United States Office of £PA-23045-83401
Environmental Protection ,_ Policy Analysis /March 1983
Agency - Washington DC 20460
<>EFA A Comparison of Alternative
Approaches for Estimating
Recreation and Related Benefits
of Water Quality Improvement
-------
"The information in this document has
been funded wholly or in part by the
United States Environmental Protection
Agency under Contract No. 68-01-5838.
It has been subject to the Agency's
peer and administrative review, and it
has been approved for publication as
an EPA document. Mention of trade
names or commercial products does not
constitute endorsement or recommenda-
tion for use."
-------
March 1983
A Comparison of Alternative Approaches
for Estimating Recreation and Related Benefits
of Water Quality Improvements
Prepared for
U.S. Environmental Protection Agency
Economic Analysis Division
Washington, DC 20460
Dr. Ann Fisher, Project Officer
Prepared by
Dr. William H. Desvousges
Research Triansle Institute
Research Triansle Park, NC 27709
Dr. V. Kerry Smith
University of North Carolina
Chapel Hill, NC 27514
and
Matthew P. McGivney
Research Triansle Institute
Research Triansle Park, NC 27709
EPA Contract No. 68-01-5838
-------
PREFACE
This research project was initiated and supported underwork agreement 68-01-5838 by the
Benefits Staff in the Office of Policy Analysis at the U.S. Environmental Protection Agency (EPA).
Throughout this research effort, the authors of this report were fortunate enough to take
advantage of research activities already in progress. One author had partially completed an
analysis of the problems of defining and measuring option value, for example, and another had
partially completed research to design a generalized travel cost site demand model. In addi-
tion, the authors also benefited from free access to any array of related working papers—many
of which have subsequently been published—that improved the final research design beyond
that possible otherwise. Finally, access to an independently developed estimator for ranked
data improved the authors' ability to make certain types of comparisons for the contingent
ranking component of the survey. Although none of these complementary activities was
contemplated when the project was initially proposed, each has played a substantial role in the
final results. We would not expect these same circumstances to be easily replicated in future
projects of comparable scale and duration.
This final report has been substantially improved through the constructive comments of
many reviewers. In particular we would like to thank Ann Fisher, the EPA project officer, for her
careful commentary and continuous support. In addition, as part of the EPA's review, six other
individuals furnished detailed comments:
Richard Bishop, University of Wisconsin
Rick Freeman, Bowdoin College
Bill Lott, University of Connecticut
Robert Mitchell and Richard Carson, Resources for the Future (RFF)
Bill Schulze, University of Wyoming.
In addition, useful comments were also received from the following individuals:
Tayler Bingham, Research Triangle Institute (RTI)
Peter Caulkins, U.S. EPA
Warren Fisher, U.S. Fish and Wildlife Service
David Gallagher, University of New South Wales
Debbie Gibbs, Bureau of Reclamation
Jerry Hausman, Massachusetts Institute of Technology
Reed Johnson, U.S. Naval Academy and U.S. EPA
John Loomis, U.S. Forest Service
Glenn Morris, RTI
Doug Rae, Charles River Associates
Liz Wilman, RFF.
The authors also have benefited from the comments of participants at presentations given at
RFF; Vanderbilt University,- the University of Missouri-Rolla; Dillon, Colorado (Visual Values
Workshop); Research Triangle Park (Triangle Econometrics Seminar); and Washington, D.C. (EPA).
We are most grateful for all these efforts. Finally, we are most appreciative of the efforts of our
editor, Hall Ashmore, and of Jan Shirley, Supervisor of RTI's Word Processing Center.
in
-------
CONTENTS
Chapter Page
Figures ix
Tables xi
1 Introduction, Objectives, and Summary 1-1
1.1 Introduction 1-1
1.2 Objectives 1-4
1.3 Summary of Results 1-5
1.3.1 Overview 1-5
1.3.2 Contingent Valuation Approach 1-5
1.3.3 Travel Cost Approach 1-7
1.3.4 Approach Comparison 1-8
1.3.5 Considerations for Future Research 1-11
1.4 Guide to the Report 1-15
2 A Brief Review of the Conceptual Basis for the Benefit
Estimation Approaches 2-1
2.1 Introduction 2-1
2.2 A Brief Review of the Conventional Theory of
Benefits Measurement 2-2
2.3 A Framework for Comparing Alternative Benefit
Measurement Approaches 2-9
2.4 The Nature of the Benefits Measured in the
Alternative Approaches 2-12
2.4.1 Travel Cost Approach 2-12
2.4.2 Contingent Valuation Approach 2-13
2.4.3 Contingent Ranking Approach 2-14
2.5 Summary 2-14
3 Survey Design 3-1
3.1 Introduction 3-1
3.2 General Description of the Monongahela River Basin . . . 3-1
3.2.1 Geography 3-1
3.2.2 Uses 3-3
3.2.3 Recreation 3-3
3.2.4 Socioeconomic Profile 3-4
3.3 Sampling Plan 3-5
3.3.1 Target Population 3-5
3.3.2 Sample Selection and Survey Design 3-6
3.3.3 Sampling Weights 3-6
3.4 Survey Plan 3-6
3.4.1 Questionnaire Design and Limited Local Pretest . . 3-7
3.4.2 Retaining Field Supervisors and Hiring
Interviewers 3-9
-------
CONTENTS (continued)
Chapter
3.4.3 Counting and Listing of Sample Segments 3"9
3.4.4 Developing Field Manuals and Conducting
Interviewer Training r~™
3.4.5 Training Session • • • • • |-11
3.4.6 Conducting Household Interviews o 10
3.4.7 Initial Contacts and Obtaining Cooperation .... 3-13
3.4.8 Household Enumeration A
3.4.9 Interviewing Procedures f
3.4.10 Interviewer Debriefing 3-16
3.4.11 Data Receipt, Editing, and Keypunching 3-18
4 Contingent Valuation Design and Results: Option Price
and User Values • • 4-1
4.1 Introduction 4-1
4.2 A Review of Design Issues in Contingent Valuation
Surveys 4-2
4.2.1 Hypothetical Bias • 4-2
4.2.2 Strategic Bias 4-4
4.2.3 Payment Vehicle Bias 4-6
4.2.4 Starting Point Bias . . 4-6
4.2.5 Information Bias 4-7
4.2.6 Interviewer Bias 4-7
4.2.7 Summary and Implications for Contingent
Valuation Research Design 4-9
4.3 Questionnaire Design 4-9
4.3.1 Questionnaire Design: Part A 4-9
4.3.2 Benefits Measures: Part B 4-11
4.4 Profiles of Survey Respondents 4-20
4.5 Option Price Results 4-27
4.6 User Value Results 4-36
4.7 Summary 4-38
5 Contingent Valuation Design and Results: Option and
Existence Values 5-1
5.1 Introduction 5-1
5.2 Contingent Claims Markets and the Modeling of
Uncertainty 5-3
5.3 Option Value: The "Timeless" Analyses 5-7
5.4 The Time-Sequenced Analyses 5-14
5.5 Recent Estimates of Nonuser Values 5-16
5.6 Measuring Option Value: Survey Design 5-21
5.7 Survey Results—Option Value 5-25
5.7.1 Option Value—Demand Uncertainty 5-25
5.7.2 Option Value—Supply Uncertainty 5-29
5.8 Existence Value Estimates 5-31
5.9 Summary 5-33
VI
-------
CONTENTS (continued)
Chapter
6 Contingent Ranking Design and Results: Option Prices. ... 6-1
6.1 Introduction . . . . 7 .... 6-1
6.2 Consumer Behavior and the Contingent Ranking
Framework 6-2
6.3 Estimation of Random Utility Models with Ordered
Alternatives 6-7
6.4 Past Applications of Contingent Ranking 6-9
6.5 Monongahela Contingent Ranking Experiment:
Design and Estimates 6-16
6.6 Benefit Estimates with Contingent Ranking Models .... 6-25
6.7 Implications and Further Research 6-28
7 A Generalized Travel Cost Model for Measuring the
Recreation Benefits of Water Quality Improvements 7-1
7.1 Introduction 7-1
7.2 Travel Cost Model 7-2
7.3 The Travel Cost Model for Heterogeneous
Recreation Sites 7-10
7.4 Sources of Data 7-22
7.5 Empirical Results for Site-Specific Travel Cost Models . . 7-30
7.5.1 The Treatment of Onsite Time 7-31
7.5.2 The Opportunity Cost of Travel Time 7-32
7.5.3 Results for the Basic Model 7-32
7.5.4 Results for the Tailored Models 7-36
7.5.5 Evaluation of Measures of the Opportunity Cost
of Travel Time 7-38
7.6 Further Evaluation of the Travel Cost Models 7-43
7.7 Analyzing the Role of Water Quality for Recreation
Demand 7-51
7.8 A Measure of the Benefits of a Water Quality
Change 7-57
7.9 Summary 7-64
8 A Comparison of the Alternative Approaches for Estimating
Recreation and Related Benefits 8-1
8.1 Introduction 8-1
8.2 The Conceptual Framework for a Comparison of
Recreation Benefit Estimation Approaches 8-2
8.2.1 Background 8-2
8.2.2 Research Design and Comparative Analysis .... 8-3
8.2.3 Past Comparisons of Benefit Estimation
Methods 8-9
8.3 A Comparative Evaluation of the Contingent Valuation,
Travel Cost, and Contingent Ranking Benefit
Estimation Methods 8-12
8.4 Implications 8-20
VII
-------
CONTENTS (continued)
Chapter
9 References
Appendixes
A Sample Design ......... „ . ............. A"1
B Survey Forms and Procedures ................ B~1
Part 1 --Household Control Form. .............. B~1
Part 2--Counting and Listing Examples ..... . ..... B~*
Part 3--Debriefing Agenda ................. B"7
Part 4--Quality Control Procedures ..... ........ B~9
C Survey Analysis: Supporting Tables ............. c~1
D Survey Questionnaires ......... . . ....... - • D~1
Part 1 --Survey Questionnaire ............... D'2
Part 2--Suggestions for Improving the Questionnaire for
Future Use ........................ D-28
E Technical Water Quality Measures: An Economist's
Perspective ......................... E-1
F Travel Cost: Supporting Tables ............... F~1
G Alternative Regression Models ................ G-1
viii
-------
FIGURES
Number Page
1-1 Effects and responses to water quality regulatory actions . . 1-2
1-2 A spectrum of water quality benefits 1-3
2-1 The demand function and the consumer surplus welfare
measure 2-3
2-2 A comparison of alternative welfare measures . 2-5
2-3 Surplus measures for a change in quantity 2-6
2-4 Smith-Krutilla framework for classifying the measurement
bases and approaches of economic benefits resulting from
improved water quality 2-10
2-5 Travel cost demand function with water quality
improvement 2-12
3-1 Map of Monongahela River and other area
recreation sites 3-2
3-2 Geographic location of survey area 3-5
3-3 Field interviewer training session agenda 3-12
3-4 Summary of completed interviews 3-13
4-1 Activity card 4-10
-4-2 Site activity matrix 4-10
4-3 Map of Monongahela River and other area recreation sites. . . 4-12
4-4 Recreation sites 4-12
4-5 Water quality ladder 4-13
4-6 Value card 4-14
4-7 Payment card 4-17
4-8 Rank order card 4-20
5-1 Optimal allocation of choice with contingent claims 5-6
5-2 Optimal allocation of choices of contingent claims
without uniqueness 5-7
5-3 Option value in Cicchetti-Freeman analysis 5-12
5-4 Option value in Cicchetti-Freeman with "no demand" 5-12
5-5 Option value with contingent claims in Graham's analysis . . . 5-13
6-1 Rank order card 6-18
7-1 Measurement of consumer surplus increment due to
water quality improvement 7-59
IX
-------
TABLES
Number
1-9
1-1 A Comparison of Mean Benefit Estimates
1-2 Regression Comparisons of Contingent Valuation and
Travel Cost Benefit Estimates "~n
2-1 Alternative Welfare Measures and Types of Consumer
Surplus Measures for Contingent Valuation Studies 2~'
3-1 Questionnaire Development Activity ^"°
3-2 Final Distribution of Sample Housing Units 3-15
4-1 Summary of Biases in Contingent Valuation Experiments. . . . 4-8
4-2 Summary of Option Price Question Formats by
Interview Type 4-18
4-3 Characteristics of Key Respondent Groups 4-21
4-4 Reasons for Zero Bids by Elicitation Method 4-23
4-5 Degree of Importance of Water Quality by Key Respondent
Groups 4-24
4-6 Respondent Attitudes About Self by Key Respondent Groups . 4-25
4-7 Logit Estimation of Zero Bids 4-26
4-8 Profile of Outliers 4-30
4-9 Estimated Option Price for Changes in Water Quality:
Effects of Instrument and Type of Respondent--Protest
Bids and Outliers Excluded 4-32
4-10 Student t-Test Results for Option Price—Protest Bids
and Outliers Excluded 4-33
4-11 Regression Results for Option Price Estimates—Protest
Bids and Outliers Excluded 4-34
4-12 Student t-Test Results for Option Price—Protest Bids
and Outliers Excluded 4-36
4-13 Estimated User Values—Protest Bids and Outliers Excluded . . 4-37
4-14 Regression Results for User Value Estimates of Water
QuaJity Changes—Protest Bids and Outliers Excluded 4-38
5-1 Summary of Mitchell-Carson Estimated Mean Annual
Willingness to Pay by Version and Water Quality 5-19
5-2 Summary of Willingness-to-Pay Questions by Type
of Interview 5-24
5-3 Summary of User, Supply Uncertainty, and Existence
Value Questions 5-24
5-4 Estimated Option Values for Water Quality Change: Effects
of Instrument and Type of Respondent—Protest Bids
and Outliers Excluded 5-26
5-5 Student t-Test Results for Question Format 5-27
XI
-------
TABLES (continued)
Number Page
5-6 Regression Results for Option Value Estimates—Protest Bids
and Outliers Excluded . 5-28
5-7 Effects of Supply Uncertainty on Option Price 5-30
5-8 Student t-Tests for the Effects of Supply
Uncertainty for Users 5-31
5-9 Estimated Existence Values 5-32
6-1 Summary of Rae/CRA Contingent Ranking Studies 6-11
6-2 Combinations of Water Quality and Payments for Monongahela
Contingent Ranking Survey 6-18
6-3 Selected Results for the Random Utility Model With
Ranked Logit Estimator 6-21
6-4 Comparison of Ordered Logit and Keener-Waldman
Ordered Normal NIL Estimator 6-24
6-5 Benefit Estimates from Contingent Ranking Models 6-27
7-1 Hedonic Wage Models 7-26
7-2 Summary of Predicted Hourly Wage Rates 7-27
7-3 The Characteristics of the Sites and the Survey
Respondents Selected from the Federal Estate Survey 7-28
7-4 Regression Results of General Model, by Site 7-33
7-5 Summary of Cicchetti, Seneca, and Davidson [1969]
Participation Models 7-36
7-6 Comparison of Basic Model with Tailored Model:
Coefficient for (TC+MC) 7-37
7-7 ' F-Test for Restriction of General Model 7-39
7-8 F-Test for Restriction of Tailored Models 7-42
7-9 Effects of Truncation on the Travel Cost Models' Estimates . . 7-45
7-10 Two-Stage Least-Squares Estimates for Selected Travel
Cost Site Demand Models 7-48
7-11 Comparison of Ordinary Least-Squares and
Two-Stage Least-Squares Estimates of Travel
Cost (TC. + MC.) Parameters 7-49
7-12 Hausman Test for Differences Between Two-Stage
Least-Squares and Ordinary Least-Squares Estimates 7-50
7-13 Description of U.S. Army Corps of Engineers Data on
Site Characteristics 7-53
7-14 Generalized Least-Squares Estimates of Determinants of
