect that com-
parative evaluations of the contingent-valuation results based on the second
hedonic model could well differ from those using the first specification. Judg-
ment is required to determine which model is the most appropriate basis for
-------
estimating a household's marginal willingness to pay for distance (as a com-
ponent of the site attributes describing the disamenities associated with haz-
ardous waste disposal sites). Such an evaluation is beyond the scope of this
chapter and the research activities associated with the first phase of this proj-
ect. However, it will be important to the development of a final comparative
assessment of contingent-valuation and property-value estimates of the risk-
avoidance activities of households. i
15,4 CONCEPTUAL DIMENSIONS OF THE DISTANCE-RISK RELATIONSHIP
We have argued throughout the conceptual and empirical analyses in' this
report that hazardous waste disposal regulations (if effective) should be treat-
ed as providing individuals with reductions in the risks of being exposed to
hazardous wastes. Consequently, our analysis has focused on estimating indi-
viduals' valuations of exposure risk reductions. Before we can evaluate the
relationship between these results and other approaches to the problem, it is
important to consider how other approaches treat "the outputs" delivered to
households from increased regulations of land-based disposal of hazardous
wastes.* To apply a hedonic property value model to estimate the values rele-
vant to a household for hazardous waste regulations, it is important to know
how the household can respond to hazardous waste disposal practices. While
there are a variety of possibilities, the most direct, and likely most compatible
with the hedonic framework, is to select a distance from a disposal site. In
effect, this distance is a proxy Tor a service delivered to the household. By
increasing the distance between its residence and a landfill with hazardous
wastes, the household is assumed to be receiving some services that enhance
its utility.
The nature of services received are important to understanding the rofe
of distance in the hedonic framework. For example, increased distance from
the disposal site can serve to reduce risk by reducing the number of pathways
through which an exposure can occur. That is, with increased distance, ex-
posure to airborne hazardous substances disposed at the site becomes less tike-
ly. Alternatively, to the extent the household gets its water from a private
*This analysis need not apply only to land-based disposal regulations.
It is general enough to apply to most hazardous waste regulations.
15-7
-------
well and therefore relies on groundwater, distance may also reduce the chance
of an exposure through contamination of the household's water supply. How-
ever, the precise outcome would depend on the relationship between the source
of the well water and the location of the disposal site. It would also be affect-
ed by the character of the disposal practices, the nature of the soil, the
wastes involved in the disposal site, and the time horizon for the analysis.
Clearly, the services provided by increased distance from a hazardous waste
site can be very complicated.
1 Despite these complications, we have assumed (as has Harrison) that in-
creased distance is a mechanism for reducing risk. This is why we have re-
ferred to these distance selections as risk-avoidance actions. In effect, great-
er distances enable the household to avoid or reduce the risk of exposure to
hazardous wastes. Linking distance and risk avoidance is a crucial assumption
because it affects our ability to connect the results of any contingent valuation
survey designed to estimate the option prices that would be paid for reductions
in the risk of exposure with comparable valuation measures derived from
hedonic property value models (see Smith [1985b]). What is at issue is the
nature of the transfer function that individuals perceive is connecting their
risk of exposure to hazardous substances with the distance of their homes from
a hazardous waste disposal site. To use the results of a hedonic model based
on these distances (or functions of them, as in the case of Harrison and Stock
[1984]) to estimate the representative individual's marginal value (i.e., incre-
mental option price) of risk reductions, we must assume that there is an
accepted, and commonly understood, transfer function between distance and
exposure risk. Since the technical factors governing this relationship are
imperfectly understood by experts on the processes governing contamination
of groundwater as a result of land-based disposal of these substances, it seems
unreasonable to expect consistency in the judgments made by the layperson.*
i *This uncertainty played an important role in the design of the Burness
et al. [ 1983] contingent valuation study. Their effort sought to use contin-
gent valuation to estimate individuals' valuations of a regulation given uncer-
tainty as to the effects of hazardous wastes. It was argued, because of the
lack of clear consistency, that the presentation of specific risks would not be
treated as credible information by survey respondents otherwise.
15-8
-------
However, this consistency in risk-perception-distance relationships across
individuals is not required if it is possible to acquire the information neces-
sary to understand the factors that determine how different individuals form
their perceptions of this relationship. Unfortunately, this possibility was ex-
plored in several focus group sessions (see Desvousges et ai. [1984a]) without
success. Thus, our a priori expectations for acquiring the desired information
as part of the contingent valuation survey were not optimistic. Nonetheless,
our questionnaire was structured to include one approach for eliciting informa-
tion on this risk perception-distance relationship.
15.5 ALTERNATIVE AVENUES FOR COMPARISON
This section discusses the two alternatives considered within our research
design for comparing the contingent-valuation-based estimates of individuals'
responses to risks of exposure to hazardous wastes with those available from
a hedonic property value model. The first alternative involves comparing the
valuation responses derived from contingent valuation on a per unit of risk
basis with the marginal valuations estimated from a hedonic property value
model. The second uses the two methods to develop a joint prediction of the
distance a specified (or hypothetical) household would select to locate in rela-
tionship to a hazardous waste site. Each alternative requires different types
of information. The first requires an estimate of distance-risk transfer func-
tion described in the preceding section and, as a result, calls for information
on the distance equivalent to risk reductions. The second requires a descrip-
tion of the behavioral responses of households in the presence of different
prices of risk avoidance activities and consequently requires information con-
sistent with a demand for distance model.
15.5.1 Eliciting the Distances Considered to be Required j
for Risk Reductions
The first alternative for comparison involves eliciting each respondent's
perceived transfer function between distance and exposure risk reduction. If
it is possible to acquire this information, then the contingent valuation re-
sponses can be "translated" into point estimates for the incremental option
prices that would be paid for increases in the distance between an individual's
home in relationship to a land-based hazardous waste disposal site.
15-9
-------
A simple comparison between these estimates and those implied by a
hedonic property value model with this distance measure would then provide
one basis for comparing point estimates of approximately consistent benefit con-
cepts, However, an exceptionally large number of qualifying assumptions are
required to establish a general correspondence between the two sets of esti-
mates.* For example, the hedonic estimates are estimates of a point on the
individual's incremental option price-risk schedule (see Smith [1985]). The
estimation of the option prices that would be paid for discrete risk changes
(i.e., increases in distance) would face modeling and estimation problems anal-
ogous to those facing conventional uses of the hedonic model (see Bartik and
Smith [forthcoming) for further discussion).
In addition, the assumptions needed to use contingent valuation responses
to derive implied valuations of distance changes are equally limiting. Perhaps
the most stringent assumption required is that the function relating the re-
spondent's option price bid to the change in his exposure risk must be com-
bined , in some fashion, with his transfer function for distance and exposure
risk changes. Combining these functional relationships is complicated not only
by conceptual issues, but also by the need to know the functional form. More-
over, it must be estimated based on a limited range of empirical evidence for
*To some extent, this is also true of the other comparative analyses as
well. However, because the results have tended to support the conclusion of
consistency between the contingent valuation and particular indirect method
findings, the required assumptions have been given less attention.
Equally important, in our case, the "commodity" involved in the contin-
gent valuation analysis is a risk change. Based on the Cummings, Brookshire,
and Schulze [1984] analysis, this could be considered a case where the appli-
cation did not satisfy their reference operating conditions (ROCs). These con-
ditions are
Subjects (or participants in the contingent valuation) must under-
stand and be familiar with the commodity to be valued
Subjects must have had or be allowed to obtain prior valuation and
choice experience with respect to consumption levels of the commodity
There must be little uncertainty
Willingness to pay and not willingness to accept valuation measures
should be elicited.
That is, depending on one's own interpretation of the ROCs, this case
could contradict three of the four reference operating conditions.
15-10
-------
these relationships. Thus, even with the ability to successfully elicit informa-
tion on the relationship between distance and perceived risk, the connection
between estimates from both methods using this comparison is crude.
However, some of these same types of problems were present with the
past comparisons of contingent valuation with an indirect method. Generally,
they were avoided by assuming that the connection between physical measures
of air or water quality and the perceived consequences of these amenities1 was
known by individuals. This approach may not be unwarranted, but it seems
a more reasonable assumption for many environmental amenities compared to
the case for relationships involving risk. For example, the whole issue of risk
perception has been controversial,* which has provoked criticism of the ex-
pected utility hypothesis, as we acknowledged in Chapter 3. Consequently,
it is prudent to consider evaluating the plausibility of the distance-risk change
responses before using them in a comparison of contingent valuation and
hedonic property model findings.
Such a comparison need not be concerned with whether the individuals
know the "true," underlying relationship beween exposure risk and distance.
Rather, the objective is to determine what each individual perceives that rela-
tionship to be. An evaluation of plausibility is then an appraisal of the con-
sistency of the responses with what would be derived if respondents under-
stood what was asked. Since the focus group sessions indicated that this level
of understanding might not be realized, we incorporated a different set of
distance-related questions used earlier by Mitchell (1982) to gauge individuals'
aversions to the siting of undesirable facilities. Mitchell used the distance at
which an individual wouid be willing to have a new facility built before desiring
to relocate his home, t This aversion presumably reflects the same motives
~For a good discussion of the problems in understanding the risk percep-
tion process from the psychologists' perspective, see Slovic, Fischhoff, and
Lichtenstein [1982],
tThe specific text of Mitchell's question is given as follows:
Finding new places to build new industrial and power plants is
sometimes difficult these days. I'm going to mention five types
of buildings or sites. Assuming that they would be built and
operated according to government environmental and safety regu-
(continued)
15-11
-------
'pothesized to govern location decisions within the hedonic property value
rr
odet. It is, of course, less carefully controlled. "Desiring to move or pro-
test an action" are activities that require more precise specification if' they
are to be associated with the tangible actions underlying the hedonic property
value model. Nonetheless, Mitchell's results indicated that individuals could
readily understand these types of questions. Our focus group sessions con-
firmed his appraisal. In using them as a gauge of the plausibility of the risk
perception responses we are implicitly assuming that our questions based on
the Mitchell approach request distances that correspond to minimally accept-
able levels of perceived risk (in our case, implicitly recognizing the costs of
moving). By contrast, our distance-risk reduction question developed for the
comparison requests increments in distance that would yield a specified risk
change. Nonetheless, it seems plausible to expect a reasonable proximity be-
tween the two with the former potentially providing an upper bound to the
latter.*
(continued)
j lations, you might or might not feel strongly about living close
to them. For each type of plant please tell me the closest such
a plant could be built from your home before you would want to
, move to another place or to actively protest, or whether it
i wouldn't matter to you one way or another how close it was?
j The specific facilities mentioned were;
| • A ten-story office building
j • A power plant that uses coal for fuel
I • A nuclear power plant
j * A large industrial plant or factory
A disposal site for hazardous waste chemicals (if the government
said disposal could be done safely and that the site would be
inspected regularly for possible problems),
*This is a conjecture and not a conclusion that could be demonstrated.
The exact relationship depends on what individuals take into account in formu-
lating their responses to the Mitchell distance questions. In particular, the
level of risk that would be regarded as acceptable (given the perceived cost
of the action moving) in comparison with the risk changes asked about would
likely be the key determinants of the relationship between these responses.
15-12
-------
15.5.2 Eliciting Oemand-for-Pistance Information
The second approach for developing comparative information does not re-
quire individuals to formally calculate distance-risk change. Rather, it
assumes that proximity is a relevant measure of the perceived risk experienced
by the individual (and of any other disamenities associated with these facili-
ties). The process is straightforward. A price increment per mile of distance
between the individual's home and an industrial site that has a landfill with
hazardous substances is suggested to the individual. This is described as
the increase in the purchase price of the home, holding all other structural,
site, and neighborhood characteristics constant. In effect, the demand-Tor-
distance approach constructs a hypothetical, partial equilibrium single market
for proximity with a constant marginal price for that proximity. It implicitly
assumes that the individual is capable of separating distance from all other
attributes. As explained in Chapter 7, our design asked different individuals
different marginal prices.
Nevertheless, our second approach assumes that a specific relationship
exists between the individual's demand for distance and his demands for other
site attributes. These other demands influence the demand for distance
through the overall price of housing. There are no specific substitution or
complementarity relationships between attributes of the structure, site] or
neighborhood and the proximity. To investigate them would require compar-
ably detailed information on the decisions with respect to any other character-
istics hypothesized to be associated with distance. Equally important, this
approach assumes a constant marginal price for distance to facilitate asking
the questions of respondents. This assumption stands at variance with the
role of most housing and site attributes in hedonic models. Indeed, it is the
nonconstancy of these marginal prices that poses problems with the use of
hedonic property value models to characterize individuals* preferences for site
attributes (see Quigley [1982] and Bartik and Smith [forthcoming]).
15.6 STRUCTURE OF THE QUESTIONS' AND DESIGN FOR
COMPARATIVE INFORMATION
There are three important features of the structure used to elicit infor-
mation for a comparison of contingent valuation estimates of an individual s
risk valuation and avoidance activities. The first of these involves the expert-
15-13
-------
mental design," Our two approaches--!. e., eliciting distances required to res-
ize risk reductions versus distances selected at constant marginal prices per
mile from a disposal site—were considered to be independent in the experi-
mental design. That is, the risk change posed in the first type question was
selected independently of the marginal price for the second. We have effec-
tively assumed that the two responses are made independently. Based on the
pretest and other experience with the questionnaire, this assumption seemed
quite reasonable. The two types of information are requested in different
parts of the questionnaire, separated by a substantial number of other ques-
tions that elicit a wide variety of additional information. Moreover, different
scenario descriptions were used for each approach.
The second feature of the process of obtaining this information involves
the selection of the risk changes used for eliciting a distance-risk change
function. Specifically, these were tied to the endpoints of the exposure risk
vectors used in the contingent valuation questions (see Chapter 7, Figure 7-2).
In effect, the starting exposure probability and the ending exposure probabil-
ity were used for the distance-risk change question. There were several
reasons for this tied design. First, and most important, if these distances
could be elicited successfully, they would correspond to the precise risk
change for which the individual expressed a valuation response. Further
assumptions concerning the functional form of the transfer function between
distance and risk would not be required to translate the bids for risk reduc-
tions to valuations of distance.
Second, the risk circles described in Chapter 8 as the basic vehicle for
plaining the commodity, a risk change, to the individual were discussed at
the outset of the questions requesting distance-risk change information. This
permitted the respondent to gain familiarity with the vehicle and helped to
ghlight the risk postulated to be capable of being controlled—the exposure
sk (or shaded portion of the first circle on the cards, see Figure 8-6 in
hapter 8).
The last feature of this dimension of the questionnaire was the selection
marginal prices for distance. Each respondent was given one of four
hi
ri
C
of
values —$250, $600, $1,000 and $1,300. These figures were described as in-
creases to the purchase price of the house for each mile it was located away
15-14
-------
from the disposal site. The specific values were selected based on an initial
review of an early version of the Harrison property value model, on experience
with this type of question in focus groups, and on the results of other hedonic
property value studies in which distance measures from other types of unde-
sirable facilities were included as site attributes. .
After explaining the concept of risk and the features of the risk circles,
the question requesting the distance-risk change information proceeded in two
parts. The first part asked if the individual thought moving would affect the
risk of exposure to a hazardous chemical in the drinking water supply. If he
answered "yes," the distance was requested. The specific text of the ques-
tions was as follows:
Ptease look at Cards A and C. The risk of exposure decreases from
1 chance in 300, or thirty-three hundredths of 1 percent chance,
on Card A to 1 chance in 1,500, or seven-hundredths of 1 percent
chance, on Card C. Since your heredity doesn't change, the middle
circles don't change. This also means the combined risk decreases
from 1 chance in 30,000 to 1 chance in 150,000, or from thirty-three
ten-thousandths of 1 percent to seven ten-thousandths of 1 percent. |
Now think about a hypothetical situation using Cards A and C. Sup-
pose that Card A shows your risk of exposure from a hazardous
chemical in your drinking water supply. Do you think that by mov-
ing you could reduce your risk of exposure to the level shown on
Card C? I am not asking would you actually move, but is it possible,
that by moving you could reduce your risk to the level on Card C?
For a positive response then:
How far do you think you would need to move to lower your risk to
the exposure level on Card C?
Of course, the precise size of the probabilities explained in the introductory
text varied with the specific design point (see Figure 7-2 in Chapter 7),
The second question used for comparative information occurred after all
risk valuation information had been requested of respondents to separate it
from the distance-risk change responses and, equally important, to avoid the
potential of the marginal prices of distance serving as informational "anchors"
and thereby affecting the individual's valuation responses for the postulated
risk changes. This question first elicited information on the average cost of
a house in the respondent's neighborhood, if the individual could not provide
an estimate, the interviewer suggested a value based on estimates for the 1980
15-15
-------
Census for each town.* Immediately following this question, the respondent
was told that he could select a location for his home in relationship to a manu-
facturing plant that disposes of hazardous wastes in a landfill at the site.
Once the distance was given, the interviewer calculated the revised cost of
tNe home and asked if this was what the individual had intended. The spe-
cific text of the two questions is given below:
I want you to think about another, completely different situation.
This is about distance from a plant or factory site with hazardous
waste and how it might affect your choice of where to buy a house.
But first, what would you say is the average cost of a house in your
neighborhood?
Now, suppose you could choose between two almost identical homes
like those in this neighborhood. That is, they have the same num-
ber and types of rooms and all their other features are the same;
and your children would go to similar schools. The only difference
between them is their distance from a manufacturing plant that dis-
poses of its hazardous waste in a landfill at the plant site. Sup-
pose you could pick any distance you would want from the hazard-
ous waste site, except that for each mile between your house and
the site, you would pay $250 more than for the same house you could
get next to the site. For example, suppose the price of a house
next to the site was (READ AVERAGE COST FROM ABOVE); then
the same house 1 mile away would cost (READ AVERAGE COST) plus
$250. At an additional cost of $250 per mile, how many miles away
from the plant site would you choose to be?
The last component of the information used in the comparative analysis
was a replication of the Mitchell [1982] distance question. The basic concept
of the question was retained, but the action was recast as an individual moving
rather than desiring to move. The specific set of facilities described to re-
spondents had several categories that overlapped with those of Mitchell's analy-
sis and some new types of facilities. The specific question was as follows:
Finding places to build new industrial or power plants, businesses
or commercial buildings, or public facilities is sometimes difficult. I
am going to name some different types of facilities. Suppose that
each of the things I name would definitely be built and would be
operated according to government environmental arid safety laws.
~Appendix G reports the specific values that were made available to inter-
viewers. However, for the sample of homeowners, the respondents were able
to provide an estimate of the average value of homes in their neighborhood.
15-16
-------
Tell me the closest distance to your home that each facility could be
built before you would move. If you wouldn't move no matter how
close it was built, please tell me,
DISTANCE DON'T
IN MILES KNOW
Ten-story office building
Large industrial plant without
hazardous wastes....................
Coal-fired power plant..............
Nuclear power plant.
Four-lane interstate highway........
Gasoline station/convenience store...
Large industrial plant with a
hazardous waste landfill.............
* = response code for does not know the answer. |
One final aspect of the process of eliciting information for the survey
should be acknowledged. The sample size available for analysis is larger with
this information because these questions were included on both the contingent
valuation and the contingent ranking variants of the questionnaire,* In what
follows, the analysis focuses on only the homeowners in the sample because
this is the group considered in the hedonic property value analysis. Moreover,
it seemed reasonable to expect that they would be more capable of responding
to the second of the two comparative questions. This restriction yields a basic
sample of 391 observations (before adjustments for nonresponses and the asso-
ciated missing information associated with each model specification).
15.7 THE HARRISON HEDONIC PROPERTY VALUE MODEL
The property value model that forms the basis of the indirect method for
estimating the extent of risk avoidance on the part of households was devel-
oped by Harrison [1983]. It is based on the sales of over 2,000 single-family
housing units in 83 towns in suburban Boston. The study included extensive
information on both the characteristics of these housing units and on the attri-
butes of their neighborhoods, including indexes of air quality, access to the
*See Chapters 7 and 9 for a more detailed discussion of the rationale for
and specific details of the experimental design underlying the survey.
*
*
*
*
*
*
*
15-17
-------
central business district, crime, the property tax rate, a measure of school
quality, and detailed information on the location of industrial sites, landfills,
and land-based disposal sites for hazardous wastes. Ten of the eighty-three
towns included landfills with hazardous wastes disposed in them. These were
industrial sites with landfills containing hazardous wastes on their premises.
A total of 11 such sites was identified in the suburban Boston area, with two
sites in Woburn. The Harrison data include information on the distances of
each property to the closest site, the type of site, and detailed information
on the sources of water for each town. Table 15-2 providfes a description of
the variables used in the Harrison model.
The housing prices are sales prices for transactions between November
1977 and March 1981, with most of the sales taking place toward the end of
the period. Prices were measured in constant dollar (1377) terms in the first
analysis of these data (Harrison [1983]). Although a variety of models were
considered, our attention will focus on the log-linear specification of the model.
The proxy measure assumed to reflect the disamenity influence of the hazard-
ous waste site was represented by two distance variables. The first of these
measures the distance of the house to the closest industrial site, whether or
not hazardous wastes were actually disposed of in a landfill at the site. The
second is an interaction variable—the product of a qualitative variable, which
was unity if the site had a landfill with hazardous waste (zero otherwise), and
the minimum distance measure. This format was selected in an attempt to sep-
arate a disamenity effect for industrial sites and a separate, additive effect
(i.e., a shift of the slope coefficient for the distance measure) that was asso-
ciated with whether hazardous wastes were actually disposed of in the landfill
at the site.
Table 15-3 lists the hazardous waste sites involved. Figure 15-1 shows
these sites in the sample-area towns, with shaded triangles indicating the loca-
tion of the hazardous waste landfill. Table 15-4 reports the hedonic function
reported in the first Harrison [1983] analysis of these data. This model formed
the basis for our comparative evaluation.