Site Demand Parameters 7-56
7-15 Recreation Sites on the Monongahela River 7-57
7-16 Dissolved Oxygen Levels for Recreation Activities 7-60
7-17 Mean and Range of Benefit Estimates for Water Quality
Scenarios 7-61
7-18 Consumer Surplus Loss Due to the Loss of Use of the
Monongahela River by Survey Users' Income 7-62
7-19 Consumer Surplus Loss Due to Loss of Use of the
Monongahela River by Survey Users' Travel Cost 7-62
xii
-------
TABLES (continued)
Number
7-20 Consumer Surplus Increments Due to Water Quality
Improvement—Beatable to Fishable by Survey Users'
Income „ 7-63
7-21 Consumer Surplus Increment Due to Water Quality
Improvement—Boatable to Swimmable by Survey Users'
Income 7-63
8-1 Predicted Demand Parameters for Monongahela Sites 8-8
8-2 Bishop-Heberlein Comparative Results for Benefit
Approaches 8-10
8-3 A Comparison of Benefit Estimates for Water Quality
Improvements 8-13
8-4 A Comparison of Contingent Valuation and Generalized
Travel Cost Benefit Estimates 8-16
8-5 A Comparison of Contingent Valuation and Contingent
Ranking Benefit Estimates 8-19
xiii
-------
CHAPTER 1
INTRODUCTION, OBJECTIVES, AND SUMMARY
1.1 INTRODUCTION
This Research Triangle Institute (RTI) report to the U.S. Environmental
Protection Agency (EPA) compares alternative approaches for estimating the
recreation and related benefits of water quality improvements. The results
provide information on the performance of various ways to estimate the benefits
of environmental quality improvements, so EPA can use such methods in pre-
paring the regulatory impact analyses required by Executive Order 12291 and
in evaluating other regulatory proposals. This report is also relevant to the
proposed revision of the Federal water quality standards regulations, which
recommends that States consider incremental benefits and costs in setting their
water quality standards. Site-specific water quality standards are likely to
play an important role in future water policy issues because they bring togeth-
er the crucial elements of appropriate stream uses and advanced treatment re-
quirements for municipalities and industries. Benefit-cost assessments can
yield valuable information for these decisions.
Evaluations of benefits and costs depend on a determination of the links
between regulatory policy, technical effects, and behavioral responses. Fig-
ure 1-1 illustrates one set of linkages--in this case for the proposed water
quality standards regulations. This report addresses the last component of
Figure 1-1, which involves estimating monetized benefits for regulatory policy.
One of the difficulties in such a task arises from the absence of organized
markets for many of the services derived from water resources.
The benefits of water resource regulations are usually measured with one
of three types of approaches: (1) market-based approaches, which use indi-
rect linkages between the environmental goods and some commodities exchanged
in markets; (2) contingent valuation approaches, which establish an institu-
tional framework for a hypothetical market; and (3) public referenda. This
report considers the first two approaches; the last is omitted since it is beyond
EPA's mandate.
Some opponents argue that benefit-cost analysis is invalid because it can-
not measure all of the benefits of environmental regulations. Nevertheless,
this report describes the measurement of several benefits from water quality
improvements, including some regarded as unmeasurable in earlier environmen-
tal benefits research efforts. Specifically, as highlighted in Figure 1-2, this
study considers both the recreation benefits that accrue to users of a recrea-
tion site and the intrinsic benefits* that accrue to both users and nonusers.
*This classification modifies the one in Mitchell and Carson [1981]
1-1
-------
Water Quality
Regulatory ActionU)"
Change Designated U«e(f)
Modify Criteria to Provide
for Designated Uie(s)
Technical Effects
of Water Quality —
Regulatory ActionU)
Changes in Effluents
Changes in Water Quality
Change in Ecological
Habitat
Effects on Economic
Agents
Behavioral Effects
of Water Quality —
Regulatory Action(s)
Behavioral Responses
of Economic Agents
Figure 1-1. Effects and responses to water quality regulatory actions.
User benefits arise from recreation uses of the river and are measured
by users' willingness to pay for the water quality levels necessary to permit
these recreation uses. That is, the valuation depends on the use of the river.
In this case, clean water in a river is worth something because recreation!sts
are going to fish, boat, swim in, or picnic along the river.
Intrinsic benefits consist of two value types: option value and existence
value. Relevant to both current users and potential future users, option value
is the amount an individual would be willing to pay for improved water quality
(over his expected user values) to have the right to use the river in the
future when there is uncertainty either in the river's availability at a particu-
lar quality level or in his use of it (with the river meeting specified water
quality conditions). For example, if an individual might use the river, but is
not sure he will, he may pay some amount each year for the right (or option)
to use it (with the river meeting specified water quality conditions). Under
some conditions, this payment, the option price, will exceed his expected con-
sumer surplus—the value he would derive from anticipated use. This excess--
the amount that the option price exceeds the expected consumer surplus—is
defined as the option Value.
1-2
-------
Potential
Water
Quality
Benefits
Current
1 Icor
Benefits
Intrinsic
Benefits
Direct
Use
Indirect
Use
Potential
Use
No
Use
In Stream —
Withdrawal —
Near Stream —
Option*
Existence*
— Recreational — fishing, swimming, boating,
rafting, etc.
— Commercial — fishing, navigation
— Municipal — drinking water, waste disposal
Agricultural — irrigation
— Industrial/Commercial — cooling, process treatment,
waste disposal, steam generation
— Recreational*— hiking, picnicking, birdwatching,
photography, etc.
Relaxation*- viewing
— Aesthetic*— enhancement of adjoining site amenities
— Near-term potential use
— Long-term potential use
— Stewardship — maintaining a good environment for
everyone to enjoy (including future
family use— bequest)
— Vicarious consumption — enjoyment from the
knowledge that others
are using the resource.
Considered in this project.
Figure 1-2. A spectrum of water quality benefits.
Existence value, on the other hand, is an individual's willingness to pay
for the knowledge that a resource exists. That is, an individual—either a user
or a nonuser—might be willing to pay something to maintain a high level of
water quality at a recreation site in a particular area, even though he will not
use it, so that his children may have future use of the site or simply to know
that the ecosystem at the site will be maintained.
This study's comparison of alternative benefits measurement approaches
estimates user values by the travel cost approach (indirect method), by four
different ways of eliciting option price in a contingent valuation experimental
design (direct method), and by a contingent ranking of water quality outcomes
and option price amounts. The central comparison evaluates whether there
are differences between approaches because "true" values for each of these
types of benefits are unknowns. In addition, since the other methods are not
suitable for measuring them, option and existence values are compared only in
terms of alternative ways for posing the hypothetical questions.
A distinguishing feature of this project is its use of a case study of the
Pennsylvania portion of the Monongahela River as the point of reference for
1-3
-------
both the comparison of approaches and the estimation of option and existence
values. The Monongahela is representative of a number of rivers in the
country, has multiple uses, and has recently been the focus of effluent g^iae-
lines for the iron and steel industry. The survey design for the Monongahela,
calling for a household survey, is a middle ground between the macro approach
for estimating benefits of water pollution controls (see Mitchell and Carson
[1981]) and the user orientation of many micro contingent valuation efforts
(see Schulze, d'Arge, and Brookshire [1981]). The design uses a representa-
tive sample of households for the region and, similar to Mitchell-Carson,
includes both user and intrinsic benefits. It also is a specific application,
considering individuals' willingness to pay for a specific river basin's water
quality.
1.2 OBJECTIVES
The potential implications of this study for water policy dictated clearly
defined objectives and a project design to achieve them. The overall objec-
tive of this project was to conduct a study comparing alternative approaches
for estimating the recreation and related benefits of different water quality
levels. In particular, the study sought to measure user, option, and exist-
ence values for the Pennsylvania segment of the Monongahela River and to
estimate the recreation and related benefits that would be derived from pro-
viding different use classifications (fishable, swimmable, boatable) for this
river segment.
In addition to meeting its own specific objectives, an environmental bene-
fits research project ideally would fit the needs of those involved in the evalu-
ation of public policy questions and the needs of the research community in
general. Since the most important direct use of natural environments is for
water-based recreation (see Freeman [1979a]), this project's general research
area considers one of the primary components of environmental benefits
research. In addition to its water quality orientation, the project is also rele-
vant to two areas Freeman identified for future research:
I think that a major research effort should be made to select an
appropriate area and water bodies for a study, to develop a properly
specified model, and to gather the necessary data. Until such an
effort is made, the practicality of the Clawson-Knetsch [1966]
[travel cost] technique for estimating recreation benefits will remain
an open question, [p. 256]
There should be carefully conducted experiments with the survey
techniques for estimating willingness to pay for reduction in pollu-
tion. These experiments should be coordinated with studies based
on other analytical techniques in an effort to provide a cross-check
or validation of benefit estimates obtained by different approaches.
[p. 265]
1-4
-------
1.3 SUMMARY OF RESULTS
This section summarizes the major findings of the research. The findings
are presented for individual approaches and for the comparison between ap-
proaches.
1.3.1 Overview
The results of this project strongly support the feasibility of measuring
the recreation and related benefits of water quality improvements. Moreover,
the benefits measurement approaches—several contingent valuation formats and
the travel cost method—show consistent results for comparable changes in
water quality. Indeed, the range of variation is generally less than that ex-
pected in models used to translate the effects of effluents in a water body into
the corresponding water quality parameters. In addition, the results also
clearly show that the intrinsic benefits of water quality improvements—espe-
cially option values—can be measured and that they are a sizable portion--
greater than half—of the total recreation and related benefits total.
1.3.2 Contingent Valuation Approach
Based on the.results of the Monongahela River case study, the general
prognosis is good for the continued use of the contingent valuation approach
to estimate the benefits of water quality improvements. Statistical analysis
using regression methods to evaluate the determinants of the variation in the
option price bids gave little indication that individual interviewers influenced
the results. The consistently plausible signs and magnitudes of key economic
variables suggest that the respondents perceived the survey structure as
realistic and did not experience problems with the hypothetical nature of some
of the questions. These findings were realized despite the fact that the sample
included households whose socioeconomic profile was comparable to demographic
groups that were found to be more difficult respondents in past contingent
valuation surveys. On average, the respondents were older, less educated,
and poorer than those in the most successful contingent valuation studies.
The contingent valuation estimates of the option price for water quality
improvements, which include user and option values, are consistently plausible
across the various analytical approaches, with estimates for the combined water
quality levels ranging from roughly $50 to $120 per year per household sampled
in the Monongahela River basin. Nonetheless, the empirical results do indicate
that the methods used to elicit the willingness-to-pay amount have a statistic-
ally significant effect on the estimates of willingness to pay. For example,
both the direct question with a payment card and the bidding game with a $125
starting point produced higher willingness-to-pay estimates than either the
direct question without an aid or the bidding game with a $25 starting point.
Thus, there is some evidence of starting point bias in the bidding game, but
the statistical analyses are not conclusive. The results comparing the two
bidding game methods as a set (i.e., those with $25 and $125 starting points)
with the nonbidding games (direct question and payment card combined) indi-
cated no differences between these two sets of approaches.
1-5
-------
The findings provide clear support for a positive, statistically significant,
and sizable option value for water quality improvements along the Mpnonganeia
River. The estimated option values for loss of the use of the area in its cur
rent condition (i.e., boatable) range from approximately $15 to $60 peryear
per household, and the option values for improving water quality to a swim-
mable level range from approximately $20 to $45 per year per nousenoia.
Thus, option value is a substantial fraction of the user's option price, ana
the value of this change in water quality generally exceeds user values.
The survey also provided estimates of existence values. Unfortunately,
respondents did not necessarily understand the distinction sought. Many bid
the same amounts as they had earlier on the option price for a comparable
change in water quality. It is not clear whether these responses were delib-
erate or a reflection of" misunderstanding of the questions. Thus, while the
findings suggest that these values are positive and statistically significant,
prudence requires they be interpreted cautiously.
Of course, it should also be acknowledged that the available estimates of
intrinsic values are quite limited. Most can be criticized for problems in the
research design, including possible flaws in the survey. The design of the
Monongahela River study relies on the use of a schematic classification of the
sources of an individual's valuation of the river (i.e., a card showing dif-
ferent types of values) in eliciting a division of user and other benefits.
Because this is the first application of this device, it was not possible to eval-
uate its effectiveness.
In addition to the more widely used bidding game and direct question
formats for contingent valuation experiments, the Monongahela River basin
survey also applied the contingent ranking format. This format requires only
that individuals rank combinations of water quality levels and option prices
and uses a statistical procedure (ranked order logit)* to estimate willingness
to pay. While other contingent valuation formats require that individuals
directly provide willingness to pay, contingent ranking asks them to rank
hypothetical outcomes. In effect, it asks a simpler task of the respondent--
only to rank outcomes—but requires more sophisticated and less direct tech-
niques to estimate the value of the outcomes.
Since use of the contingent ranking format to estimate the benefits of
environmental quality improvements is quite new, the behavioral model under-
lying its estimation procedures is also early in its development. Although this
project provides a description of these underpinnings, its evaluation of the
theoretical properties and practical issues is incomplete. Overall, the findings
of this study suggest that, even though the behavioral models used to derive
benefits estimates with the contingent ranking format were somewhat arbitrary,
the results from the ranking format closely parallel other contingent valuation
estimates.
*ln more technical terms, the procedure uses a specification for the indi-
rect utility function together with a maximum likelihood estimator.
1-6
-------
The mean estimates derived from the contingent ranking format—roughly
$60 annually per household for improving water quality in the Monongahela to
fishable and approxiamtely $50 more annually for improving it to swimmable--
appear larger than those derived with other contingent valuation formats.
However, these differences are not statistically significant. In addition, the
benefit estimates from all contingent valuation formats are comparable across
individuals, with the primary differences between contingent ranking and other
methods stemming from the questioning format used in the other methods.
1.3.3 Travel Cost Approach
This study also developed and used a generalized travel cost model to
predict the recreation benefits of water quality improvements at recreation
sites.
The travel cost model assumes that site features or attributes affect both
an individual's ability to participate in recreation activities at any particular
site and the quality of his recreation experiences at the site. In considering
the demand for a recreation site as a derived demand, the common sense ra-
tionale of the model suggests that a recreation site's features or attributes
will influence-the demand for its services. Since the level of water quality is
a site attribute, a basis is established for relating water changes to shifts in
demand for a recreation site's services.
The generalized model was estimated from data on 43 water-based recre-
ation sites in the Federal Estate Survey component of the 1977 National Outdoor
Recreation Survey. This survey provided information on recreation use pat-
terns at each site during a single season. Based on sample sizes for each
site that ranged from approximately 30 to several hundred respondents, the
survey described individuals' recreation behavior, socioeconomic characteris-
tics, travel time necessary to reach the site, residential location, and a variety
of other factors.
Several advantages of this travel cost model include:
Deriving individual estimates for the time associated with
traveling to the site as well as the roundtrip distance for each
trip.
Using the opportunity cost of time to evaluate travel time and
estimating opportunity cost for each individual based on his
characteristics, including age, education, race, sex, and
occupation.
Considering for each site the potential effects of individuals'
differences in onsite time per visit.
The generalized model was used to estimate the benefits for users of the
Monongahela River, as identified in the survey of the basin. The travel cost
model predicted a value of $83 per year per user household if a decrease in
1-7
-------
water quality is avoided and a value of $15 per year for each user household
if water quality is improved to a swimmable level.
Several features of the generalized travel cost model are of particular
importance: it provides a framework for estimating the value of water quality
improvements for a substantial range of sites, and its site-specific orientation
is especially relevant for water quality standards applications. Finally, it in-
cludes the effect of key site features—like access and facilities—and can use
data frequently available in the public domain.
1.3.4 Approach Comparison
One of the primary objectives of this research has been to compare avail-
able approaches for measuring the benefits of water quality improvement. Such
a comparison—reflecting the assumptions inherent in each approach—will show
the plausibility of the required assumptions as descriptions of real-world be-
havior and constraints. However, since the "true" value of water quality im-
provement benefits is unknown, a comparison cannot be interpreted as a vali-
dation of any one approach. On the other hand, an evaluation of the com-
parability of estimates across approaches that considers the reasons for their
consistencies and differences provides a basis for an improved use of benefit
methodologies. Consistency also would give increased flexibility in matching a
method to available data for each particular application.
Based on the research for the Monongahela River basin case study, the
comparison between the travel cost and contingent valuation approaches is the
most interesting. Estimates of benefits from water quality improvement are
compared for the 69 users identified in the survey of households in the basin
are£. Previous comparisons of approaches relied on the use of mean estimates
from each method. When these means are compared, it is assumed that all
individuals can be treated as drawing from populations with the same mean
benefits. Differences in individuals or error in the pairing of means can lead
to a confounding of the benefit comparisons. In contrast, this study compared
each household's user value, derived from the contingent valuation survey,
with the corresponding estimate for that household from the travel cost model.
Thus, this study gives a more controlled comparison than was possible in
earlier studies.
Table 1-1 shows the mean benefit estimates of user values for the travel
cost and contingent valuation approaches. On theoretical grounds, the contin-
gent valuation estimates of compensating surplus should be less than the travel
cost estimates based on ordinary consumer surplus, but the differences should
be slight due to the small income effects found in the research. However,
this is not the case for three out of four contingent valuation estimates for
improvements in water quality. Only the estimates derived with the $25 bid-
ding game format are less than the travel cost estimates, although the travel
cost estimates are within the range of contingent valuation estimates. For the
loss of the area, the means comparison conforms to theoretical expectations,
with the travel cost estimates larger than the contingent valuation estimates.
Most of the differences between approaches exceed the size expected from
theory. At best, simple comparisons of mean estimates—augmented by a priori
information—are rough judgments of plausibility. On the basis of this compar-
1-8
-------
Table 1-1. A Comparison of Mean Benefit Estimates (1981 Dollars)
Approach
Contingent
Loss of area
Boatable
to fishable
Boatable
to swimmable
valuation
Direct question
Payment card
Iterative
Iterative
Generalized
bidding
bidding
travel
($25)
($125)
cost
19.
19.
6.
36.
82.
71
71
59
25
65
(17)
(17)
(19)
(16)
(94)
21.
30.
4.
20.
7.
18
88
21
31
01
(17)
(17)
(19)
(16)
(94)
31
51
10
48
14
.18
.18
.53
.75
.71
(17)
(17)
(19)
(16)
(94)
aThe travel cost estimates were converted from 1977 to 1981 dollars using the
consumer price index for December 1981, the last month of the survey.
The numbers in parentheses after the means are the number of observations
on which each of these estimates was based. The number for the travel cost
estimates exceeds the sum of the sample size for the contingent valuation
results because some users visited more than one Monongahela River site.
ison, however, the Monongahela River basin estimates are plausible, but not
precise.
A more discriminating comparison of the travel cost and contingent valua-
tion' approaches, one that judges how the two approaches compare across indi-
viduals, is also possible with the Monongahela River basin benefit estimates.
In this comparison, presented in Table 1-2, the contingent valuation measure
of user value was regressed on the travel cost estimate (see Chapter 8 for
details). The a priori expectations of comparability in methods can be struc-
tured as two statistical tests. These models also take account of the effect of
question formats used in the contingent valuation survey.
The results from the regression tests generally reinforce the earlier con-
clusions based on comparing the means estimated from each method. Several
additional conclusions are possible from these comparisons:
The contingent valuation estimates of water quality improve-
ments overstate willingness to pay-in contrast to the theoret-
ical expectations—but the results do not permit a judgment of
statistically significant differences between the two sets of esti-
mates. Some caution is required, however, because the prop-
erties of the statistical tests are approximate.
The travel cost model overstates—by an amount greater than
theory would predict—willingness to pay for the loss of the
area, and the estimates are not comparable to the contingent
valuation estimates.
1-9
-------
Table 1-2. Regression Comparisons of Contingent Valuation and
Travel Cost Benefit Estimates
Water duality change
Independent
variables
1 ntercept
Travel cost-benefit
estimate
Loss of area
21.86
(1.37)
0.33
(-4.36)b
Boatable to
game fishing
33.99
(1.90)
-3.67
(-1.20).
(-1.71)
Boatable
to swimming
59.57
(2.02)
-2.71
(-1.14)
(-1.79)°
Qualitative variables
Payment card
Direct question
Iterative bid ($25)
R2
F
-32.64
(-2.55)
-14.60
(-1.27)
-31.82
(-2.55)
0.10
2.42
(0.05)c
51.76
(2.64)
12.96
(0.75)
-11.24
(-0.60)
0.12
3.00
(0.02)C
77.01
(2.36)
21.00
(0.73)
-21.82
(-0.69)
0.11
2.62
(0.04)C
aThe numbers below the estimated coefficients are t-ratios for the null hypoth-
esis of no association.
These statistics are the t-ratios for the hypothesis equivalent to unity for
the slope coefficient^for Ordinary Consumer Surplus (OCS) after adjustment
is made for the fact that Compensating Surplus (CS) is measured in 1981
dollars and OCS in 1977 dollars.
The number in parentheses below the reported F-statistic is the level of sig-
nificance for rejection of the null hypothesis of no association between the
dependent and independent variables.
The comparative performance of the contingent valuation ap-
proach in relationship to the travel cost method is sensitive to
differences in question format—with the clearest distinctions
found between the payment card and the bidding game with
the $125 starting point.
The explanatory power of the models used in the comparison
are not high, but the null hypothesis of no association between
methods is clearly rejected at high levels of significance (based
on the F-tests reported at the bottom of the table).
1-10
-------
1.3.5 Considerations for Future Research
The findings of this project also suggest that there are a number of areas
for future benefits research, including both general and specific issues—
especially those concerned with particular benefits measurement approaches.
General Issues
Option and existence values remain the most difficult general issues to
address adequately. The research design for this project relied on the indi-
vidual to divide the hypothetical option price payment into its user and option
value components and then to add existence values to these option price bids
as an incremental premium. Other studies (Brookshire, Cummings, et al.
[1982] and Randall, Hoehn, and Tolley [1981]) have elicited preservation
values—including both option and existence values—as additions to user
values. Mitchell and Carson [1981] found user values by subtracting non-
user's option price payments from user option price payments. Regardless of
the procedures, however, all these studies have found option and existence
values to be substantial—greater than half of the total benefits of environ-
mental improvements. The choice among elicitation procedures remains an open
question.
One question that arises from the results of this and other recent studies
of intrinsic benefits is, "Why worry about measuring option value when it is
possible to elicit option price bids that include it?" Empirical estimates are of
interest because of the controversy over the sign and magnitude of option
value that has arisen in the theoretical literature. In addition, many practical
applications of benefit methods do not measure intrinsic benefits, suggesting a
need for empirical estimates to gauge the extent of the omitted portion of
benefits from particular environmental policies. The early theoretical work
seemed to imply (without explicitly stating this conclusion) that option values
would be small in comparison to user values. Recent theoretical work by
Freeman [1982] makes a case for positive option values and confirms this pre-
sumption by suggesting that option values should be small under almost all
conditions. Only by attempting to distinguish between option and other intrin-
sic values will it be possible to bring some empirical evidence to bear on this
question.
Proportional relationships between user and intrinsic values from earlier
studies have often been used in attempts to infer the size of the omitted bene-
fits when the intrinsic values are not directly estimated. The limited resources
available for many public policy evaluations is the primary reason for the wide-
spread use of the proportional approach. Since it is unlikely that these con-
straints on evaluations will ease in the future, more empirical research on the
use and size of these proportions might be productive. For instance, deter-
mining how (and if) the proportions differ for certain classes of assets—
ranging from unique natural environments to waterbodies with numerous
substitutes—would provide useful guidance for applying these proportions.
Moreover, attempting to distinguish between option and existence values for
different classes of environmental assets may indicate the feasibility—and
need—for such distinctions (see Fisher and Raucher [1982] for a review).
1-11
-------
The research in this project has skirted another important issue"be"e™*
aggregation. The travel cost model used in this project predicts recreation
site benefits for "the representative" user. By assuming that all sites are
possible substitutes (because one site's attributes can be "repackaged to oe
equivalent to any other site), it implicitly maintains a simplistic view or ine
relationship between recreation sites within a region. Individuals always select
the site providing the desired mix of attributes at the lowest implicit price.
Clearly, not all sites adhere to these relationships. For example, a mstonca
monument at the site may make it unique. What is needed is a more general
characterization that would accommodate sites not conforming to the aggrega-
tion rule used to relate effective site services to site attributes. Such a
framework would explain the relationship between, an individual's patterns of
site usage for facilities permitting very different types of recreation activities
(e.g., water-based recreation versus skiing). Nevertheless, consistent
regional and national benefit estimates will require a careful description of the
interrelationships between the individual's demands for different types of rec-
reation sites.
Another unresolved issue involves regional aggregation of local benefits
estimated with the contingent valuation approach. Conventional practice in
statistical surveys is to use statistical weights, which reflect the probability
of selecting a particular sampling unit, to estimate aggregate benefits for the
representative population (see Mitchell and Carson [1981]). However, this
approach raises fundamental problems with the conventional practice in eco-
nomic modeling that assumes common (and constant) parameters across indi-
viduals for correctly specified behavioral models. The definition of a repre-
sentative sample is often based on a description of statistical models, leading
to observed data that are at variance with conventional economic modeling.