In subsequent analysis, Harrison and Stock [1984] have respecified the
model, adding new potential determinants of housing prices and--more signifi-
cant for our comparative analysis--changing the definition of the measure of
tHe disamenity (i.e., the measure of the perceived exposure risk) due to the
15-18
-------
TABLE 15-2. DESCRIPTION OF SELECTED VARIABLES IN
HARRISON PROPERTY VALUE DATA SET
Variable name
Description
MINDISTT
Ml NDlSTI
LOT
STORIES
YRBLT
HEAT (FORCED AIR)
HEAT (HOT WATER)
HEAT (STEAM)
CONST
COND
LOT
BATH
SPACE
ROOMS
BASE
Fl REP
COVPARK
CRIME
TAX
Distance to nearest hazardous waste site
Distance to nearest industrial site
Lot size
Number of stories of house
Year house was built
Type of heating system, variable = 1, then forced
air, 0 otherwise
Type of heating system, variable = 1, then hot wjater,
0 otherwise
Type of heating system, variable = 1, then steam
0 otherwise
An index of the quality of construction, variable is:
0 = poor, 5 = excellent
An index of the present condition, variable is;
0 = poor, 5 = excellent
Lot size in square feet
Number of full bathrooms
Living area in square feet
Number of rooms
Percentage of basement that was finished
Number of fireplaces
Qualitative variable for covered parking, variable = 1
for covered parking, = 0 otherwise
Crime rate
Full value tax rate
(contir
ued)
15-19
-------
TABLE 15-2 (continued)
Variable name
Description
PTRATIO
Pupil to teacher ratio
STAT
Fraction of low status in tract population
TOXIC
Area of nearest hazardous waste site
ACC
Index of access to employment centers
NOXO
1980 air pollution measure in the Census tract
RAD
Index of access to radial highways
CHAS
Qualitative variable, variable = 1, indicating tract
borders the Charles River, = 0 otherwise
THREAT
Qualitative variable on whether municipality's water
supply has been threatened by a hazardous waste
site, variable = 1 if threatened, = 0 otherwise
FILL
Area of nearest landfill
D
Distance to nearest industrial or hazardous waste site
DW
Interaction variable between D and a qualitative
variable = 1 if site contains hazardous wastes and
= 0 otherwise
aThe omitted category
is described as "other" heating systems.
15-20
-------
TABLE
15-3. HAZARDOUS WASTE SITES IN THE
SMS A IDENTIFIED BEFORE 1982
BOSTON 1
Town
Name
Approximate
land area
(acres)
Date cf
discovery
Acton
W, R. Grace Company
400
Dec 1978
Ashland
Nyanza, Inc.
30
1987
Bedford
BSAF Industries
5
May 1978
Beliingham
Benzenoid Organics
4
Oct 1980
Cambridge
W. R. Grace Company
10
Mar 1373
Canton
Indian Line Farm
25
Dec 1980
Kingston
Marty's GMC
1
Apr 1980
Salem
Salem Acres, Inc.
180
Sept 198b
Weymouth
Agrico
10
May I98tj
Woburn
Industriplex 128
300
June 1979
Woburn
Wells G and H
200 (plume)
0.005 (wells)
Sept 137f)
Source; Harrison
and Stock [1984].
15-21
-------
L C«CNO
Figure 18-1. Map of survey area showing industrial sites
with landfills containing hazardous wastes.
15-22
-------
TABLE 15-4. HEDONIC PROPERTY VALUE FUNCTION; INITIAL VERSION
Variable Coefficient t-statistic
Dependent: Log (HV) '
Constant
7.677
22.12
STORIES
-0.006
-0.61
HEAT (FORCED AIR)
0.048
3.26
HEAT (HOT WATER)
0.098
6.39
HEAT (STEAM)
0.068
3.40
CONST
0.056
9.23
COND
0.034
6.91
YRBLT
-0.002
-3.03
YR8LT2
0.. 00006
6.21
log (LOT)
0.073
9.42
log (BATH)
0.088
5.29
log (SPACE)
0.338
18.49
BASE
0.0004
1 .92
FIREP
0.096
13.79
COVPARK
0.085
8.64
log (TAX)
-0.336
-12.23
log (PTRATIO)
-0.264
-6.10
log (STAT)
-0.006
-3.25
log (ACC)
-0.204
-13.43
log (NOXO )
-0.575
-6.78
log (RAD)
0.059
6.18
CHAS
0.079
2.94
THREAT
0.031
1.07
FILL
-0.0007
-4.59
D
0.013
1.59
DW
0.054
3.17
Adjusted R2 = ,72
Total degrees of freedom = 2186
Source: Harrison [1933}.
15-23
-------
location of the hazardous waste landfills in relationship to private homes. In
heir revised analysis, Harrison and Stock considered two indexes of perceived
risk of exposure to hazardous wastes. These indexes are based on functions
of the inverse of the square of the distance of each of the 11 sites to the
house. The first index is simply the sum of these inverses across all sites.
The second weights these inverses of the squared distances by the area of
the site as a proxy for the volume of the chemicals at the site. The rationale
for their specification is based on a simple physical model of the exposure
process. Under the assumption of a uniform dispersion through a homogeneous
medium, the concentration of contaminants at any house's location will decline
with the inverse of the square of the distance of the contamination source from
the house.*
It is important to note that this framework may not be relevant to expo-
sure through an individual's water supply if the houses are served through
municipal water supplies. In that case, the relevant measure would relate to
the location of the source in relation to the wells providing each town's water.
Harrison and Stock acknowledge this possibility, but argue that the effects of
risks of this type cannot be distinguished from the "town effect" measures for
each town represented by qualitative variables in their revised model. These
variables were also not part of the original Harrison specification. In addition,
two further modifications were introduced. Since the sales took place over a
period in which interest rates and other financial factors changed substantially,
with corresponding influences on both housing prices and sales activity, the
authors introduced qualitative variables for the quarter in which the sale took
(ilace. Finally, a variable to reflect "income effects" was included in the
hedonic price function. They explain their rationale as follows:
Our formulation implies that the willingness to pay to avoid the risks
of living near a hazardous waste site would depend on income. To
account for such a relationship, we included interaction terms in
which the two hazardous waste variables [the risk measures described
: earlier] were multiplied by the predicted price obtained from an ini-
tial regression [i.e., hedonic price function]. We anticipated that
the interaction term would be negative to reflect an increasing mar-
ginal value of waste cleanup as income (and hence predicted house
price) rises. (Harrison and Stock [ 1984 ], pp. 18-19).
*This explanation paraphrases the Harrison-Stock [1984] explanation.
They also used the number of industrial sites, within various concentric circles
around each housing location, to measure disamenity effects.
15-24
-------
This additional variable seems somewhat controversial. Based on the
authors' explanation, their specific rationale for including it is not clear.
The hedonic price function should not reflect the buyer's income as a measure
of willingness to pay. The hedonic function is an equilibrium relationship
describing the locus of market equilibria for a differentiated product. It may
well be that the authors intended to postulate segmented markets by income
level, similar to Thaler and Rosen's [1975] suggestion of differentiated implicit
markets for risk in their hedonic wage model. In that case, one might still
wish to question their proxy variable because of its construction. |
Nevertheless, the revised model does not yield statistically significant
estimates of the parameters for the risk terms. That is, based on a simple
examination of the estimates, some of the specification changes do not appear
to be clear improvements in the earlier format. Consequently, further compar-
ative evaluation of the two models will be necessary before it is possible to
unambiguously conclude that the revised model is a superior basis for describ-
ing household risk-avoidance activities.
However, a detailed evaluation of the two models is beyond the scope of
this report; the first has been accepted as the initial basis of performing an
initial comparison and describing the factors that may be important to the re-
sults of such an evaluation. |
15,8 RISK CHANGE AND DISTANCE
Table 15-5 reports the mean responses for our version of Mitchell's dis-
tance questions by town for two types of facilities — a large industrial plant
without hazardous wastes and one with a landfill containing hazardous wastes.
These responses are reported in the second and third columns of the table
with the estimated standard deviation in parentheses below the mean. The
fourth column provides the average distance reported for the risk change from
the exposure level associated with Card A to that of Card C. Consequently,
this mean is based on distances for quite different risk changes. That is, as
we noted earlier, this risk change was specifically tied to the experimental
design for the valuation changes measured with the contingent valuation meth-
od. This implies that we are adding together the distance response percep-
tions for quite different risk changes, depending upon the composition of the
design points that happen to be contained in each town.
15-25
-------
TABLE 15-5. AVERAGE DISTANCE RESPONSES, DISTANCE RISK CHANGE RESPONSES,
AND ACTUAL DISTANCES BY TOWN3
Ol
KJ
Ol
Town
Distance from
industrial plant, mi
Plant Plant
Number without with
of obser- hazardous hazardous
vations wastes wastes
Actual distance
from industrial
plant (minimum), mi
Distance
for risk .
change, mi
Number
of
sales
Plant
without
hazardous
wastes
Plant
with
hazardous
wastes
Abington
7
2.71
24.50
2.00
13
4.96
8.36
(1.98)
(22.60)
(1-09)
(1.65)
Acton
114
3.64
16.02
100.66
31
2.03
1.94
(9.80)
(21.07)'
(134.15)
(0.99)
(0.96)
Arlington
2
3.00
39.67
100.00
58
2.31
2.32
(3.00)
(51.38)
(0.67)
(0.66)
Belmont
6
6.67
58.29
50.00
63
1.91
1.83
(2.58)
(40.99)
<0.49)
(0.48)
Beverly
10
16.80
24.90
200.00
21
2.54
13.12
(46.83)
(29.98)
(267.71)
(1.35)
(3.53)
Braintree
6
2.67
9.33
83.33
43
2.25
5.53
(2.58)
(7.79)
(28.87)
(1-16)
(1.79)
Brookline
5
2.20
7.20
70.00
32
4.44
4.36
(1.64)
(4.09)
(51.96)
(0.90)
(0.89)
Cambridge
4
2.00
17.25
53
1.95
2.06
(2.00)
(21.99)
(0.71)
(1.23)
Canton
5
1.40
16.60
310.00
9
1.41
9.64
* (2.07)
(10.38)
(410.12)
(0.80)
(1.51)
Carlisle
8
6.25
26.88
S
6.31
4.58
(2.55)
(22.98)
(1.00)
' (0.42)
Cohasset
4
6.25
16.25
26.67
9
6.36
15.28
(4.79)
(11.09)
(17.56)
(0.66)
(2 59)
Framingharn
12
25.13
25.31
320.00
74
2.48
4.27
(79.20)
(35.16)
(391.46)
(0.92)
(1.14)
Franklin
5
1.40
14.00
68.78
34
2.12
8.49
(1 .14)
(20.36)
(89.29)
\
(0.72)
(3.52)
Hoi brook
1
10.00
20.00
18
5.46
6.30
(0.55)
(0.55)
(continued)
-------
TABLE 15-5 (continued)
Actual distance
Distance from from industrial
industrial plant, mi plant (minimum), ml
Town
Number
of obser-
vations
Plant
without
hazardous
wastes
Plant
with
hazardous
wattai
Distance
for risk
change, mi
Number
of
sales
Plant
without
hazardous
wastes
Plant
With
hazardous
waste
Kingston
3.00
(2.83)
s.oo
(S.00)
10.0®
32
2.77
(1.56)
17.25
(5,05)
Lexington
11
5.58
(5.43)
10.83
- (8.26)
126.25
(134.45)
33
5.26
(0.98)
3,99
(0.94)
Lynn
t
1
1.50
(2.18)
17.50
(1768)
50.00
202
2.96
(0.63)
8.34
(2.59)
Maiden
10
2.75
(3.16)
24,20
(15.98)
130.00
8 >.90)
18
2,19
(0.58)
4.84
(0,88)
Marblehead
1
0.00
0.00
9
2.17
(0.39)
13.55
(0.80)
Marshfield
2
1.00
(1 41)
1.50
(2.12)
30.00
31
8.60
(1.38)
IS.81
(2.89)
Med ford
1
7.50
(3.54)
15.00
(7.0?)