More research following the work of Porter [1973] is needed to consider the
relevance of this issue for the extrapolation of contingent valuation estimates.
Another general research issue involves comparing alternative benefit
estimation approaches. This project's comparison, which examines benefits
predicted with the generalized travel cost model and contingent valuation will-
ingness-to-pay estimates for the same individuals, permitted a fairly direct
comparison of estimates with theoretical bounds. However, this study used
estimates from only 69 users of the Monongahela River. A comparison having
a larger number of users and based on a water-based recreation site with a
greater diversity of users would provide a more revealing comparison. Indeed,
following Bishop and Heberlein, attempts to compare simulated market results
with the results of this project also may shed light on the relationships among
the estimation approaches. Before these comparisons are made, however, more
systematic attention should be given to the theoretical underpinnings of the
approaches, following the work of Schulze et al. [1981], Smith and Krutilla
[1982], and Bockstael and McConnell [1982].
Future research should also reconsider the economic principles underlying
comparisons of economic welfare—particularly the measurement basis (ordinary
consumer surplus and the more precise Hicksian-based measures). The com-
parisons made in this project have involved expenditures of such a small per-
centage of individuals' budgets that the differences between the measures is
1-12
-------
insignificant. Since some, and perhaps many, environmental issues may in-
volve large price and quantity changes with more significant income effects,
the empirical application of various measures becomes significant. Bockstael
and McConnell [1980] have raised some empirically based issues, but a more
extensive effort such as Willig's [1976], comparing recent approaches proposed
by Hausman [1981], McKenzie and Pearce [1982], and Takayama [1982], may
yield guidance for applications with these large changes.
A final general issue on the research agenda that, unfortunately, was
beyond the scope of this project—and too many other benefits analyses--is the
distribution aspect of benefit policies. By neglecting distribution concerns,
economists are unable to appreciate many policy objections expressed in the
political arena. For example, attention to the distributional effects of alterna-
tive water pollution policies would be a valuable complement to the efficiency-
oriented questions that constitute the primary focus of benefits analysis.
Further rationale for such efforts stems from Executive Order 12291, which
recognizes the importance of distribution effects by requiring them in regu-
latory impacts analyses.
The future research agenda for the individual benefits estimation ap-
proaches contains items ranging in subject from experimental design and sam-
pling to the behavioral models that underlie several approaches. Some of the
agenda items are already being studied in various quarters, while others will
involve substantial funding—e.g., basic data collection—for any progress to
be made.
Specific Research Issues
The travel cost model developed in the project raises as many research
questions as it answers. The main answer is that the model can be used to
estimate the benefits of water quality improvements in a way consistent with
economic theory.* However, many problems were encountered on the way to
answering this fundamental question. For example, in the survey data used
to estimate the travel cost model, as in many surveys involving noneconomic
data, the data were heaped at specific points, possibly presenting problems for
ordinary least-squares regression analysis. Specifically, all visitors who made
only one visit to a site were heaped at the zero point for the logarithmic trans-
formation of the dependent variable, while the visitors who made the maximum
were heaped at the other end point. The maximum is the value (8) assigned to
the open interval for five or more visits. The remaining visitors were arrayed
at specific intervals in between. The need, obviously, is for a statistical esti-
mator that can handle this problem. In terms of the absolute magnitude of the
estimated values, which is important for estimating benefits, the differences
may be small, but this is a fundamental question requiring statistical analysis
rather than judgment. Equally important, the fact that all respondents have
used the site at least once implies that this study fails to consider the demands
of individuals whose maximum willingness to pay falls below their travel cost.
This truncation can, as suggested in the report, lead to biased estimates of
This is one of the items on Freeman's [1979a] research agenda cited
earlier.
1-13
-------
site demands. It is important to evaluate the implications of amending the
statistical models to directly account for these effects for the benefit estimates
derived for water quality Improvements.
Many of the items on the travel cost research agenda stem from limited
data. This project used the 1977 Outdoor Recreation Survey's Federal Estate
component, which surveyed visitors at various recreation sites on reaerai
lands. Although in many ways these data are far better than those in earlier
survey efforts, they omit many items important for the travel cost model. For
example, there were no questions on substitute sites that respondents had
considered—or even visited at other times—before visiting a particular site.
While the generalized model assumes that site attributes are capable of ""eTIect-
ing substitution potential, the model would be considerably improved if it had
a better measure of substitutes.
The travel cost model also assumed that the sole purpose of an individ-
ual's trip was to visit a particular site. However, Haspel and Johnson [1982]
point out the potential for overstating benefits when there are multiple pur-
poses for a trip, suggesting the need for more research using itinerary infor-
mation to assess the importance of multipurpose trips. Also needed for the
travel cost model are more data on the types of time allocations the individual
considered in making the trip. For example, was work time forgone or com-
pulsory vacation time? Each may have a different opportunity cost. With
answers to these questions, it will be possible to improve the calculation of an
individual's time costs for recreation.
Including site attributes in the travel cost model created several data-
related questions. Specifically, because water quality data from the standard
storage system (STORET) were inadequate for many recreation sites, obser-
vations were missing on key parameters, and the monitoring station information
was frequently unreliable. Clearly, more comprehensive data are needed,
especially for water quality parameters relevant to recreation activities. Data
on other site attributes such as access or size were available for the U.S.
Army Corps of Engineers' sites through the Corps' Resource Management
System. However, to apply the model to other recreation sites—e.g., sites
managed by the U.S. Forest Service—would require similar information on
important site attributes. Presently, such data are not readily available.
The future research agenda for the contingent valuation approach is
aimed at a more systematic treatment of issues involving the design of the
hypothetical market. The research questions are in the general area that
economists have termed "framing the question" (see Brookshire, Cummings,
et al. [1982])—an area generally called "context" in the psychological litera-
ture. The definition of the commodity to be valued, the question format used
to elicit the value, the ordering of various valuation and nonvaluation ques-
tions, the means of payment in the market, and the information provided in
the survey questionnaire are all important elements in this framing process.
More attention to these issues is likely to substantially improve the under-
standing of the approach and provide results that are easier to interpret.
1-14
-------
This project addressed several general contingent valuation issues by
comparing several question formats—bidding games with two starting points,
direct question, and the unachored payment card—both to each other and to
results from the contingent ranking format. Different payment cards, such
as the anchored card used by Mitchell and Carson [1981], were not compared.
In addition, the contingent ranking format was always used in conjunction with
another question format, 'which limits the independence of the conclusions.
Both of these are good candidates for future research.
This survey was conducted in a specific river basin, making the orienta-
tion more micro in scope than Mitchell and Carson's [1981]. A more systematic
comparison of their results for overall national water quality and the results
of this study for the Monongahela River basin may be useful. Moreover, the
general framing questions are especially relevant to the macro approach, where
it is more difficult to define the hypothetical commodity. If policy decisions
require basin-specific results, either specific surveys (or the ability to trans-
fer results between basins) or the ability to infer estimates for specific river
basins from the macro approach will be required.
Recently, Brookshire, Cummings, et al. [1982] introduced the ideas of
environmental accounts and budget constraints as part of the framing issue.
The accounts question aims at determining whether people give an overall
environmental quality bid in a survey or a bid for the specific hypothetical
commodity. The budget constraint requires that individuals provide rough
budget shares for their monthly incomes and then reallocate these categories
to provide the budget amount for the hypothetical commodity. The preliminary
results in Brookshire, Cummings, et al. [1982] suggest this is a useful avenue
for learning more about framing influences.
• Finally, improving efficiency in defining hypothetical markets is a neg-
lected area in the contingent valuation literature. One promising approach is
the use of focus groups (from market research literature) to obtain impressions
about terminology, visual aids, and other framing issues. Applying these mar-
keting research ideas to contingent valuation may indicate their overall merits.
Research agendas must continually evolve, producing new avenues from
deadends that once offered promise. The present agenda tries to map some
of these new avenues. The passage of time and the fruits of future research
will mark its ultimate usefulness.
1.4 GUIDE TO THE REPORT
This chapter has introduced the report by highlighting the project objec-
tives and summarizing the findings of the research. Chapter 2 provides a
brief review of some of the theoretical issues of comparing alternative benefit
estimation approaches. After describing the Monongahela River basin, Chap-
ter 3 summarizes the sampling and survey plans for the contingent valuation
and contingent ranking approaches used in the case study. Chapter 4 builds
on the contingent valuation foundation laid in Chapter 3 by presenting the
research design for the contingent valuation survey, by profiling key groups
of respondents, and by summarizing the empirical option price results, includ-
1-15
-------
ing the effects of question format, starting point, and interviewer bias.
Chapter 5 synthesizes the theoretical underpinnings of option value, giving
particular attention to the role of supply uncertainty, and presents empirical
results for both option and existence values. Chapter 6 reviews the theory
underlying the contingent ranking approach, provides a critical summary of
its previous applications, considers appropriate measures of benefits, and sum-
marizes the empirical findings from its application to the Monongahela River
basin. Chapter 7 presents the development of a generalized travel cost model
and describes its application to predict the recreation benefits of water quality
improvements in the Monongahela River. The development of the model treats
the empirical significance of model specification, site time costs, simultaneity
in visit/site time decisions, and statistical biases in its predicted values.
Chapter 8 compares the alternative approaches for estimating recreation and
related benefits, in light not only of previous comparison attempts but also of
a priori expectations. In addition, Chapter 8 also describes paired compari-
sons of the contingent valuation and travel cost approaches and of the contin-
gent valuation and contingent ranking approaches using multivariate regression
techniques.
1-16
-------
CHAPTER 2
A BRIEF REVIEW OF THE CONCEPTUAL BASIS FOR THE
BENEFIT ESTIMATION APPROACHES
2.1 INTRODUCTION
An ideal comparison of benefit estimation approaches would begin with a
detailed theoretical appraisal of each approach, showing how each is derived
from a common conceptual framework. However, this kind of appraisal is
beyond the scope of this project. Instead, this chapter highlights the assump-
tions, information, and types of benefits measured by each of three approaches
and compares these features on general, rather than on common, theoretical
grounds.
The definition of economic benefits based on theoretical welfare economics
has closely followed the model of consumer behavior, which suggests that indi-
viduals can acquire utility only through consuming goods or services. This
framework leads to definitions of economic benefits best suited for describing
user benefits associated with improvements in environmental quality. However,
since the work of Krutilla [1967], nonuser, or intrinsic, benefits have been
increasingly recognized as playing an important role in the aggregate values
for certain environmental resources.
Intrinsic benefits are generally viewed as arising from two sources. The
first source suggests that an individual can realize utility without direct con-
sumption of a good or service. Rather, other motives can be satisified with
allocation patterns for certain resources, and these motives—"stewardship"
and "vicarious consumption" in Freeman's [1981] terms—can lead to utility,
therefore providing nonuser benefits. An alternate view can be derived by
redefining the act of consumption to admit what might be considered indirect
use of the services of an environmental amenity.
The second source of intrinsic benefits is derived by relaxing one of the
assumptions underlying conventional consumer behavior models. The simplest
treatment of the conditions for efficient resource allocation assumes that all
goods and services—whether they provide positive increments to utility or
decrease it--are available with certainty. Of course, this is not the case in
the real world. Indeed, in some circumstances—e.g., the degree of reversi-
bility in water quality conditions—uncertainty may well be the most important
element of the public policy problem. In these cases, therefore, consumer
behavior models must be amended to reflect how households react to uncer-
tainty and whether they would be willing to pay for action that would reduce
it.
2-1
-------
A second relevant feature of the definitions of economic benefits presum-
ably arises from the early focus on goods or services exchanged in -\Pr™e
markets. These definitions developed measures of benefits for Pr'ce.c u 9"l
Since environmental policy has dealt mostly with quantity (or quality) changes
in services provided outside of private markets, these measures must oe
adapted to meet policy needs.
The purpose of this chapter is to briefly review the theoretical concepts
generally used in measuring economic benefits. Specifically, Section 2.2 deals
with the theoretical basis of benefit measurement based on the concept of an
individual's willingness to pay and describes alternative ways to measure
changes in consumer welfare. Section 2.3 outlines the framework for compar-
ing different benefit estimation approaches—an adaptation of the Smith-Krutilla
[1982] framework for classifying the different approaches and summarizing
their conceptual bases. Section 2.4 describes the welfare measurement bases
underlying the two benefit estimation approaches compared in this study--
travel cost and contingent valuation (including the contingent ranking format).
Finally, Section 2.5 prvides a brief summary.
2.2 A BRIEF REVIEW OF THE CONVENTIONAL THEORY OF BENEFITS
MEASUREMENT
The primary emphasis for this study of recreation and related benefits of
water quality improvements focuses on the measurement of benefits that accrue
to individual households. Fortunately, the theory of consumer behavior pro-
vides a framework for measuring these benefits. This section briefly reviews
this framework to set the stage for the comparison of approaches that follows.
The first guidepost for the definition and measurement of economic bene-
fits that the theory of consumer behavior provides is the individual demand
function, shown in Figure 2-1. This function describes for any good, X, the
maximum amount an individual would be willing to pay for each quantity of X.
The downward slope of the curve indicates that individuals are willing to buy
more of X at lower prices than they are at higher prices. The simple two-
dimensional diagram in Figure 2-1 assumes all other factors that might influ-
ence demand—income, the prices of related goods, etc.--do not change. Thus,
according to the demand function, if the market leads to a price of P0/ the
individual will purchase Q0 of X and make a total expenditure equal to P0AQ0O.
Since the demand curve measures the individual's maximum willingness to pay
for each level of consumption, the total willingness to pay for Q0 can be
derived—total expenditures plus the triangle P0P-A. This difference between
the amount they are willing to pay and what individuals actually pay with a
constant price per unit is defined as the consumer surplus—the conventional
dollar measure of the satisfaction individuals derive from consuming a good or
service, exclusive of what they pay for it.
As a dollar measure of individual welfare, however, consumer surplus is
not ideal. The most direct way of understanding its limitations is to consider
2-2
-------
Price
($/unit)
P=
•*• Quantity/time
Figure 2-1. The demand function and the consumer surplus welfare measure.
the measurements underlying a conventional demand function. An individual's
demand function describes the maximum an individual with a given nominal
income would be willing to pay for each level of consumption of a particular
good. Specifically, if the price paid changes, it will affect not only what the
individual can purchase of this good, but also the purchases of all other com-
modities through its effect on the remaining disposable income. Thus, move-
ment along a conventional demand function affects the level of satisfaction an
individual will be able to achieve with a given income. For example, suppose
the price of hypothetical good X declines to Px. The individual can purchase
the same quantity of X at its new price as indicated in Figure 2-1 by the area
OPiBQo and have income remaining, as given by PjPoAB, to purchase more X
or more of other goods and services. The movement to a consumption level of
OQj describes the increased selection of X under the new price. This change
leads to a higher utility level because more goods and services can be con-
sumed with the same income. For consumer surplus to provide an "ideal"
dollar measure of individual well-being, however, the conversion between dol-
lars and individual utility levels must be constant for every point on the
demand curve. According to this example, then, each point on a conventional
demand function in principle corresponds to a different level of utility. Thus,
no single conversion factor links consumer surplus and utility.
In his seminal work on consumer demand theory, Hicks [1943] noted that
an ideal measure would require that utility be held constant at all points along
the demand curve. As a practical matter, however, the difference between
the area under such an ideal, Hicksian-based demand curve and that under a
conventional demand curve depends on the size of the income effects accom-
panying the price changes associated with movements along the ordinary
demand curve. As suggested earlier, price reductions lead to more dispos-
able income. To judge the association between the two measures of welfare
change, all aspects of the choice process that affect the size of the change in
disposable income must be considered. For example, if the price change for
X is small and the share of the budget spent on the good X is also small, the
2-3
-------
change in disposable income is likely to be small. Thus, little difference will
exist between the ordinary measure of consumer surplus and the measure
derived from Hicks' idealized demand curve. However, the same outcome
arises either if income has little effect on the demand for X or if an individ-
ual's preferences are such that the demand responsiveness to income is equal
for all goods (i.e., unitary income elasticities of demand).
Of course, each of the conditions described above is a special case.
When ordinary demand functions are used to measure the benefits of an action
in practical applications, the factors influencing the demand function's relation-
ship to an ideal dollar measure of welfare change must be identified. Fortu-
nately, Willig [1976] and Randall and Stoll [1980] have derived such guidelines
for cases involving price and quantity changes, respectively. To understand
these guidelines, the possible theoretical measures of individual welfare change
must first be defined in more precise terms.
Hicks' [1943] theoretical analysis of measures of welfare change provides
the basis for developing a set of rigorous measures and, with them, the error
bounds for ordinary consumer surplus. The four Hicksian welfare measures
for a price decrease are summarized below:
Compensating variation (CV) is the amount of compensation that
must be taken from an individual to leave him at the same level
of satisfaction as before the change.
Equivalent variation (EV) is the amount of compensation that
must be given to an individual, in the absence of the change,
to enable him to realize the same level of satisfaction he would
have with the price change.
Compensating surplus (CS) is the amount of compensation that
must be taken from an individual, leaving him just as well off
as before the change if he were constrained to buy at the new
price the quantity of the commodity he would buy in the absence
of compensation.
Equivalent surplus (ES) is the amount of compensation that must
be given to an individual, in the absence of the change, to make
him as well off as he would be with the change if he were con-
strained to buy at the old price the quantity of the commodity
he would actually buy with the new price in the absence of com-
pensation.
•
As a simplified comparison, Figure 2-2 highlights the essential differences
between the Hicksian variation measures and the ordinary consumer surplus
measures when the price of a good decreases. The two Hicksian demand
curves holding utility constant (at levels U0 and ux with Ut > U0) are shown
as H(U0) and H(Ut), the prechange and postchange levels of utility, respec-
tively. The ordinary demand curve—also known as the Marshallian demand
curve—is shown as D, where income, and not utility, is held constant. The
2-4
-------
Price
\
H1U,)
NOTE: D • ordinary dtmand cum
H(U0). H(Un) - Hiduim demand cumi
Ordinary eomunwr Mnphn » a + b
Compmrtniwtotion-*
Equhnlmt nriition • • + b + e
Figure 2-2. A comparison of alternative welfare measures.
compensating variation measure, labeled as area a, uses the original level of
utility as its reference point and indicates the amount of compensation that
must be taken from an individual to leave him at the original level of utility
when the price changes from P0 to Pt. The equivalent variation measure is
represented by area a+b+c. It measures the change in income equivalent to
the change in prices and thereby permits an individual to realize the new level
of utility with old price P0. The change in ordinary consumer surplus is the
area under the ordinary demand curve, D, between P0 and Px. In Figure 2-2
it is shown as areas a+b.
The concepts of compensating surplus and equivalent surplus were origi-
nally defined as measures of the welfare change resulting from a price change,
given that the quantity of the good whose price has changed is not allowed to
adjust. However, it is also possible to interpret these concepts as measures
of the welfare change associated with a quantity change (see Randall and Stoll
[1980]). Just, Hueth, and Schmitz [1982] have recently offered a diagrammat-
ic illustration of compensating and equivalent surplus in a format similar to
that used above to describe compensating and equivalent variation. However,
in the present example, the price is assumed constant at some arbitrarily low
value (effectively zero for Figure 2-3), and D is interpreted as an ordinary
demand curve (i.e., as if the quantities consumed could be realized only at
the corresponding prices and not the constant price). In Figure 2-3 a change
in the quantity of the good available from Q0 to Qt leads to a compensating
surplus of c+f and an equivalent surplus of a+e+c+d+f+g. The ordinary con-
sumer surplus is c+d+f+g, which is d+g more than the compensating surplus
measure and a+e less than the equivalent surplus.
2-5
-------
Price'
^•Quantity
Source Just, Hutth, mi Schmitz (1982].
Nott: Ordimry conninwr wrplus -c+d + f + g
Compensating nirplus • c + f
Equinlmt turplui • i + c + d+e + f + g
Figure 2-3. Surplus measures for a change in quantity.
Table 2-1 relates the welfare measures under different conditions of will-
ingness to pay/accept, showing quite clearly that no one unique measure
exists. Rather, the appropriate measure is determined by the particular situ*
ation. Table 2-1 reinforces this point by presenting the types of welfare
measure in relation to different situations. For a price decrease, for example,
the following relationship holds between the alternative welfare measures:
ES > EV > CV > CS .
For a quantity increase, the equivalent surplus measure will be greater than
the compensating surplus measure. The primary reason for the differences be-
tween welfare measures is that the equivalent surplus and equivalent variation
are not bounded by an individual's income constraint, while the compensating
variation and compensating surplus measures are. It should also be noted that
the measures are symmetrical and that, for a price increase or quantity de-
crease, the relationship between the measures is exactly the reverse.
It is important to recognize that the compensating and equivalent meas-
ures of welfare changes differ because they imply a different assignment of
property rights to the individual and therefore are based on different corres-
ponding frames of reference. For example, with a price decrease, the compen-
sating variation measure takes the initial price set as an individual's frame of
reference and asks, in effect, "What is the maximum amount he would be will-
ing to pay to have access to the lower prices?" By contrast, equivalent varia-
tion takes the new, lower price set as an individual's frame of reference and
2-6
-------
Table 2-1. Alternative Welfare Measures and Types of Consumer
Surplus Measures for Contingent Valuation Studies
WTP
WTA
NOTE:
CS is
Price
decrease
CV; CS
EV; ES
the amount of come
Price Quantity
increase increase
EV; ES CS
CV; CS ES
•ensation that must be taken from an
Quantity
decrease
ES
CS
individual,
leaving him just as well off as before the change if he were constrained
to buy at the new price the quantity of the commodity he would buy in
the absence of compensation.
CV is the amount of compensation that must be taken from an individual
to leave him at the same level of satisfaction as before the change.
ES is the amount of compensation that must be given to an individual, in
the absence of the change, to make him as well off as he would be with
the change if he were constrained to buy at the old price the quantity
of the commodity he would buy in the absence of compensation.
EV is the amount of compensation that must be given to an individual, jn
the absence of the change, to enable him to realize the same level of
satisfaction he would have with the price change.
WTA is the amount of money that would have to be paid to an individual to
forego the change and leave him as well off as if the change occurred.
WTP is the amount of money an individual will pay to obtain the change and
still be as well off as before.
describes the minimum amount an individual would be willing to accept to relin-
quish his right to the lower price. These measures bound the range of dollar
values for the welfare changes because they describe the results obtained from
the perspectives of the initial utility level and the final utility level. Conse-
quently, Willig [1976] uses this feature to establish conditions under which
conventional consumer surplus would approximate "ideal" measures for the wel-
fare change associated with a price change. Moreover, Randall and Stoll
[1980] follow essentially the same logic to gauge the relationship between ordi-
nary consumer surplus measures for a quantity change and the corresponding
compensating and equivalent surplus measures.
Equations (2.1) and (2.2) provide the basis for the Willig bounds for the
difference between the ordinary consumer surplus measure and the equivalent
2-7
-------
variation and compensating variation measures of a change in welfare due to a
price change:*
CV - PCS „ IQCSl N (2.1)
|OCS| ~ 2 M
PCS - EV „ IOCSI N , (2 2)
IOCSI ~ 2 M
where
OCS = ordinary consumer surplus measure of welfare change
N = income elasticity of demand = -x*- / -^-
M = initial level of income.
These relationships can be evaluated at different values for the income elas-
ticity of demand over the region for the price change and thereby provide
bounds for the magnitude of the discrepancy between ordinary consumer sur-
plus and the equivalent and compensating variation welfare measures. Equa-
tions (2.1) and (2.2) assume that the income elasticity of demand (N) is
approximately constant over the region for the price change (see Willig [1976],
pp. 592-593, for a discussion). If this assumption is relaxed, the bounds can
be stated as inequalities for the percentage difference between ordinary con-
sumer surplus and the corresponding measures of welfare, as in Equations
(2.3) and (2.4):
|ocs| NS cv _ QCS |ocs| NL
2~M - JOCSJ~ 2 M ' (2-3)
OCS N _ pCS N
S * OCS EV ^ L f (2>4)
2 M - |OCS| = 2 M
where
N_ = the smallest value of the income elasticity of demand over the region
for the price change
*lt is important to note that the direction of the price change affects the
sign of ordinary consumer surplus, compensating variation, and equivalent
variation and, thus, the interpretation of Equations (2.1) through (2.4). This
formulation adopts Willig's convention that ordinary consumer surplus is posi-
tive for a price increase and negative for a price decrease so that it corre-
sponds to the interpretation of compensating variation or equivalent variation.
See Willig [1976], p. 589.
2-8
-------
= the largest value of the income elasticity of demand over the region
for the price change.
The Willig approximation is reasonable if the value of N ^ 0.05. If this
value is greater than 0.05, Willig has provided a table of error bounds
based on the relationships used to derive these approximate bounds.*
2.3 A FRAMEWORK FOR COMPARING ALTERNATIVE BENEFIT
MEASUREMENT APPROACHES
Comparing alternative approaches for estimating the recreation and related
benefits of water quality improvements at first seems formidable because of
the wide range of consumer behavior outcomes described by each. However,
despite this diversity, all approaches adhere to a consistent general model of
consumer behavior: individuals allocate their monetary and time resources to
maximize their utility subject to budget and time constraints. As noted at the
outset, a complete comparison of the methods could derive each method from
this common conceptual basis. However, this section simply provides a taxo-
nomic framework that eases the comparison of approaches by drawing clear
distinctions between the assumptions underlying each.
Figure 2-4 presents the Smith-Krutilla [1982] framework for classifying
the alternative approaches for measuring the recreation and related benefits
of water quality improvements. This framework considers linkages between
changes in water quality and observable actions of economic agents that affect
the information available for measuring water quality benefits. In particular,
Smith and Krutilla suggest that all approaches for measuring the benefits of a
change in an environmental resource can be classified as involving either
physical or behavioral assumptions.
The category associated with physical assumptions in this framework main-
tains that the association between the environmental service of interest (i.e.,
water quality) and the observable activities (or changes in goods or services)
is a purely physical relationship. The responses are determined by either
engineering or technological relationships. Thus, the evaluation of water qual-
ity changes in such a framework must begin by identifying the activities
affected by water quality. Analysts must then focus on measuring the techni-
cal relationships, sometimes referred to as damage functions, assumed to exist
between water quality and each activity. Because water quality improvements
can be associated with the support of gamefish, swimming, and the use of
water for human consumption, the physical approach seeks to specify the tech-
nical linkages between water quality levels and permitted amounts of recrea-
tion fishing, swimming, and water consumption. Another example of the
*Two excellent discussions of the practical implications of the Willig
bounds for benefit measurement are available in Freeman [T979a], pp. 47-50,
and in Just, Hueth, and Schmitz [1982], pp. 97-103.
2-9
-------
No Rote for
Behavioral Responses
of Economic Agents
Bahavioral Responses
of Economic Agents
Are Essential
Types of Linkage
Between Water Quality
Change and Observed
Effects
Physical
Linkages
Bahavioral
Linkages
Indirect
Links
Direct
Links
Types of
Assumptions
Required
Responses are
determined by
engineering or
" technological"
relationships
Restrictions on the
nature of individual
preferences OR
associations in the
delivery of goods
or services
Institutional
Measurement
Approaches
Damage Function
Hedonic Property Value
Travel Cost*
Contingent Valuation*
(including
Contingent Ranking*)
•d in ttm study.