18
2,03
(0.40)
2.28
(0.81)
Melrose
6
2.38
(3.79)
7.67
(4.97)
183.33
(28.87)
19
3.03
(0,37)
4.88
(0.63)
Millis
1
2.00
10.00
12
4.39
(0,65)
8. SO
(0.65)
Milton
2
8.00
(5.66)
22.50
(3,54)
100.00
54
4.27
(0.58)
6.29
.(0.91)
Natick
3
2.00
CI.73)
7 00
( 7. 21)
75.00
(43.30)
51 •
3.29
(0.84)
6.48
(1.13)
Needham
2
0.00
(0.00)
1.00
(1.41)
19
7.78
(0.29)
9.50
(0,89)
Newton
6
3.67
(3.78)
178.84
(402,35)
268.75
(124.79)
48
5 78
( 1,00)
5.76
(1.08)
Norwell
6
8.50
(797)
23.50
(21.39)
157.00
(158.TO)
28
4.62
(1.80)
12,21
(2.89)
(continued)
-------
TABLE 15-5 (contlniMd)
Town
Number
of obser-
vations
Distance from
industrial plant, ml
Distance
for risk .
change, mi
Number
of
sales
Actual distance
from industrial
plant (minimum), mi
Plant
without
hazardous
wastes
Plant
with
hazardous
waste*
Plant
without
hazardous
wastes
Plant
with
hazardous
wastes
Norwood
1
5.50
8.50
22
1.74
11.51
(6,36)
(4.95)
CO.50)
(2.19)
Peabody
5
6.83
39.17
40.00
66
1.60
8.16
(8,01)
(38.33)
(14.14)
(1.00)
(2.17)
Quincy
8
3.42
9.17
100.00
162
2.36
6.67
(5.10)
(9.17)
(1.00)
(1.66)
Reading
3
2.00
21.67
78.33
33
0.93
2.04
(1.00)
(12.58)
(105.S3)
(0.48)
(0.74)
Revere
2
0.15
1.80
6
1.73
7.88
(0.21>
(0.71)
(1 18)
(0.92)
Salem
i
18.56
52,89
SB, SO
a
0.67
11.65
(31.78)
(44.18)
(89.72)
(0.48)
(101)
Saugus
6
3.17
17.00
28.33
16
3.84
6.62
(2.79) '
(19.34)
(20.21)
¦
(0.57)
(0.83)
Scituate
4
1.75
30.00
200.§0
46
8.56
14.80
,
(2.36)
(46.90)
(1.28)
(2.62)
Sherborn
4
25.20
56.B0
100. Si
4
3.07
5.52
(42.54)
(41,82)
(140.71)
(1.00)
(1.28)
Somerviilfl
3
'2.33
20.67
38
1.47
1.90
12.52)
(25.72)
(0.47)
, (0.72)
Sloneham
1
1.00
10.00
20.00
16
1.19
2.76
(0.43)
(0.71)
Slough Ion
5
2.00
7,50
35.00
20
2.85
9.95
(2.45)
(11.61)
(13.23)
(0.89)
(1.85)
Swampseott
i
168.00
32.00
100.00
5
2.44
11. za
(407.60)
(37.80)
(0.22)
(0.33)
T opsfield
5
2.80
22.80
800.00
/
2.87
13.25
(2.28)
(18.71)
(0.00)
(0.41)
(0.35)
< continued)
-------
TABLE 15-5 (continued)
T own
Distance from
industrial plant, mi
Number
of obser-
vations
Plant
without
hazardous
wastes
Plant
with
hazardous
wastes
Distance
for risk .
change, ml
Number
of
sales
Actual distance
from induct r>*l
plant (minimum), mi
Plant Plant
without with
hazardous hazardous
wastes w«*tes
Wakefield
4
1.75
(2.22)
56.25
(95.86)
160.00
(121.66)
13
2.01
(0.75)
3.54
(0.77)
Watpole
5
0.92
(0.66)
35.00
(40.62)
80,00
(98.99)
13
1.82
(0.82)
12.61
(2.01)
W«ltham
4
3.75
(2.22)
28.25
(23.58)
30.00
(28.28)
21
4.99
(0.89)
4.98
(0.37)
Watertown
6
3.08
(3.93)
35.00
(23.24)
106.00
(92.63)
16
2.82
(0.64)
2.71
(0.64)
Wayiand
1
1.00
1.00
18
1.88
(0.72)
7.41
(0.95)
Wellesley
3
3.33
(2.08)
11.00
(12.29)
95.00
(77.78)
27
5.10
(0.98)
9.04
(0.66)
Weslwood
8
4.00
(2.20)
30.63
(39.11)
31.75
(45.60)
8
3.24
(0.71)
11.55
(1.02)
Weymouth
79
1.57
(0.74)
7.29
(2.88)
Wilmington
5
5.00
(3.54)
57.00
(41.77)
203.33
(260.83)
17
1.87
(0,68)
3.63
(1.17)
Winchester
1
15.00
50.00
14
2.13
(1.01)
3.50
(0.39)
Winthrop
8
4.31
(3.84)
134.50
(349.79)
100.00
3
2.74
(0.39)
8.99
(0.83)
Woburn
8
1.00
(1.69)
23.50
(32.54)
33.00
(36.98)
21
1.79
(1.21)
2.49
(1.19)
The first four columns relate to the survey responses; the last three are based on the Harrison proper-
.Xy ——cJa&ts—Mmpl - THn—(ujfltbttrt—-
deviations
In those cases where no standard deviation is reported, there was only one respondent to I he question.
-------
To compare the distance for a risk change with the distance responses to
our Mitchell questions, it is necessary to consider the variation in the distance
responses according to the risk change described. Table 15-6 reports the
mean distances perceived for the exposure risk changes corresponding to each
design point in the sample (including both the contingent valuation and the
contingent ranking formats). With the exception of the subset associated with
the third design point under the contingent ranking format (given in the third
row of the table), the means for our Mitchell-distance questions calculated with
these subsamples are of the same general order of magnitude as the means esti-
mated using the town-specific subsamples. The presence of hazardous wastes
at an industrial site's landfill yields a consistent increase in the distance indiv-
iduals would require (the third design point is again the only exception to
this conclusion and results from outlying observations). The distance increase
ranges from 2 to over 25 times larger than the minimum "acceptable" distance
from an industrial plant without hazardous wastes in landfills at the site.
The responses on the distance/risk change questions are more difficult
to understand. When one considers the means across subsamples, there does
not appear to be a consistent relationship between the size of the risk change
and the distance response. In some cases equal percentage changes, regard-
less of the initial level of the exposure probability, yielded distance responses
in fairly cfose proximity. However, this was not uniformly true. Indeed,
there are several important exceptions. For example, the first two design
points of the contingent ranking imply the same percentage risk change as
the first two with contingent valuation question format, yet the distance re-
sponses are quite different. Even within a question format type there are a
couple of exceptions to the assumption that individuals focus on the percent-
age change in the odds and not the size of the actual risk change. Thus, it
appears that either the sample respondents did not understand or they were
unable to formulate a consistent perceived relationship between distance to a
hazardous waste site and the associated risk changes. This conclusion is rein-
forced by the fairly close proximity between the average distance responses
to the Mitchell questions and the average of actual distances from sites for
1
houses involved in sales in each town, shown in Table 15-5. The results are
15-30
-------
table 15-6. MEAN DISTANCE RESPONSES GROUPED BY THE SIZE OF THE RISK CHANGE8
Minimum distance from industrial plant, tni
Exooiurt risk chanae .
n c— — " Av«r»oe distance Pi»nt without Plant with
Version Scenario Change for risk cnwige. mi hazardous wiiKi hazardous »mim
Ranking
1/10
to
1/50
0.08
155.05
17.29
33-35
{159.69}
(37.71)
(35.85)
flanking
1/10
to
1/50
0.08
154.12
7.96
24.50
< 163 . M)
(22.21)
, <30.73)
Ranking
l/?0
to
1/60
0.03
128.SO
64.24
2B.20
(155,12!
(224.04)
(31,12)
Ranking
1 /20
to
1/60
0.03
137.05
3.94
1-4,91
*
{208.043
(3.48}
(13.68)
Direct
1/10
to
1/50
o.ae
39.8?
3.03
80.12
question
(126-37)
(4.61 )
(238.80)
Direct
1/10
to
1 /SO
0,08
61.83
2.35
20.72
question
(50.84)
(4.50)
125.39)
Oirect
1/5
to
1/2S
0.18
61.00
3.33
22,33
quastion
{S3.21)
(2.53)
(32,83)
Direct
1/5
to
1/25
0,16
155,00
1.92
16.92
question
(313.33)
(1,38)
<25,JO!
Direct
1/30
to
1/150
0,023
83.21
3.27
84.73
Question
(135.10)
(4,18)
<253,68)
Direct
1/30
to
1/150
0.023
153.85
4.64
13.49
question
(147.49)
(6.38)
(13.88)
D'ract
1/300
to
1 /1,500
0.0023
97 <3
3.81
27.06
question
(165.65)
(3.54)
(47.29)
Direct
1/300
to
1/1,500
0.0023
72 40
5.10
24.69
guastion
(98 80)
(7.32)
(35.37)
*rm numbers in pir«ni>imi below ttw »«nple means «ra the estimated standard deviation#,
^Ranking designates the contingent ranking version# of the questionnaire, and direct question refers to the
remaining versions used in the research aasign
CTh«r» are multiple replications of the same exposure risk change reported because t*«y correspond 10
Cases where there was another reason 'i.e., vantoi* that changed) to distinguish the design point. How-
ever, thtse other changes were not intended to tht distant* Questions u**cJ for ifi«h results.
15-31
-------
also consistent with our focus group experiences, which led us to develop the
alternative demand-for-distance approach.
Finaliy, the mean distances for subsamples organized by town seem im-
plausible. They are generally a good deal larger than both the averages of
the actual distances and the averages for the Mitchell questions. Thus, ali
the available informal evidence suggests that the use of these responses either
in translating the contingent valuation responses to a valuation per unit of
distance or in the estimation of a distance-risk change function based on them
is not likely to provide an adequate basis for comparing the contingent valua-
tion responses with estimates from the Harrison hedonic property value model.
13.9 THE DEMAND FOR DISTANCE FOR RISK AVOIDANCE
Given the unsuccessful nature of our attempt to elicit the perceived dis-
tance-risk change function of our sample respondents, the questions requesting
information consistent with a partial equilibrium demand-for-distance function
formed the basis for our comparison of the survey and the hedonic models.
Table 15-7 reports the results of the statistical analysis of these responses
with three different functional forms for the demand function — linear, semi-log
(with the fog of distance as the dependent variable), and double-log. The
basic specification included the housing price reported by respondents for the
average house in their respective neighborhoods, the postulated marginal price
of distance, and the household income. The results clearly favor the nonlinear
specifications with both the semi-log and the double-Jog forms exhibiting sta-
tistically significant parameter estimates. In ail cases, the signs of the basic
model's parameter estimates agree with a priori expectations.
There are a wide variety of other determinants that might be considered
in attempting to improve the fit of these models as well as our understanding
of the household responses. Indeed, the design of the questionnaire provides
a reasonably wide range of variables that should be considered potential deter-
minants of these responses--including measures of the individual's attitude
toward risk; his socioeconomic characteristics; the number of children in his
household; the years he has been living in the home; and, based on the recent
experience in this area, the town in which he lived. This analysis is clearly
warranted for further research. However, it is not relevant to this compari -
15-32
-------
TABLE 15-7, DEMAND-FOR-O'STANCE MODELS3
Basic mooet
Selected aiternatwe sp*.fication*
Iindependent
variables
Lineer
Semi-log
Double-
log
Semi-log
Semi-log
Semi *log
Ooubie-
log
Douo.e-
log
Doubie-
«4o
intercept
-0.169
(-0,014)
2.003
(12.114)
1.1S0
(1.S54)
2.181
(10.732)
2-054
{5.528)
1,904
(4.880)
1.829
d »8C)
2,026
C1.751 )
t n
775
509 i
Housing price0
0,08?
(0.930)
0.003
(2.294)
0,416
(2.774!
0.003
(1.989)
0.003
(2.219)
0.003
<2.003)
0.411
(2.640)
0.433
(2,851)
oUo?
(21.654!
Marginal price of
distance
-0,008
c~0.8«4)
-0.0003
{-2.815)
-0.193
(-2.506)
-0.0004
C-2.784)
-0.0004
C-3.158)
-0.0004
(-3.217)
" -0.204
(.-2.523)
-0.216
(-2.834)
-oi
{ -i
220
880.
income6
0.306
(1,626)
0.005
(1.924)
0.124
(1.474)
0.QD5
CI-687)
0.002
(0.648)
0.002
(0.703)
0,094
(1.056)
-0.001
(-0.011>
0t 001
(01014)
Education
•
.
-
-
0.037
(1754)
0.039
(1.878)
-
0.441
(1.720)
oi 46$
C11810}
Age
-
-
-
-
-0.008
(-2.334)
-0.006
(-1.647)
-
-0.391
(-2 338!
-0
<-
317
777 S
Children <1? years
-
-
-
-
-
0.0S3
(1.23?)
-
-
0
(1
054
ISO)
v««r»d
-
-
-
-0.068
(-1.810)
-
-
-0.105
(-1,839)
¦
"
R2
0,028
0.089
0.08S
0.101
0.122
0.128
0 096
0.116
128
F
2 827*
9.579"
9.011**
7.647**
8.078**
8.93§»
7.167**
7,605"
6; 585"
s
58.121
O..S26
0.828
0,829
0.813
0.813
0.831
0.816
0
816
n*
296
296
296
275
296
296
27$
296
296
*Th« numbers In P'®renth«s®$ below the estimated coefficients are t-ratios for the null hypotftesis of no associafij>n.
bTh* housing price is m«#«ui*e
-------
son, because independent measures of these potential determinants are not
available in the Harrison data set and could not be acquired from other sources
such as the 1980 Census. At best, the census reported income, race, educa-
tion, and family composition measures that can be developed at the census
tract level to consolidate to the town level for use with Harrison's information,*
Thus, what is important for the objectives of this research is whether
the omission of these variables seriously biases the parameter estimates for
the variables on which independent information is available'. To address this
issue, we report in the second half of Table 15-7 a selected set of the ex-
panded specifications for the semi-log and double-log models. The principal
interest in these models is the sensitivity of the estimated parameters for the
price of housing, income, and the marginal price of distance to the inclusion
of additional variables. For the most part, it is fairly limited with the semi-log
specification. The estimated parameters for the variables that can be meas-
ured for our comparative analysis are quite stable across any of the three
specifications reported here and, indeed, more generally over several others
that have been considered as part of our preliminary analysis of these re-
sponses, The results are somewhat less encouraging with the double-log
model. In this case, the estimated parameter for income is quite unstable.
Iri in one case, the estimated coefficient is negative but insignificant.
Clearly, such informal comparisons cannot establish whether there will be
substantial biases associated with omitted variables when our comparison is
forced to rely on the basic model with a limited specification. Nonetheless,
they do suggest that the judgments that must be made in selecting a final
specification for the demand for distance model are, in the case of the semi-log
specification, less likely to be important to the "accepted" parameter estimates
for what are clearly among the most important of the economic variables deter-
mining these distance responses. Moreover, this interpretation is consistent
with James-Stein [1961] type estimators, which have influenced much of the
recent work on pretesting and model selection. The James-Stein approach
5 *lt is important to distinguish the information used to estimate the Harri-
son [1383] and Harrison and Stock [1984] hedonic functions from that used in
our comparison. The former are data on individual sales transactions. The
latter is simply a summary of these results providing the mean of the individ-
ual records for each town.
15-34
-------
develops estimates as weighted averages of the results of restricted and unre-
stricted estimators. The restrictions involved can be the exclusion restrictions
associated with differing model specifications. If the estimated parameters do
not vary greatly with alternative treatments of other potential determinants of
the distance responses, the results derived for the weighted estimator will be
approximately the same as the basic model's estimates. This approach also
seems to be in the spirit of Learner's [1983] proposals for reforming the prac-
tices used in reporting econometric results.
Unfortunately, the same conclusions cannot be drawn in the case of the
double-log model. In this case, the results are much more sensitive to the
specification selected. However, this may not be crucial to our further anal-
ysis. If one were required to select a final model, then, based on conven-
tional criteria of minimum standard error of estimate (see TheiI [1357]), the
semi-log would appear to have a slight advantage over the double-log models
with the comparable specifications. Accordingly, while the comparison in the
next section reports the results for both models in a variety of alternative
prediction forms, our primary focus will be on the results with the semi-log
specification,
Before turning to those results it is important to acknowledge the encour-
aging findings from this simple and preliminary analysis of distance responses.
The results clearly indicate the types of tradeoffs implied by the hedonic mod-
els that assume distance will serve as a proxy for the disamenity effects (in-
cluding the perceived risk) associated with proximity to hazardous waste sites,
15.10 A COMPARATIVE EVALUATION OF THE CONTINGENT
VALUATION AND HEDONIC MODELS
As we acknowledged at the outset of this chapter, the primary results
reported in this section are not a comparative evaluation of estimated values
from the contingent valuation and hedonic approaches to benefit estimation.
Instead, we propose to judge the consistency or compatibility of the methods
by using them together to predict the distances that the "average" (or repre-
sentative) household in each of 54 towns would select from an industrial site
with a landfill containing hazardous wastes. These predictions can then be
compared with the averages of the actual distances selected in these towns.
Our application of the two methods accepts as a maintained hypothesis the
15-35
-------
assumption that households treat distance as a proxy for the disamenity effects
of these sites (including, but not necessarily limited to, their perceived risks
of exposure to these hazardous wastes). Therefore, a close correspondence
between the actual levels of distance and the predictions from' a framework
that uses both methods to derive the predictions would yield indirect evidence
of the relative compatibility of each framework's description of the decision
process. It is not a validation of the methods. Moreover, it faces many of
the problems of past comparisons in that a finding of incompatibility of actual
and predicted distances does not provide insight as to which aspect of either
method is at fault.
The specific details of the prediction begin with the basic specifications
estimated for the demand for distance models based on the responses given in
the contingent valuation survey. With these models and information on the
determinants of these distance demands it is possible to project the distance
that the "average" individual would select. These projections are based on a
constructed average household in each of 54 towns in suburban Boston, This
household is assumed to have a housing demand that corresponds to the aver-
age sates price (in 1984 dollars) experienced with the homes in the Harrison
data set. The household income level corresponds to the average of the family
incomes reported for the census tracts in each of these towns for the 1980
Census (using the consumer price index to convert it to 1984 dollars). The
marginal price is calculated from' the Harrison hedonic price function as the
derivative of the function with respect to distance. The marginal price for
the specification of the Harrison model given in Table 15-4 is given in Equa-
tion (15.1) below :
3D ~ ^aD + °bw^ P ' (15.1)
wh
ere
cfp = estimated parameter for distance to the nearest industrial
site regardless of whether it included a landfill with hazard-
ous wastes
«Dw = estimated parameter for the interaction variable of distance
and a dummy variable identifying the site as containing haz-
ardous waste
P = the average price of the houses sold in a town in 1384 dol-
lars.
15-36
-------
There are several aspects of this calculation that are important to the
interpretation of the results. First, we have included both the disamenity
effects of an industrial site (as reflected in a^) and the differential effects
per mile of the presence of hazardous wastes in the calculation of the marginal
price for distance. This specification was selected because of the description
of the scenario used in our question (as reported above in Section 15.4).
Second, there are at least two options for the value of the housing price
used in these calculations including the average of the actual sales prices
(converted to 1984 dollars) and the predicted price based on the average char-
acteristics of houses sold in each town. There are a number of reasons why
these two measures will be different and will imply quite different assumptions.
Use of the second measure, for example, constructs a housing type with the
average characteristics of houses that sold in the town in the Harrison cata
set. A specific house with those features may not exist. With a nonlinear
hedonic price schedule this will not correspond to the average price. How-
ever, if we assume that the hedonic model is correctly specified, it does reflect
only the factors considered to be important to housing prices. It omits other
factors that might have influenced the sale prices in a particular town at a
specific time that would more reasonably be considered random error. There-
fore, it uses the model more specifically than the first strategy. )
However, this specificity is a mixed blessing. Since the hedonic price
function is nonlinear with the log of the price, a function of the levels of some
variables, and the logs of others, we can expect that estimates of the price
based on it wilt be biased (i.e., by Jensen's inequality). There are some ad-
justments that reduce the extent of or eliminate the bias for some cases {see
Goldberger [1968]), but the information required for these calculations was
not available for our estimates of these prices.
Finally, the hedonic price function by its specification implies a marginal
price schedule rather than a constant marginal price as implied by our formu-
lation of the demand for distance survey question. It also does not necessarily
imply that decisions will correspond to the partial equilibrium framework under-
lying the description of the survey questions.
Some of these problems must, by design, be treated as maintained hypo-
theses. Others can be investigated in development of the comparison. (The
15-37
-------
postulated structure of the decision process and approximate constancy of the
marginal price are maintained hypotheses. To deal with the effects of the pro-
cedures used to estimate the marginal price, we consider two separate esti-
rrates-~one based on the average of the actual prices and one using the pre-
dicted price using the average characteristics.
Table 15-8 compares the marginal prices implied by the two approaches
with MP1 being the prediction based on the average characteristics of homes
sold in each town and MP2 using the average of the actual prices. This table
also compares these calculated marginal prices with the average of the design
marginal prices that were asked of the survey respondents in each town
(column 4 in the-table). Clearly, the calculated marginal prices are uniformly
larger than the average of our design prices and fall outside the range of the
prices used. This is unfortunate because it implies that the demand for dis-
tance models estimated from the survey responses are less likely to be relevant
to these cases.