Figure 2-4. Smith-Krutilla framework for classifying the measurement bases and
approaches of economic benefits resulting from improved water quality.
physical approach to evaluating the effects (and, ultimately, the benefits) of
a water quality change can be found in the dose-response models used to eval-
uate the health risks associated with certain forms of water pollution (see
Page, Harris, and Bruser [1981] for a review of these models). Although
these models ignore economic behavior and postulate that the relationships
involved can be treated independently of the motivations of economic agents,
they may well provide reasonable approximations of the actual effects on water
quality for certain classes of impacts. However, these models are unlikely to
be adequate when economic agents can adjust their behavior in response to
the water quality changes and, as a result, are not considered in this study.
The behavioral category of valuation methodologies in the Smith-Krutilla
framework relies on the observed responses of economic agents and on a model
describing their motivations to estimate the values (or economic benefits) asso-
ciated with a change in a nonmarketed good or service. Within this class,
direct or indirect links identify three classes of assumptions that can be used
to develop measures of individual willingness to pay. The first type of
assumption used within the indirect behavioral framework requires restrictions
on the nature of the individual's utility function and is usually associated with
Maler's [1974] weak complementarity. This type of assumption maintains that
an individual's utility function is such that there is a specific association
between the nonmarketed good (or service) and some marketed commodity such
2-10
-------
that the marginal utility of an increment to the consumption of the nonmarketed
good is zero when the individual is not consuming some positive amount of the
associated, marketed commodity. This assumption maintains that a type of
"jointness" exists in the formation of the individual's utility, which, in turn,
constrains the feasible responses to price changes for the marketed good (or
changes in the availability of the nonmarketed good). Thus, the selection of
the two goods is joint, and market transactions for one good can be used to
determine demand for the other. Of course, this approach depends upon the
plausibility of the restriction on an individual's utility function. Researchers
have used this restriction to justify both hedonic property value and travel
cost studies.
Smith and Krutilla [1982] argue that the weak complementarity behavioral
restriction is not necessary for these approaches and that the observed tech-
nical associations between marketed and nonmarketed goods are responsible
for the ability to use these methods to measure benefits of changes in a
•nonmarketed good. In the case of the technical assumptions, the availability
of the nonmarketed service is tied to some marketed good by the nature of its
natural delivery system, making the linkage an observable phenomenon rather
than a feature of an individual's preferences. For example, water-based
outdoor recreation is undertaken using the services of recreation sites on
rivers or lakes. Each recreationist is interested in the water qualities only at
the sites considered for his recreation use. By selecting a site for these
activities, an individual is also selecting a water quality, because the two are
"technically linked" or jointly supplied. Thus, where there is a range of
choice (i.e., several different combinations of recreation sites and water
quality), how an individual values the nonmarketed good or service can be
seen through his observable actions, including such decisions as the selection
of a residential location or visits to specific recreation facilities (see Rosen
[1974] and Freeman [I979c]). This study specifically considers the travel cost
method, which uses this technical association as its basis for measuring water
quality benefits.
The last case of behavioral approaches to benefit estimation involves
direct linkages between water quality and an individual's actions. The assump-
tions made to ensure these linkages are labeled institutional, a designation
somewhat more difficult to understand than previous descriptions because it
encompasses the contingent valuation and contingent ranking methods for
measuring an individual's valuation of environmental amenities. Specifically,
the institutional assumptions arise because the analyst assumes that individual
responses to hypothetical decisions (or transactions) are completely comparable
to individual responses revealed in actual transactions. The term institutional
is used for this class because the organized markets in which goods and serv-
ices are exchanged are institutions that provide the information on individuals'
marginal valuations of the commodity involved. With the survey approach,
the interviewer poses the survey questions to construct an equivalent institu-
tional mechanism in the form of a hypothetical market. Both the contingent
valuation and the contingent ranking methods will be considered under this
approach.
2-11
-------
2.4 THE NATURE OF THE BENEFITS MEASURED IN THE ALTERNATIVE
APPROACHES
This section highlights the nature of the benefits measured in the travel
cost and contingent valuation approaches.
2.4.1 Travel Cost Approach
The travel cost approach measures the change in ordinary consumer sur-
plus for a water quality improvement, represented for an individual incurring
travel costs per trip of OPX by area ABCD in Figure 2-5. To empirically
develop the ordinary consumer surplus estimate, the travel cost approach
assumes both that travel to a recreation site reveals a respondent's reservation
price for that site's services and that water quality is jointly supplied along
with the other site attributes. If other variables are held constant, and if
sites are placed on a common measurement scale,* area ABCD can be measured
by observing individuals' site selections across sites with varying levels of
water quality, thus revealing the effect of water quality on site demand.
Therefore, while both Freeman [I979b] and Feenberg and Mills [1980] maintain
that conventional travel cost models cannot measure benefits associated with
water quality change,t the generalized travel cost model developed for this
Travel costs
($/Qx)
WQ2 > WQ1
Qx/t (visits/year)
Figure 2-5. Travel cost demand function with water quality improvement.
*The rationale for this measurement approach is presented in more detail
in Section 7.3.
tTheir models do not gauge the demand change that accompanies a water
quality change.
2-12
-------
study (see Chapter 7) uses the responses of individuals at different locations
to both travel cost and water quality levels to infer benefits of water quality
changes. The information provided by these responses allows the change from
D(WQ±) to D(WQ2) in Figure 2-5 to be distinguished (where WQt and WQ2
represent different levels of water quality, with WQ2 > WQt).
2.4.2 Contingent Valuation Approach
The contingent valuation approach directly measures an individual's will-
ingness to pay for water quality in an institutional arrangement that approx-
imates the market for water quality. Unlike the travel cost approach, contin-
gent valuation does not require observations of individuals' decisions on use
of recreation sites with given "implicit" service prices, but it does assume an
individual's response in the hypothetical market is the same as it would be in
a real market. That is, respondents are assumed not to behave strategically,
not to give unrealistic responses, and not to be influenced by the survey
questionnaire or the interviewer who administers the survey questionnaire.
Furthermore, the contingent valuation approach imposes an institution that
leads to a hypothetical change in an individual's budget constraint by requir-
ing that the individual "pay" for the specified water quality improvement.
Thus, the new budget constraint for the utility maximization process includes
both the prices and quantities of market goods and the hypothetical price and
defined quantity of water quality.
The institutional design underlying contingent valuation surveys requires
that ownership of the property rights for water quality at the recreation site
be determined in the specification of the question, thus affecting the appro-
priate measure of consumer welfare. Specifically, consumer ownership of
property rights would indicate a willingness-to-accept measure as the appro-
priate valuation concept, and industry ownership would dictate a willingness-
to-pay measure. Although currently boatable throughout, the Monongahela
River—the site used for this study (see Chapter 3)--supports swimming and
fishing only upriver from Pittsburgh, and property rights are in a state of
flux with considerable confusion over ownership (see Feenburg and Mills
[1980]). Thus, a reasonable allocation for this study's survey of Pittsburgh
residents is that consumers own the rights to boatable water (which suggests
an equivalent surplus measure), while no one yet owns the rights to fishable,
swimmable water along the entire river (which indicates a compensating surplus
measure).
While using a willingness-to-accept measure for maintaining a boatable
water quality level and a willingness-to-pay measure for the value of moving
to fishable and swimmable levels is consistent with current Monongahela prop-
erty rights, willingness-to-accept measures have proven difficult in hypothe-
tical market experiments, thus creating serious problems in the development
of a workable survey methodology. For example, respondents have either
refused to answer, given infinite bids, or refused to accept any compensation
for reductions in environmental quality [Schulze, d'Arge, and Brookshire
[1981] and Bishop and Heberlein, 1979]. To cope with this problem, this
study employs a willingness-to-pay (equivalent surplus) measure for the
decrease from boatable water quality and a compensating surplus measure for
improvements from the same level.
2-13
-------
2.4.3 Contingent Ranking Approach
Like the other contingent valuation formats, contingent ranking relies on
individuals' responses in a hypothetical market situation. However, instead of
requiring an individual to respond with the maximum willingness to pay for a
water quality improvement, contingent ranking requires that individuals rank
outcomes—consisting of a hypothetical payment and a corresponding level of
water quality—from most preferred to least preferred. The implicit argument
underlying contingent ranking is that an individual is better able to respond
to the hypothetical market when both outcomes are specified. In the utility
maximization framework underlying the contingent ranking approach, an indi-
vidual ranks the alternatives based on their implications for his ability to max-
imize utility with a given income, the prices of other goods, and the proposed
combination of payment and water quality. Analytically, this choice can be
described by comparisons of the indirect utility functions arising from each of
these sets of decisions. An appropriate compensating surplus measure can
then be derived from estimates of the indirect utility function.
2.5 Summary
Partly because they are all based on the common standard of constrained
utility maximization, the travel cost, contingent valuation, and contingent
ranking approaches can each develop measurements of changes in consumer
welfare. The travel cost approach measures the change in ordinary consumer
surplus, the contingent valuation approach measures equivalent and compensat-
ing surpluses, and the contingent ranking format yields a compensating sur-
plus welfare measure.* The relationship between each of these methods' meas-
ures of the welfare changes associated with water quality changes is consid-
ered in the comparison analysis reported in Chapter 8.
*lt should be noted that, for the contingent valuation approaches, ques-
tions have been formulated to include both user and nonuser values. Strictly
speaking, both approaches measure the option price, but the contingent valua-
tion approach permits the user value component to be identified.
2-14
-------
CHAPTER 3
SURVEY DESIGN
3.1 INTRODUCTION
Estimating the recreation and related benefits of water quality improve-
ment with the contingent valuation approach requires an Integrated survey
design. This chapter describes the survey design for the case study of the
Monongahela River. Specifically, Section 3.2 describes the general back-
ground of the Monongahela River basin area, Section 3.3 highlights the
sampling plan for the project, and Section 3.4, a discussion of the survey
plan, concludes the chapter with detailed information on the survey field
procedure.
3.2 GENERAL DESCRIPTION OF THE MONONGAHELA RIVER BASIN
This section describes the Monongahela River basin, providing a general
description of river geography, river uses, river-related recreation activities,
and a socioeconomic profile.
3.2.1 Geography
Formed by the confluence of the West Fork and Tygart Rivers near
Fairmont, West Virginia, the Monongahela River drains an area of 7,386 square
miles in southwest Pennsylvania, northern West Virginia, and northwest
Maryland. (See Figure 3-1 for a map of the area.) It flows northerly 128
miles to Pittsburgh, where it farms the Ohio River headwaters with the
Allegheny River, and has two major tributaries, the Youghiogheny and Cheat
Rivers. All 128 miles of the Monongahela are navigable year round by motor-
ized commercial traffic.
Characterized by steep banks and rugged terrain, the Monongahela River
basin lies in five Pennsylvania Counties (Allegheny, Greene, Fayette, West-
moreland, and Washington) and two West Virginia counties (Monongalia and
Marion) in the Appalachian Plateau and the Allegheny Mountains. The
Monongahela basin currently supports four major reservoirs:
Deep Creek Reservoir—A privately owned Maryland facility
operated on a Youghiogheny River tributary to generate 51
megawatts of electric power.
Lake Lynn Reservoir--A privately owned West Virginia facility
operated on the Cheat River to produce 19 megawatts of electric
power.
3-1
-------
Figure 3-1. Map of Monongahela River and other area recreation sites.
Tygart River Reservoir—A facility operated by the U.S. Army
Corps of Engineers to provide flood control, recreation, and
low flow augmentation. This reservoir provides most of the
Monongahela's augmented flow, a minimum of 340 cfs in the
upper river.
Youghiogheny River Reservoir—A facility operated by the U.S.
Army Corps of Engineers to provide a minimum flow of 200 cfs
for the Monongahela River.
Comprising nearly 30 percent of the river basin's seven-county area, the
following urban areas and boroughs (listed below with 1970 census population)
line the Monongahela's banks:
Pittsburgh
McKeesport
Clairton
Duquesne
Monessen
Monongahela
Morgantown
Fairmont
520,117
37,977
15,051
11,410
17,216
7,113
29,431
26,093
Donora
Charleroi
Brownsville
Braddock
Glassport
Munhall
Port Vue
West Miff in
8,825
6,723
4,856
8,795
7,450
16,574
5,862
28,070
3-2
-------
3.2.2 Uses
As part of the Mississippi River Waterway System, the Monongahela has a
9-foot-deep navigation channel from Pittsburgh to Fairmont to support both
commercial and recreation river traffic. This navigation channel ranges in
width from a minimum of 250 feet to nearly full river width at the river's
mouth and is currently maintained by a series of nine lock and dam struc-
tures. The heaviest barge traffic occurs at Structures 2 and 3. Use of the
locks and dams for generating hydroelectric power is currently under consid-
eration and would provide an estimated total capacity of 96.2 megawatts. To
support river traffic, the Monongahela's banks have a boat dock concentration
approaching one dock per mile. However, these docks—which numbered 147
in 1979—are mostly single-purpose, single-user facilities.
Industrial activity along the Monongahela is dominated by the primary
metals industry, which accounts for over 31 percent of the area's total manu-
facturing employment, including 29 percent of all Pennsylvania's steel industry
employment. Also important with respect to industrial activity along the
Monongahela are significant amounts of natural resources, including oil and
gas, limestone, sandstone, sand and gravel, and coal. Area coal reserves
are estimated at approximately 23 billion tons, and the Monongahela River
region alone accounted for 24 percent of total 1977 coal production in
Pennsylvania and West Virginia. Underground mining in the area produced
78 percent of this total, with strip mining operations accounting for the
remainder.