The table also reports the survey respondents' estimate of the average
price of homes in their neighborhood in comparison with the average prices of
sates in that town in the Harrison data set (in 1984 dollars). These prices
are generally consistent, though there are a few notable exceptions, as, for
example, the case of the town of Brooktine. The discrepancies are in both
directions and should not be interpreted as reflecting on the ability of re-
spondents to gauge housing values. Our sample is a representative sample of
the population in suburban Boston, not each town. Moreover, even a repre-
sentative sample of each town's homeowners would not necessarily be repre-
sentative of the prices and characteristics of houses that were selling in any
given period.
The last component of the table is the average distance selected by re-
spondents as their answers to the distance demand question in comparison to
the actual distances of the homes selling in these towns. With the exception
of the distance response for Belmont and Lexington, these two distance meas-
ures are comparable in order of magnitude terms. Of course, this does not
represent a confirmation of either modeling framework. Rather, at best, it
can be interpreted as one indication that respondents understood the features
of the contingent valuation question.
15-38
-------
TABLE 15-8. A COMPARISON Of THE REGION I SURVEY RESPONSES AND
HARRISON DATA BY TOWN
Survey respormi Average of
msnimam
Number Average
of prict of
Towrv oS»«rv»tior>j house, $
Distance
selected,
mi
Average
marginal
price, $
Average
price of
house, S1 MP1
MP2
distance to
hazardous
waste site, mi
Abingtort
7
51,625
6,29
669
62,832
7,779
4,210
8,36
Acton
114
120,102
12.72
788
125,884
8,239
8.434
1.36
Arlington
3
93,000
S.67
9SQ
108.455
6,308
7,267
2.32
Beimont
5
124,375
S3.00
706
147,820
8,073
9,904
1.83
Bev«rly
i
90,000
8.50
798
85,044
13,005
5,698
13.12
Sraintree
6
87,500
18.67
900
77,349
7,052
5,182
5.53
Brookline
6
240,833
11.50
633
98,593
5,874
8,605
4.37
Cambridga
4
171,250
12.50
m
109,845
4,916
7,360
2.06
Canton
5
71,1)00
6.20
700
123,563
13,809
8.279
9.64
Carlisle
7
179,28S
21. S?
681
132,201
11,686
8,858
4.50
Cohasset
3
78,750
11.67
813
130,157
19,083
8,721
15.28
Framingham
14
117,153
22.43
673
102,416
7,472
6,862
4.27
Fr*nklin
S
84,200
14.60
920
73,727
10,601
5,342
8,43
HoibrooK
1
55,000
10-00
1,300
67,980
6,599
4,555
6.30
Kingston
3
56,333
3.33
983
77,277
15,785
5,178
17.25 ,
Lexington
10
158,792
112.00
929
145,315
9.073
9,736
3.99
Lynn
3
43,333
8.33
983
54,331
6,023
3,680
8-34
MMden
10
73,111
15.50
625
66,870
5,843
4.480
4.84
M*rbl*h»»d
1
80,000
0.00
650
145,299
14,506
9,735
13.55
MarshfiBld
2
66,000
1.50
1,150
83,627
14,271
5,602
15.81
Murtford
2
65,000
10.00
825
81,512
6,165
5,461
2.29
Melrose
6
92,500
7.83
1,142
86,357
6.658
5,786
4,88
Millis
1
60,000
10.00
650
94,347
9,745
6,321
S. 50
Midori
2
145,000
6.00
1,300
105,003
9,838
7,035
6.29
N«tick
4
73,333
6.50
975
91,596
8.245
8,137
6 48
Needham
3
126,687
10.67
600
119,540
12,677
8,009
9.50
Nswtort
6
153,714
25-00
929
142,091
8,534
3,520
5.76
(continued)
39
-------
TABLE 15-8 (continued)
Survey responses
T own
Number Average
of price of
observations house, $
Distance Average,
selected, marginal
mi price, $
Av«rage
price of
house, $ MP1
Au«r|gf of
minimum
distance to
hazardous
MP2 waste site, mi
Norweli
6
92,500
15.00
925
104,511
13,878
7,002
12.21
Norwood
2
75,000
5.50
S25
89,983
11,743
6,029
11 58
Peabody
6
83,333
7.83
S7S
84,328
8,127
5,650
8.17
Quincy
6
70,000
10.83
517
69,862
6,234
4,681
6,67
Reading
3
90,000
14.67
350
37,727
6,504
6,547
2,04
Revere
2
30,000
8.00
250
68,380
6,118
4,578
7.88
Salem
8
76,667
9.63
750
68,236
8,003
4,572
11.65
Saugus
6
80,833
7.83
875
72,368
7,660
4,848
6.62
Seituate
3
IIS,000
20.00
713
33,179
12,742
6,243
14.80
Sherborn
5
18.8,000
11.40
560
176,338
10,237
11,319
5.52
Som«rvill«
2
122,500
13.50
600
65,184
4,356
4,367
1.90
Stoneham
1
75,000
10.00
1,000
88,928
S, 146
5,958
2.76
Stoughton
6
58,333
9.00
725
69,713
9,211
4,671
9.95
Swamp scott
6
62,166
12. SO
817
111,119
11,143
7,445
11.29
Topsfisld
5
146,000
12.00
820
150,5«
20,103
10,000
13.25
Wakefield
4
101,250
7.50
625
84,259
6,497
5,645
3,54
Walpole
5
72,083
21.40
725
87,669
13,416
5,874
12.61
Waftham
4
85,000
10.00
800
84,828
7,036
5,684
4.98
Wattrtown
5
99,250
12.00
683
117,052
6,856
7,843
2.7'
Waylarxl
1
200,000
20.00
1,300
128,623
10,916
8,631
7.41
Wailatley
3
166,667
11.67
51?
154,742.
14,939
10,368
9.04
Westwood
8
102,500
12.50
S38
116,017
17,378
7,773
11.55
Wilmington
4
32,000
38.50
920
14,207
5,935
4,372
3.63
Winchester
1
137,500
20.00
1,150
140,456
7,676
5,411
3.50
Winthrop
8
100,006
9.50
869
57,304
3,728
4,509
8.93
woburn
8
68,625
8.00
706
84,652
6,801
3,671
2.49
These prices are averages of the sale prices from November 197? to March 1981. Harrison provided
the mean price by town in 197? dollars. These have been converted to 1984 dollars (the year of the
survey) using the total shelter component of the housing component of the consurrvar price n^ix for
'97? and June 1984.
15-40
-------
Table 15-9 presents the predicted distances implied by each of the two
basic models (the semi-log and double-log specifications in Table 15-?) with
each measure of the marginal price and with a correction to adjust for the bias
induced by using the semi-log and double-log functions to predict the level of
distance.* The specific features defining each type of prediction in Table 15-9
are given in Table 15-10, Several overall observations can be made based on
casual inspection of these results. First, the predictions from the semi-log
specifications (i.e., D3» D4, D7, and D8), our preferred model, are uniformly
less than those with the double-log model. Moreover, they are usually less
than the average of the minimum distances for the houses in each town in the
Harrison data set (i.e., what we are interpreting as the actual distances se-
lected), Both the use of the actual sales price and the adjustment to reduce
the bias in each model's estimates of distance tend to increase the predicted
distance. While the selection of the actual price for housing as the price com-
ponent in Equation (15.1) does not always increase the marginal price esti-
mates, it does increase the majority of the estimates for both the semi-log and
the double-log models.
Comparing the estimates with the average of the actual distances for the
houses in Harrison's sample is difficult. In somewhat less than half the cases
(20 of the 54 towns), our range of estimates considering all three of the fac-
tors that distinguish them—model, housing price, and bias adjustment--does not
include the average of the"actual distances to the nearest disposal site contain-
ing hazardous wastes. Since these ranges are quite large, greater in many
*The specific correction involves using Goldberger's {1968] suggestion
for reducing the bias by predicting distance, D, with the conditional expec-
tation :
Dk = exp (XkP) • exp (| a2) ,
where
D. = the predicted distance for the kth town I
K /
X, = 1xN vector of the determinants of distance demand (in linear
form for the semi-log model and log form for the double-log
specification)
p = Nxl parameter vector
o2 = variance in the error for the distance demand function.
15-41
-------
TABLE 15-9. PREDICTED ANO ACTUAL DISTANCE TO HAZARDOUS WASTE SITES BY TOWN
Actual distance,
Predicted distance, mi
Town
D1
D2
D3
04
05
06
07
D8
Average minimum
distance, industrial
Average
minimum distance,
hazardous waste
6.92
7.7S
1,43
4,17
4.96
8.36
9.51
9.47
1.60
1.51
2.04
1.95
9.14
8.90
2,59
1,94
2,31
2.32
10.14
9.7S
1,79
1.03
1.91
1.83
7.25
8.36
0.43
2.87
2.54
13.12
7.73
8.26
1.90
3 32
2.25
5.53
8,75
8.55
2.89
2.32
4.44
4,37
8.80
8.14
3,65
1.75
1.95
2,06
8.40
9.28
0.29
1.54
1.41
9.64
9.44
9.95
0.62
1.45
6.31
4,58
8.31
9.66
0.06
1.44
6.36
15.28
8.71
8.85
1.81
2.18
2.48
4.27
7.31
8.34
0.65
3.17
2.12
3.49
7.40
7.95
2.07
3,83
5.46
6.30
6.50
8.06
0.13
3.22
2,77
17,25
10.03
9.90
1.36
1,11
5.26
3.99
6.67
7.33
2.29
4.62
2.96
8.34
7.37
7 76
2:54
3.82
2.19
4.84
9.04
9.76
0.26
1.08
2,17
13.55
6.85
8.21
0.22
2.93
8.SO
15.81
7.96
8.IS
2.43
3.00
2.03
2.29
8.28
8.SO
2.20
2.86
3.03
4.88
7.98
8.68
0.89
2.48
4.39
8, SO
8.42
9.00
0.S1
2.14
4.27
6.30
8.16
8.64
1.39
2.62
3.29
6.48
8.68
9.49
0.42
1.72
7.76
9.50
9.98
9.78
1.57
1.16
S. 78
5,76
7,as
8.96
0.27
2.15
4.62
12.21
7.54
8.57
0.48
2.67
1,74
11,58
7.78
8.35
1.38
2.91
1,60
8.17
7.43
7.86
2.29
3,66
2.36
6.67
8.85
8.84
2,42
2.39
0,83
2. 04
7.31
7.73
2.32
3.69
1.73
7,88
7.05
7,86
1,34
3.74
0.6?
11.65
7.43
8.11
1,54
3.57
3.84
6,61
7.56
8,6?