3.2.3 Recreation
Because it essentially is a series of large pools—ranging from 400 to 1,741
.surface acres—created by its nine lock and dam structures, the Monongahela
offers substantial opportunities for recreation. In fact, although the lower
20 river miles, subjected to heavy industrial and urban development, offer
limited recreation opportunities, the remaining 108 miles have seen dramatic
increases in recreation usage over the last 10 years, partially because of im-
proved water quality. As a result of this increased recreation usage, numer-
ous public and private facilities have been developed along the Monongahela,
ranging from single-lane boat launching ramps to boat club docks, commercial
marinas, and community parks.
The primary recreation activities along the river are power boating and
fishing. Because power boating is more popular, many recreation facilities
have been constructed primarily to serve it. Partially as a result, the
Monongahela River comprises a substantial portion of the water acreage avail-
able in the region for unlimited horsepower boating.
Although it is second to power boating in popularity, fishing occurs over
a greater number of water acres in the area when small lakes and streams are
considered. In fact, fishing accounts for approximately 12 percent of all
current uses of the Monongahela. Fishing in the river is encouraged by
special programs in both Pennsylvania and West Virginia to stock warmwater
fish, and fish sampling has revealed the presence of up to 47 separate species,
3-3
-------
plus 3 hybrids. Of special interest, the U.S. Environmental Protection Agency
(EPA) and the Pennsylvania Fish Commission, which have monitored fish
population trends in the Monongahela since 1967, have reported a dramatic
increase over an 11-year period in species' diversity and biomass, particularly
in the upper reach.
In addition to power boating and fishing, the Monongahela also offers
other recreation opportunities at several major facilities, including two con-
structed by the U.S. Army Corps of Engineers at the Maxwell and Opekiska
pools; the Tenmile Creek Recreational Area (adjacent to the Maxwell Pool),
which showed increased visitor days from 1972 to 1975; and the Prikett Bay
Recreational Area (at Opekiska Pool), which has also experienced increased
visitation from 1972 to 1975. Recreation activities offered by these sites
include picnicking, camping, boating, and swimming. Despite its length and
general popularity for recreation, the Monongahela nowhere offers either
campgrounds or State parks for potential recreationalists, who are forced to
the substitute sites offered by the Youghiogheny River Reservoir and the
Allegheny River. Both of these substitutes offer better water quality than
the Monongahela and, perhaps, more scenic settings for recreation.
3.2.4 Socioeconomic Profile
In 1977, population for the seven-county area of the Monongahela River
basin totaled 2,417,885, which results in an average population density of 518
persons per square mile. Although density is greatest along the river, there
is a recent trend to move into other areas. However, population changes in
the basin vary according to State: several Pennsylvania counties have experi-
enced a noticeable population decrease in the period from 1960 to 1977, but
Monongalia County in West Virginia experienced a dramatic population increase
during the same period. In general, the basin has a greater percentage of
urban population than either the Pennsylvania or West Virginia State averages.
Per capita income within the basin is lower than either the Pennsylvania
or West Virginia State averages, and the basin in fact contains a higher
percentage of persons living below the poverty level than does either State
generally. Not surprisingly, then, much of the basin's housing stock is gen-
erally Considered substandard, and, in 1970, 70 percent of it was more than
25 years old.
The average education level, which has steadily increased since 1950, is
higher in the basin than it is in either Pennsylvania or West Virginia or in
the United States generally. However, the difference between the basin and
the nation has almost disappeared, eroded by a steadily rising U.S. education
level. Another steadily eroding difference between the basin and the nation
as a whole is in the percentage of the workforce made up of craftsmen and
laborers. Specifically, due primarily to the area's heavy concentration of
primary metals and extraction industry, the basin still has a higher con-
centration of blue collar workers than does the nation generally, but this
difference has greatly diminished during the last 20 years.
3-4
-------
3.3 SAMPLING PLAN
The following subsections describe the sampling plan implemented to
accomplish the objectives of this study. A single-stage, area household
sampling design was used to contact approximately 384 sample households in a
four-county area of southwest Pennsylvania. Appendix A contains additional
details of the survey design, sample selection, and weight calculation.
3.3.1 Target Population
Five counties comprised the sample area for this study (outlined in
Figure 3-2): Allegeny, Fayette, Greene, Washington, and Westmoreland.
These counties were selected because they contain the reach of the Mononga-
hela River within Pennsylvania. The random nature of the sample resulted in
no sample segments being chosen in Greene County. The target population
consisted of all households in this five-county area. Group quarters were not
included, and only adult (persons 18 years and older) household members
were eligible for interview. One adult was selected for the interview from
each household.
WU.KES BARRE HAZLETON
iror.0 ^ f
LEGEND:
0 Placaa of 100,000 or more inhabitants
• Placaa of 50,000 to 100,000 inhabitant!
O Central cities of SMSAs with fewer thin 50,000 inhabitant*
O Places of 25,000 to 50,000 inhabitants outsida SMSAs
Standard Metropolitan
Statistical Areas
Figure 3-2. Geographic location of survey area.
3-5
-------
3.3.2 Sample Selection and Survey Design
The design was a single-stage, stratified cluster sample. The sampling
units (SUs) were noncompact clusters of approximately seven households
each. The clusters were developed by partitioning all the block groups
(BGs) and enumeration districts (EDs) within the five-county area into non-
compact clusters. The clusters were nonoverlapping and, when aggregated,
completely accounted for all of the households in the five-county area.
The sampling units were stratified into three disjoint groups: (1) Pitts-
burgh, (2) not in a place, and (3) a place other than Pittsburgh. Fifty-one
clusters with an average of 7.78 sample housing units (SHUs) each were
selected, yielding 397 SHUs. A roster of all adults was compiled for each
SHU. One adult was randomly selected from each SHU for interview.
3.3.3 Sampling Weights
The probability structure used to select the SHUs and the adults within
each SHU allows calculation of the selection probability for each individual
interviewed. The sampling weights, reciprocals of the probability of selec-
tion, were then calculated. Because interviews were not obtained from all
selected SHUs (80.59 percent response), the sampling weights were adjusted
for the nonresponse.
3.4 SURVEY PLAN
This project required a detailed survey plan to enable the successful
completion of a full range of survey tasks. The following subsections discuss
the procedures and methods developed to carry out these tasks. The major
field tasks were as follows:
To design and perform a limited local pretest of the survey
questionnaire.
To retain field interviewers.
To count and list households within the randomly selected area
segments. (Two field supervisors and two interviewers per-
formed this task.)
To develop a field procedures manual and interviewer training
materials.
To conduct a field interviewer training session.
To administer the benefits instrument at randomly selected
households within the area segments. (One questionnaire was
to be administered by an in-person interview at each sample
household. The desired number of interviews to be conducted
was 305.)
3-6
-------
To develop and implement onsite and offsite quality control
procedures on the work performed by the field staff.
To conduct an interviewer debriefing.
To develop and implement data receipt, data editing, and
keypunch procedures for all resultant data.
3.4.1 Questionnaire Design and Limited Local Pretest
The design of the benefits questionnaire involved the combined talents of
RTI staff knowledgable in benefits analysis and questionnaire design, the
EPA project officer, and selected consultants. Efforts to design the ques-
tionnaire centered on satisfying the two primary objectives:
To collect the data required for analysis
To collect the data in such a way that reliability and validity
are enhanced.
In meeting these objectives, the number and types of questions included in
the instrument and the format that those questions took were determined by
several interrelated factors: Those factors included:
The precise analytic goals of the survey.
The adequacy of the project budget to support the data collec-
tion required.
The facility of the interviewers in administering the instrument.
The tolerance of potential respondents of the time and effort
required to answer the questions.
The ability of respondents to provide the data requested.
Table 3-1 outlines questionnaire development activity. After the data
collection was completed and the interviewers debriefed, it was clear that the
careful attention given to questionnaire design had reaped substantial rewards.
The nuances of the questions and intricate skip patterns made necessary by
anticipated responses necessitated a considerable investment of time early in
the questionnaire development.
Another factor that had a considerable effect on the overall quality of
the instrument was the variety of skills brought to bear on the wording of
questions. The economic concepts, of course, resided with the economists.
However, the wording of questions was critiqued by survey specialists for
sensibility and administrative ease and further reviewed by staff experienced
in questionnaire formatting and overall survey methodology. The net effect
of these efforts was a questionnaire that was more comprehensible than the
economists could have ever produced themselves and more sophisticated than
the survey specialists alone would have designed.
3-7
-------
Table 3-1. Questionnaire Development Activity
Activity
Date (1982)
Review existing survey work: Resources for the
Future, Inc. (RFF) (Mitchell);
Colorado State; Wyoming
Develop first draft for presentation at workshop
Revise draft for review by EPA project officer,
consultant, and survey specialist
Incorporate revisions from review
Review by survey staff
Send revisions to EPA project officer for review by
EPA survey liaison officer
Perform limited pretest in Raleigh area
Revise instrument based on pretest
Submit draft instrument to EPA for review
Revise instrument based on additional pretest
Submit Office of Management and Budget
(OMB) package
Incorporate OMB suggestions
OMB approval
August 5
August 10
August 17
August 20
August 22
August 24
August 26
August 28
September 2
September 6
October 9
October 27
November 5
After the instrument was developed, it was administered on a limited pre-
test basis in the Research Triangle Park, North Carolina, area. Further lim-
ited pretesting of the instrument was completed in Pittsburgh after the Office
of Management and Budget (OMB) package was submitted for EPA review.
The Research Triangle Park pretest was conducted on people from the
Pittsburgh area to detect major faux pas in the instrument that Triangle-area
residents could not perceive. As a result of this pretest, several recreation
sites were added to the site list, the groups of activities were rearranged,
and a better map was developed. Most of the benefits from the pretest came
from finding flaws in the logic of the questionnaire. The pretest was espe-
cially helpful in determining what subsequent questions were appropriate for
zero bidders and for bidders who gave a zero to only certain parts of the
questionnaire.
3-8
-------
A limited budget prevented extensive pretesting in the target area. In
future surveys this activity should be budgeted. Because of the logical
consistency desired across all items in the questionnaire, a pretest in the
survey area would reveal potential logical inconsistencies only sample area
residents could expose via their responses. Researching the river and the
sample area was a viable substitute-, but a pretest in Pittsburgh would have
been a valuable complement.
3.4.2 Retaining Field Supervisors and Hiring Interviewers
The project used two field supervisors experienced in hiring and training
interviewers and in managing survey fieldwork to supervise and carry out the
count-and-list task and to recruit the field interviewers who performed the
household interviewing task. Because much of the cost of a data collection
effort is due to count-and-list activities and to interviewer recruiting, using
offsite field supervisors made the project's field operations more economical.
The survey task leader closely monitored the field supervisors in the count-
and-list and recruiting activities, which were carried out during the week of
October 19, 1981.
Project staff and the field supervisors worked together to select the
interviewers from among experienced applicants who had previously performed
well on similar surveys. Top prospects in the Pittsburgh area were screened
by telephone to verify general qualifications, availability, and interest.
During the count-and-list activity, the field supervisors interviewed some of
the best qualified applicants in person. Personal and work references were
checked before final selections were made. Relevant selection criteria included
interest in the objectives of the study, availability of dependable transporta-
tion, perceived ability to relate well to the sample population of interest,
input from personal and work references, and interviewing skills (e.g.,
ability to read questions clearly, to follow instructions, to use nondirectional
probes, to record responses accurately and legibly, etc.).
The selected interviewers were nine professionals who had extensive
experience in household surveys, focus groups, census work, and a variety
of other interviewing activities. These interviewers performed admirably
throughout the data collection process, overcoming inclement weather, a few
irate refusals, and an approaching holiday season. This was done with a
refreshing enthusiasm and reinforced the confidence of the project team
members. The interviewers were aware of all the things that can possibly
bias a respondent and were careful to follow the procedures outlined in the
manual and covered in the training session. In summary, the importance of
using experienced, professional interviewers cannot be overstated.
3.4.3 Counting and Listing of Sample Segments
Two field supervisors and two experienced interviewers conducted £ll
counting and listing of sample segments. This task involved:
Locating the segment
Identifying segment boundaries
3-9
-------
Counting the housing units
Listing all eligible housing units.
The count-and-list task was completed in 1 week and the materials
returned for an in-house check and preparation of interviewer assignments.
Appendix B shows samples of the results from the count-and-list activities.
Details of how these materials were used by the interviewers are provided in
the Field Interviewer's Manual, available from Research Triangle Institute.
3.4.4 Developing Field Manuals and Conducting Interviewer Training
Because the interviewers were supervised from the Research Triangle
Park during the household interviewing phase, a high degree of administrative
organization of field personnel was required for the project. Interviewers
were carefully informed of reporting and communications channels, procedures,
schedule requirements, documentation of nonresponse, reassignments, quality
control techniques, and other operating procedures required to complete the
project in a timely, cost-effective manner. The Field Interviewer's Manual
provided the details of the organization of the field procedures and covered
the following topics:
Purposes and sponsorship of the project
Role of the interviewer
Data collection schedule
Field sampling and locating procedures
Contacting and obtaining cooperation from sample members
Reporting results of attempts to secure interviews
Documentation of nonresponse
Validations, field edits, and other quality control procedures
Disposition of completed cases
Completion of administrative forms (e.g., field status reports,
reassignment forms, and production and expense reports)
Communications with central office staff.
In addition to the Field Interviewer's Manual, a series of administrative
forms was developed including a household control form (see Appendix B),
which served the following functions:
Provide assignment information for the interviewer (i.e.,
sample household address).
3-10
-------
Provide the interviewer with an introductory statement explain-
ing the survey.