0.37
2,58
8.56
M.80
10.82
10.52
1.10
0.69
3.07
5.52
7,58
7.57
3.88
3.87
1.47
l 90
S. 47
8.52
2.57
2.72
1.19
2,76
7.01
7.99
0.96
3.72
2.85
9.95
8.53
9.22
0,64
1.93
2.44
11 .29
8.74
9.98
0.05
1.01
2.87
13.25
8,14
S.36
2.26
2,92
2.01
3.54
7.37
B. 65
0.29
2.84
1,33
12.61
a,00
8.33
1.91
2,87
4,99
4.98
9.12
8.89
2.21
1.64
2.82
2.71
9.30
9,73
0,76
1.51
1,88
7.41
9.49
10.18
0,25
0,98
5.11
9.04
7,97
9.30
0,10
1,81
3.24
11.55
7.38
8.05
1.57
3.58
1.58
7.29
8.03
8.31
2.69
3,59
1.87
3.63
10,12
9.73
2.02
1.20
2.13
3.50
6.92
7.86
1,03
3.83
2.74
8.99
8.09
8.38
2.06
2.89
1.79
2.49
Atoington
Acton
Arlington
Belmont
Beverly
Blr«irrtre«
Brookiine
Cambridge
Canton
Carlisle
Cohasset
Fran ingham
Fr-anklin
Hoi brook
Kingston
Lexington
Lynn
Maiden
MerDlefveed
Mershf iele
Mjadford
Milrote
Mllifs
Helton
Nfclick
Njeedfiam
Nfcwton
Nprwell
Norwood
Pisa body
Qtlirscy
Rjeading
Rlevere
S»l«*n
Saugus
Scituate
Sherborn
Somerville
Stonehajn
Stoughton
S*vamp5CQtt
Topsfieid
Wakefi#ld
Walpole
Waltham
Waiertown
waytand
wellesley
wjestwooel
Wtaymouth
Wilmington
Winchester
wlnthrop
wjabum
4.91
6,75
6-49
7.20
5.15
S.S3
6.21
6.25
5.97
6.70
5.30
6.IS
5.19
S. 26
4,62
7.12
4.73
5.24
6.42
4.87
5 65
5.88
5.67
5.98
5.80
6.17
7 09
5.57
5.35
5.53
5.28
6.28
5.19
5.01
5.27
5.37
7.68
5.38
6.01
4.98
6.06
6.20
5.78
5.23
5.68
6.48
6.60
6.74
5.66
5.24
5.70
7 18
4.92
5.75
5.S3
6.72
S.32
6.92
5.94
5.87
6.07
5.78
6.59
7.07
S. 86
6.28
5.92
5.65
5.72
7.03
5.21
5.51
6.93
5.83
5.79
6.04
6.16
6.39
6.13
6.74
6.94
6,36
6.08
5.93
5.58
6.28
5.49
5.58
5.76
6.16
7.47
5.38
6.05
5.67
6.55
7.08
5.94
6.. 14
5.92
6.31
6.91
7,23
6.61
5.72
5.90
6.91
5,58
5.95
1.02
1.14
1.84
1.27
0,31
1.35
2,05
2.60
0.21
0.44
0.05
1.29
0.47
1.47
0.10
0.97
1.63
1,81
0.18
0,15
1.73
1.56
0.63
0.64
0.99
0.30
1.11
0.19
0.34
0.98
1.63
1.72
1.65
0.95
1.09
0,26
0.78
2.76
1.83
0.66
0.4S
0.04
1.61
0.21
1.36
1.57
0.54
0.18
0.07
1.12
1.91
1,43
0.77
1.47
2.96
1,08
1.38
0,73
2.04
2.36
1.65
1.25
1.00
1.03
1.03
1.S5
2.2S
2.72
2.29
0.79
3.29
2,72
0,77
2,08
2.13
2.03
1.76
1.52
1.86
1.22
0.83
1.53
1.90
2.07
2,60
1,70
2.62
2.66
2.54
1,83
0.49
2.75
1.93
2.64
1 .37
0.72
2.08
2.02
2.04
1.17
1.07
0.70
28
.55
55
0.85
2.74
2.06
15-42
-------
TABLE 15-10. FEATURES OF THE MODELS FOR PREDICTING
DISTANCE FROM HAZARDOUS WASTE SITES
Name
Model
Marginal
price
Bias
correction
P1
Double-log
MP1
No
D2
Double-log
MP2
No
D3
Semi-log
MP1
No
D4
Semi-fog
MP2
No
D5
Double-log
MP1
Yes
D8
Double-log
MPZ
Yes
D7
Semi-log
MP1
Yes
D8
Semi-log
MPZ
Yes
15-43
-------
cases than ±50 percent of the actual distances, this is not a particularly good
performance pattern.* This conclusion would not have been as apparent from
simple comparisons of,the overall average predictions with the average actual
distances. These results are reported in Table 15-11. However, student-t
tests of the equality of the means, under the assumption of independence of
the two variables and equality of their variances, suggest that all but one
(D5—the double-log model using the predicted price and the bias adjustment)
reject the null hypothesis of equality of means.
These findings are to some extent qualified by considering the movements
in the actual distances in comparison to the predictions across towns, This is
accomplished using a simple regression approach originally proposed by Theil
[1961] for evaluating forecast accuracy. It involves regressing the actual val-
ues of the distance on each of the predictions and testing two null hypothe-
ses--that the intercept is zero arid the slope unity. In other words, the
points are assumed to cluster around a 45° line when actual distances are
plotted as a function of the predicted distances. Table 15-12 reports these
results for the eight approaches considered throughout our analysis. For all
of the models that used the predicted housing price to estimate the marginal
price of distance, both of these hypotheses are rejected. By contrast, the
actual housing price used in these calculations does not allow a rejection of
one of the null hypotheses. That is, the results seem to indicate that the
actual and predicted values cannot be argued to diverge from a 45° line based
on the slope parameter, but clearly exhibit a constant displacement of the
intercept. These findings must be interpreted cautiously for several reasons.
Failure to reject the null hypothesis of unity for the slope parameter is not a
strong conclusion when the slope parameter is imprecisely estimated, it can
also be interpreted as an indication of no association between the actual and
predicted values of the distance.
The results using the models based on predicted housing price should be
interpreted carefully for another reason. They include the actual distance on
both sides of the equation. That is, distance influences the predicted housing
~The ±50 percent was selected to parallel the proposal of the Cumrnings,
Brooks hi re, and Schulze [1984] methodology for assessing the reference accu-
racy of contingent valuation estimates.
15-44
-------
TABLE 15-11
OVERALL MEANS FOR PREDICTED AND
ACTUAL DISTANCES
Distance to nearest
industrial plant with Standard deviation
hazardous wastes
Mean
t
of the mean
D1
5.83
2.74
.097
D2
6.19
2.11
.074
D3
1.02
11.11
.094
D4
1.79
9.77
.097
D5
8.22
-1.41
.136
D6
8.72
-2 29
,104
D7
1.43
10.28
.131
D3
2.52
8.33
.136
b
Actual distance selected
7.40
.566
3
t-ratio for null hypothesis of equality of means under the assumption of
equality of variances.
This distance is the average of the minimum distances from hazardous waste
sites based on the housing sales in the Harrison data. j
15-45
-------
TABLE 15-12, A COMPARISON OF ACTUAL AND PREDICTED DISTANCES3
Models
1 1 lucpcl IUCI l L
variables
D1
D2
D3
D4
D5
D6
D7
D8
1 ntercept
23
.03
6,51
12,40
6.96
23,03
6.51
12.40
6.96
(5.
¦ 42)
(0.39)
(20.66)
(4.45)
(5,42)
(0.99)
(20.66)
(4.45)
Prediction (D.)
-2,
.68
0.15
-4.92
0.25
-1.90
0.10
-3.50
0.18
i
( ~3.
.70)
(0.14)
(-10.03)
(0.30)
(-3.70)
(.14)
(-10.03)
(0.30)
t b
li
-5,
.09
-0,31
-12.06
-0.93
-5.65
-1.19
-12.89
-1.42
R2
0.
209
0.00
0.66
0.00
0.21
0..00
0,66
0.00
F
13.
72
0,02
100.51
0.09
13.72
0.02
100.51
0.09
3
^ The numbers in parentheses below the estimated coefficients are t-ratios for the null hypothesis of
no association.
Oi
is the t-ratio for testing the null hypothesis that the slope parameter is unity.
-------
price and therefore is present in a nonlinear relationship determining the (pre-
dicted distance. There is no way to avoid this outcome if the predicted price
is to form the basis for the estimates of the marginal price for distance. There
are, however, other factors changing across towns, so it is not ensured'that
the models based on predicted price would riot yield a tautological relationship
for the actual-predicted regression models.
While the overall findings of this analysis at this stage indicate inconsis-
tency between the results of the two models, this does not necessarily imply
that either of them is incorrect. It can easily be a reflection of the differing
assumptions underlying each framework,*
*lt should also be noted that the demand-for-di stance models could be
modified to estimate the implicit price for distance and compare implicit values.
We also performed these comparisons for both models. The results were com-
pletely implausible with the double-log specification and therefore will not be
reported here. With the semi-log model, the findings were more plausible,
but not in close correspondence with the estimated marginal prices from the
hedonic model. For example, the overall averages across the 54 towns for the
marginal prices MP1 and MP2 are given below in comparison with the results
of inverting the demand for distance function and predicting the constant mar-
ginal price (designated as VAL),
Estimated of
Marginal Value
MP1
MP2
VAL
Mean
9,498.61
8,860.34'
2,354.82
Standard
Deviation
3,692.87
1,938.39
2,212.65
-12.21
-10.76
Allowing VAL to play the role of the actual marginal price, regressions compar-
ing VAL and the marginal prices derived from the hedonic models are as fol-
lows:
VAL
VAL =
5,868.85 -0.370 MP1
(8.835) (- 5.66)
(-15.31)
581.92
(0.527)
+0.269 MP2
( 1.75)
(-4.75)
R2 = 0.381
F =32.06
R2 = 0.058
F =3.06
The results of tests of equality of means (given in the column labeled t)
cate that the null hypothesis must be rejected at the .01 significance le
The regression analysis is also clear in indicating incompatibility between
ndi-
yel.
each
15-47
-------
ljs.11 SUMMARY
It must be acknowledged that a large number of assumptions and judg-
j
ments were required to complete this comparison. They have been enumerated
throughout this chapter, but a few of the more important considerations are
reiterated here. One of the most important arises with the property value
model. It was acknowledged to be an early version of the hedonic function
initially developed by Harrison [ 1983J and revised in Harrison and Stock
[1984]. Nonetheless., the specification used was a plausible formulation for
the model and was not obviously inferior to the revised model. Further evalu-
ation of the two property value models is necessary before a clear choice can
be made.
A second consideration concerns the compromises required in the repre-
sentation of the nature of the constraints facing the individual. One of the
most important of these involved treating the marginal price of distance as con-
stant. Most hedonic property value models do not make this assumption. In-
deed, it is inconsistent with both versions of the Harrison property value
model. However, there was no simple way to avoid this problem. Communica-
tion of the demand for distance question in the survey required a direct and
simple explanation of the price of distance. It was felt that presentation of a
function or price schedule would have decreased the chances for successfully
administering the question and lengthened an interview that was felt to be too
long already.
Despite these qualifications and limitations, the research has highlighted
a number of dimensions of comparative analyses that were implicit, maintained
assumptions in past studies. Equally important, it has suggested an alterna-
tive means of judging the compatibility of two models. Rather than comparing
each approach's estimate of an unknown marginal valuation, we used the models
together to predict an observable response that could then be used to evaluate
the methods' compatibility by gauging the relationship of these predictions with
of the method's estimates of the marginal valuation of distance. The statistic
in parentheses immediately below the estimated parameters is- the t-ratio for
the null hypothesis of no association. Below this statistic for the case of the
slope parameters is the t-ratio for the null hypothesis that each slope parame-
ter is unity.
15-48
-------
actual responses. Of course, this procedure does not allow for the identifica-
tion of which method is "at fault,11 However, it does provide an alternative
standard for judging compatibility and a clearer criteria for what that compati-
bility might mean by allowing one to consider the performance of these types
of predictions in relationship to other types of predictive performance of other
economic models,
This procedure demonstrates the use of a reference group or relative
standard in judging the compatibility of benefit estimation methods. Past com-
parisons have focused on each method's estimates of an economic valuation con-
cept that can never be known--e. g., the true willingness to pay for reductions
in air pollution or reductions in the risk of exposure to hazardous wastes.
Therefore, the precision (in conventional terms) of these methods' estimates
cannot be gauged with real-world data. We have suggested that since each
method describes a choice and valuation process, the methods may be combined
to predict some economic outcome that can be observed. These predictions can
then be compared with actual choices.
15-49
-------
-------
CHAPTER 16
POLICY IMPLICATIONS AND RESEARCH AGENDA
16.1 INTRODUCTION
This chapter has two objectives—first, to consider the policy implications
of the valuation estimates derived from the contingent valuation survey and,
second, to outline the further research that follows from the activities under-
taken during Phase I of the project. Our discussion to this point has deliber-
ately avoided consideration of both the potential policy uses for these estimates
and a comparison of them with earlier efforts to value risk changes. There are
several reasons for holding the discussion of these issues until the end of the
sized
ary.
report. The most important of these arises from the caveat empha
throughout the presentation of our empirical results: They are prelimir
The data have been used for detailed econometric and comparative analyses to
begin the process of developing finai estimates. These efforts have identified
a large number of assumptions required to use the sample responses in a par-
ticular task. In many cases it will be necessary to evaluate the implications of
alternative sets of assumptions and to refine the econometric techniques used
with these assumptions and the data before a final set of results can be devel-
oped. Indeed, the preliminary statistical analyses have highlighted problems
that would not have been apparent from the summary statistics used to describe
the survey results. In the analysis of the contingent valuation results, for
example, the problems posed by missing values for key attitudinal and risk
perception variables were identified as important to the development of models
reflecting these influences. Equally important, heteroscedasticity in the errors
associated with the respondents' valuations for the two risk reductions appeared
to be an especially important consideration for further econometric modeling of
these marginal valuations. Finally, treatment of the zero bids and the truncated
nature of the distribution of valuation responses will require further analysis.
18-1
-------
; It should be acknowledged that these issues are important to both the
econometric modeling of valuation responses and the development of summary
statistics to describe the distribution of these responses. They cannot be
avoided by focusing exclusively on the summary statistics for the valuation
responses.
A second reason for postponing discussion of policy interpretations stems
¦om the objectives of the research:
To develop a set of estimates of the individuals' valuations for
risk changes in a format that allowed analysis of the factors
influencing these valuation decisions.
To compare the contingent valuation and hedonic approaches
for describing the behavioral responses to {and the valuations
of) changes in the risks of exposure to hazardous wastes.
These objectives relate primarily to technical issues associated with the model-
ing and measurement of the values for risk changes. They do not extend to
he tasks associated with using valuation information in decisionmaking with
respect to the specific policy actions that are associated with hazardous wastes
and would be expected to lead to changes in the risk of exposure to these
substances. Accordingly, this chapter's discussion of policy issues addresses
only one issue: "How do these valuation estimates compare with those fre-
quently in use for policy decisions involving risk change?"
16.2 GUIDE TO THE CHAPTER
Section 16.3 of this chapter offers a comparative assessment of our valua-
tion results. Section 16.4 considers the research issues that have emerged
from our preliminary analysis. In contrast to the work undertaken under
Phase I, most of these tasks are empirical. Where further conceptual analysis
is required, it is motivated by the empirical research. Finally, Section 16.5
contains a brief summary of the chapter.
16.3 THE INCREMENTAL VALUES FOR RISK CHANGES USED IN POLICY
| ANALYSES
One of the most obvious questions that might be posed in interpreting
our results is, "How do they compare with estimates of the values for 'statis-
16-2
-------
tical lives1?"* While we will attempt to discuss this issue in what follows, it
is important to recognize a larger number of qualifications that must be raised
with any such comparison.
Most of the frequently cited estimates of the values for statistical! lives
are the result of empirical models for labor market compensation. In [these
studies, wage rates (or earnings) are specified as a function of the individual
worker's characteristics and his job characteristics.! One of these job charac-
teristics is a risk measure. The hedonic wage model is assumed to describe
the compensation (in higher wages) that is required in equilibrium (based on
the existing distribution of individuals and their respective attitudes toward
risk and jobs) to have the marginal individual accept an increment to his risk
of injury or death on the job. While there have also been studies of other
individual or household decisions involving risk (see Viofette and Chestnut
[1983] for a review), these will riot be considered here.
Table 16-1 summarizes the highlights from four frequently cited studies
of labor market job risk decisions. The values for statistical lives derived
from these studies range from $630,000 (in 1984 dollars) in the Thaler-Rosen
[1975] analysis to $6,300,000 (in the same units) in the Viscusi [1981] analy-
sis. The use of the term "value of a statistical life" to describe these incre-
mental values is in many respects unfortunate. It is simply an index of the
rate per unit of risk at which workers are compensated. It does not imply
the acceptance of any of the estimates as "the value of a life." Rather, there
are two equivalent ways of interpreting the index: (1) the scaling of a mar-
ginal risk premium by the level of risk to derive the dollar value per unit of
*Most estimates of the value of a statistical life come from the aggregate
willingness to pay by many people for small reductions in their own small risks
of death. As a result of the aggregation, the willingness to pay to save one
statistical life refers to that for some person who cannot be identified. Thus,
these values may differ from those expressed for the lives of individuals who
can be identified (e.g., trapped coal miners).
THedonic wage models have also been used in the valuation of environmen-
tal amenities such as air quality (see Bartik and Smith [forthcoming] for a
discussion of these models). These analyses would imply that, for wage func-
tions estimated with a cross-section of workers extending outside a single geo-
graphic area, site attributes ought to be an additional determinant of real wag-
ers, See Smith [1983] for further discussion.
16-3
-------
TABLE 16-1. LABOR MARKET ESTIMATES OF VALUE
OF UNIT OF RISK REDUCTION
Study
Risk measure
Mean
fatality
rate
Implicit value
(1984 dollars)3
Thaler-Rosen [19751
Smith 11976]
Viscusi [1979]
iscust [1381]
Occupational death 0.001
rate
BLS Industry 0,0001
death rate
BLS industry fatal 0.0001
and nonfatal risk
BLS industry fatal 0,0001
and nonfatal lost
workday risk
$830,000
$3.5 million
$3.2. to $4,2 million
$6.3 million
Source: Viscusi [1383] .
3
These values were converted from the estimates reported in Viscusi, using
the consumer price index.
16-4
-------
risk or (2) treating the value as the sum of the marginal values for a specified
risk reduction over a group large enough to yield a reduction in accidental
deaths by one life in expected value terms. In either case, the wage premiums
are based on a specific institutional mechanism for compensating workers to
induce them to assume risk. Since there are constant payments irrespective
of the state of nature realized; this mechanism is an option price format. (See
Smith {1385] for a discussion of this point in relation to hedonic property value
models.) Equally important, the risks involved are annual risks of death from
an on-the-job accident. Thus, in equilibrium, the wage premium is an ex ante
measure of the incremental option price. :
Our contingent valuation responses are also ex ante measures designed
to correspond to an option price. Nonetheless, there are some important dif-
ferences between our results and those of the earlier studies. First, the per-
ceived mechanisms for adjustment may well be different in the two cases. [ For
the worker, there may be opportunities for compensation of families after a
fatal accident affects one of its members (e.g., subsidized company insurance,
benefit programs through unions, etc.). In the scenarios posed as part of
the contingent valuation analysis, these mechanisms are assumed to be non-
existent in the description of the question, but this format does not in itself
necessarily prevent the respondents from adjusting their reported valuations
to reflect their own perceived opportunities for state dependent adjustments.
Second, the analysis from labor market studies is conditional upon a selec-
tion process that has matched individuals willing to accept greater risk with
those providing it. Consequently, the marginal valuation of a risk change
will be sensitive to the composition of the sample used to estimate the wage
model that forms the bases for the estimates,* This opportunity was not avail-
able to the respondents to our survey. While it is possible that those individ-
uals least willing to- accept risk will have moved from locations with the highest
perceived risk, this is not relevant to our contingent valuation questions. In
these questions, risk levels are posed to individuals, It is, of course, ton-
*lt will also be sensitive to the information on risk available to individuals
and how the job risks are perceived. Some preliminary evidence on this issue
is presented in Viscusi [1979]. More recent findings are in Viscusi anc
O'Connor [1984].
16-5
-------
ceivable that an extreme aversion to risk might lead to rejection of the hypo-
thetical scenario. However, our analysis of protest bids did not indicate this
type of response,
1 Third, the job risk measure is a risk of death as if an accident on the
job was always fatal. White this view is not correct, the wage models have
not attempted to describe the decision process as a multi-stage tottery. Both
Viscusi [1381J and Smith [1983] have incorporated nonfatal accident rates, but
neither study has attempted to explicitly describe the process in the same form
as the description of the risks presented in our contingent valuation analysis.
The question requesting values for a risk reduction does identify the process
as a multi-stage lottery in which the event exposure must first be realized
before the risk of the health effect (death) must be considered. Depending
upon how Job risks are perceived, this is a potentially important distinction.
Psychological research has suggested that individuals can have difficulty in
dealing with multi-stage lotteries (e.g., Schum [19801.)
Finally and perhaps most Importantly, the job risk is an annual risk for
any empirical analysis. Once an individual leaves the work environment each
day, the risk of an on-the-job accident is eliminated for that day. By con-
trast, the risk of exposure to hazardous wastes in a given location is contin-
uous—always present so long as the individual is present at the location.
Moreover, the outcome is also different. In the on-the-job accident, the out-
come is a fatality at the time of the event. With the contingent valuation, the
outcome is a fatality 30 years after the exposure. As a consequence, time
plays a very different role in the two cases.
With these qualifications, Table 16-2 reports the calculated values per
unit of risk in a format comparable to the values for statistical lives reported
in Table 16-1. They are based on the mean valuation responses from the sam-
ple with protest bids excluded (see Table 11-7). Appendix H contains a table
with identical information that includes protest bids. Three different calcula-
tions are reported—the annual incremental value per unit of risk and two dif-
ferent values of what we have designated the annuity value of the risk change.
The annual value is simply the annual bid scaled by the unit change in the
risk of death implied by the reduction in the risk of exposure and the condi-
tional probabilities that were posed for each design point in the contingent
16-6
------- |