Provide appropriate household enumeration questions and
queries to obtain demographic data on persons in the sample
household.
Provide the interviewer with instructions for selecting a house-
hold member to be interviewed.
Require the interviewer to document all attempted and success-
ful contacts with the sample member.
Provide an appropriate set of result codes for describing
interim and final results for each case.
Require the interviewer to record certain information required
for validation of completed interviews and noninterviews.
The training materials developed for the project included background on
benefits analysis and administrative procedures. The Interviewer's Manual
and a copy of the questionnaire were sent to the interviewers prior to their
classroom training. A specified amount of time was authorized for advance
study, and interviewers were expected to read the manual and specifications
prior to the training session.
3.4.5 Training Session
The extensive experience of the interviewers enabled the project team to
focus on the unique aspects of the project and to highlight the technical
details of the interviewing procedures. The agenda, shown in Figure 3-3,
shows the variety of topics covered in the 2-day session on November 11 and
12, 1981.
In addition to covering the project objectives, the training session pro-
vided an opportunity for personal interaction with the interviewers. The ses-
sion focused on benefits, EPA water policy, the water pollution basics, and
mock interviews with all versions of the questionnaire. The mock interviews
included zero bidders, recalcitrant and reluctant bidders, use of the payment
card, and procedural problems that might be encountered. The interviewers
were reminded not to provide supplemental information but to reread an item
as many times as necessary. Each interviewer received a healthy dose of
information on benefits methodology and the important policy implications of
the project. The participation by the project officer in the training also con-
veyed the feeling that the interviewers were important to the successful com-
pletion of the survey.
3.4.6 Conducting Household Interviews
Face-to-face interviews were conducted between November 13 and
December 20, 1981. Conducting the interviews involved a series of inter-
related operations, which included taking steps to obtain the desired number
3-11
-------
Field Interviewer Training Session Agenda
Study for Estimating Recreation and Related Benefits
of Water Quality
November 11,1981
9:00 a.m.
9:10 a.m.
9:15 a.m.
9:45 a.m.
10:15 a.m.
11:00 a.m.
11:15 a.m.
12:00 a.m. -
1:00 p.m.
1:00 p.m.
1:30 p.m.
2:30 p.m.
2:45 p.m.
3:00 p.m.
5:00 p.m.
November 12, 1981
9:00 a.m.
9:30 a.m.
10:00 a.m.
10:30a.m.
12:00 a.m.-
1:00 p.m.
1:00 p.m.
2:00 p.m.
Introduction of RTI staff and field interviewers Kirk Pate
Review of training agenda Kirk Pate
Project administrative procedures Kirk Pate
Break/picture taking and IDs
Explanation of the Benefits Study Bill Desvousges
Overviewof field interviewer responsibilities Kirk Pate
Locating sample housing units Kirk Pate
Lunch
Completing household control form and selecting
sample individuals Kirk Pate
Questionnaire administration Kirk Pate
Demonstration interview Kirk Pate/
Bill Desvousges
Break
Mock interview—Version A Group
Adjourn
Questions and answers/discussion of yesterday's Kirk Pate/
session Bill Desvousges
Water pollution: Dimensions of a problem Bill Desvousges
The Benefits Study Dr. Ann Fisher
Mock Interview—Version C Group
Lunch
Questions and answers
Distribution of assignments
Adjourn
Figure 3-3. Field interviewer training session agenda.
of interviews, instituting interviewer assignment and reporting procedures,
making initial household contacts and obtaining cooperation, enumerating
household members, and administering the instrument.
Initial assignments of cases to interviewers were made on the basis of
each interviewer's location and characteristics. Generally, assignments were
made on the basis of the interviewer's geographic proximity to the sample
segments. That was, of course, a cost-effective practice and usually resulted
in interviewers sharing some characteristics with the people to be inter-
viewed .
Efforts were made to equalize interviewer workloads; however, individual
assignments were made after careful consideration of factors related to the
difficulty of the areas assigned to each. Based on an assumed equal distri-
bution of cases per interviewer, the average number of cases initially assigned
per interviewer for the 6-week data collection period was 40. Under Number
of Cases Assigned, Figure 3-4 shows the final case load for each interviewer
after adjustments in the field.
3-12
-------
RTI Project 2222-2 FIELD DATA COLLECTION WEEKLY STATUS REPORT
ESTIMATING BENEFITS OF WATER QUALITY
Week I 6 Dates Covered: 12 / 15 / 81. to 12 / 21 / 81 Date Report Prepared: 12 / 22 / 81
FI Name
TOTAL
Number
of Cases
Assigned
42
39
57
36
64
48
61
6
44
20
397
*Slatus Codes:
02- No Enumeration Eligible
06 Enumeration Refused
05 Language Barrier
06 Vacant SHU
07 Not an Sllll
08 Other
No
Action
Taken
0
0
0
0
0
0
0
0
0
0
0
Cases in
Progress
42
39
57
36
64
48
41
6
46
20
397
t
Enumeration
Final Status Code'1
02
1
3
1
1
1
1
I
9
06
3
1
6
1
,
1
2
17
05
^Status Codes:
06
4
4
2
1
6
1
IB
07
2
1
3
08
1
,
1
3
Home 20 Completed Interview
22 Interview fireakoff
23 Hot at Home/No Contact
24 Refused
25 Language Barrier
26 Other
Interview
Final Status Code4*
20
29
78
60
31
56
14
33
5
36
13
303
22
,
1
2
23
3
7
1
1
6
1
1
1
14
24
1
4
3
,
3
2
3
I
5
1
26
25
1
1
Distribution Lint:
26
,
1
1
3
B. DesvousgeR
K. Pate
D. Smith
SOC Dept. 2 Files (2222-2)
Figure 3-4. Summary of completed interviews.
Once interviewer assignments were identified, interviewers' names were
associated with each household control form. Thus, manual control of assign-
ments was established and maintained. This control of assignments was
updated weekly on the basis of status reports and receipt of completed work.
Once assignments were issued at the conclusion of training, rigid report-
ing procedures were implemented. At a specified time each week, each,inter-
viewer telephoned the survey specialist and reported the status of each as-
signed case, using the current status code from his copy of the household
control form. The staff member entered the codes on a field status form for
the reporting period and discussed each active case showing no progress or
indicating a problem.
3.4.7 Initial Contacts and Obtaining Cooperation
Obtaining cooperation depended upon the persuasiveness of interviewers,
who, as a result of training and experience, were able to communicate to
respondents their own convictions regarding the importance of the study.
There was no major problem in obtaining respondent cooperation. Inter-
viewers indicated that people who were uncooperative for this project were no
different from other survey experiences in the Pittsburgh area.
3-13
-------
3.4.8 Household Enumeration
Once the interviewer made contact with an eligible household member, he
proceeded to enumerate all individuals residing in the household. This pro-
cedure ensured that each age-eligible individual was given a chance to be
selected for interviewing. All reasonable field efforts were made to interview
all sample individuals. The following situations were anticipated and were
handled as indicated below:
Field efforts were discontinued once it was determined that a
sample member had moved outside the sample counties.
Field efforts were discontinued upon learning that sample
members were deceased or institutionalized.
When non-English-speaking respondents were encountered, an
attempt to identify a close relative to serve as interpreter was
made in an effort to complete the interview. There was only
one interview with a language barrier, so no special effort was
made in this area.
An initial call and at least three additional callbacks were made
at different times of the day and different days of the week in
an attempt to establish contact with sample individuals to com-
plete the interview.
Contacts with neighbors were made after the second call to
obtain "best time to call" information.
. The enumeration process was facilited by the design of the household
control form (see Appendix B), which contained procedural instructions,
questions, and recording mechanisms to assist the interviewer in identifying
and listing household members and determining sample status. Procedures for
assigning appropriate unique identifiers were also included.
3.4.9 Interviewing Procedures
Interviewers were instructed to attempt to conduct interviews immediately
following the enumeration process when the sample member was identified and
if he were available. If necessary, appointments were made to return at a
time convenient for the sample member. All interviews were completed by
means of a face-to-face interview. The average length of a completed inter-
view was approximately 35 minutes.
Table 3-2 highlights the final tally from the field data collection. The
final number of sample housing units was 397 due to the discovery by field
interviewers of 13 housing units not listed during the listing phase of the
3-14
-------
Table 3-2. Final Distribution of Sample Housing Units
Result category
Number
Percentage of SHUs
Out-of-Scope SHUs.
Vacant
Not an HU
21
18
_3
21
4.53
.76
5.29 (of 397
SHUs)
No enumeration eligible at home
Enumeration refused
Other enumeration result
Completed interviews
Interview breakoff
Sample individual not at home
Sample individual refused
Language barrier
Other interview result
W f V
9
17
3
303
2
14
24
1
3
376
2.39
4.52
.80
80.59
.53
3.72
6.38
.27
.80
100.00
^Out-of-scope refers to sample housing units not included in response rate
calculation.
In-scope refers to sample housing units included in response rate calcula-
tion.
project.* The interviewers completed 303 interviews during the data collec-
tion period of November 13 through December 20, 1981--two interviews short
of the desired goal. The response rate (80.59 percent) was ever so slightly
above the anticipated 80 percent rate, while the refusal rate equaled 10.90
percent.
The count-and-list process is an imperfect one because interviewers
are not required at that stage to actually knock on each door in an
effort to identify housing units (HUs). Procedures for discovering HUs
missed during the listing process are implemented during the household
interviewing stage. The inclusion of each missed HU in the survey
improves the statistical representativeness of the initial sampling frame.
3-15
-------
Twenty-three sample households either did not complete the interview or
refused to cooperate. These were 23 cases in which either no one was at
home to provide the enumeration or the enumeration of the household members
was obtained but the sample individual was never available to complete the
interview. The crush of the Christmas holidays and a week of inclement wea-
ther conditions prevented resolution of these cases. Without either of these
hindrances, it is not unreasonable to expect that an additional 15 to 20 inter-
views could have been obtained by the interviewers.
3.4.10 Interviewer Debriefing
The project staff and the project officer conducted a 1-day debriefing
session in mid-December. This session provided an opportunity for the
interviewers to evaluate survey procedures and the questionnaire relative to
their other interviewing experiences. The overall conclusion of the debriefing
session was that the questionnaire was generally easy to administer and that
there were few major problems.
The comments that follow represent general impressions and evaluations
of the interviewers. There is no way to validate them, but they certainly
provided valuable insight for the project staff. The debriefing session was
highly valuable for project staff, both in terms of current project and ideas
for handling problems in future efforts.
Training Materials
More background on water pollution and recreation would have
been helpful.
Background and policy setting provided "keys" for getting in
doors. Interviewers simply found it easier to pique people's
interest because they understood the project objectives better.
More explanation of the payment vehicle—how people are cur-
rently paying for water pollution in higher prices and taxes--
would have been helpful to the interviewers.
Interviewing Process—General Comments
Count-and-list maps and materials worked well.
Drinking water was a major concern of many people, especially
the elderly. This was not addressed in our instrument because
of the recreation focus.
There were occasions in which a spouse intervened or cri-
tiqued the interview responses of the sample individuals. The
interviewers felt, however, that the respondents gave responses
that reflected the households' views.
3-16
-------
Refusals were generally three types: busy, timid, or nasty.
This was no different from other household surveys, according
to interviewers.
Thirty minutes was the ideal length both in terms of adminis-
tration and getting critical cooperation of respondents.
Evaluation of Specific Parts of Questionnaire
Section A, with activities listing and sites, worked very well.
Easy to administer and established interest of many respon-
dents—especially recreators.
Section B introduction is still wordy, especially B-1 intro-
duction. "Season" ticket needed after advance in introduction.
B-2. needed a skip pattern for nonrecreators.
Few problems with B-3 or B-4.
There was some confusion in B-5 as to how to interpret zero
response to this question. Does it mean no change or a com-
plete reduction? This will require careful attention in analy-
sis. There was also some confusion over how the water quality
might be bad sometimes and not at other times.
Few problems with B-6.
There was some concern in B-7 whether the amount given was
the total amount already given, a new amount independent of
other amounts, or an amount in addition to those given earlier.
Visual Aids
Map and water quality ladder worked well.
Visual aid showing how (but not how much) people are cur-
rently paying was needed to aid less perceptive respondents.
Rank order card design was effective. People had little trou-
ble connecting levels and dollar amounts, but cards should
have been larger for easier use.
Numbers on scale in water quality ladder were too small for
elderly respondents.
There could have been several more sites on the site listing.
A better visual aid is needed for "use—might use," perhaps
with color and/or larger print.
3-17
-------
Questionnaires
The direct question of willingness to pay without a payment
card was the most difficult version to administer because
people often seemed uncomfortable without some aid (consistent
with Mitchell and Carson's [1981] finding). The payment card
was the easiest to administer.
The bidding games usually reached an amount quickly as
respondents supplied amounts after seeing how the process
worked. The $125 starting point for each level was high rela-
tive to many bids making this slightly embarrassing for the
interviewers to administer. Reason for high amount was to test
for bias due to starting points.
Specific suggestions for revising the questionnaire are pre-
sented in Appendix D.
3.4.11 Data Receipt, Editing, and Keypunching
The last phase of the survey process required careful handling of the
survey data, coding, editing, and keypunching. Appendix B provides the
details of this process. In general, completed questionnaires were received
from the interviewers on a flow basis during the data collection period.
In-house editing was performed by the survey specialist for the purpose of
detecting any irregularities. As necessary, irregularities were discussed with
the appropriate interviewer.
The only major coding of responses that was required involved the
occupation questions. The verbatim responses were coded into the occupation
classes from the Bureau of the Census.* Household control form and ques-
tionnaire data were keypunched on cards and verified before analysis began.
*March 1971, publication from the Census of Population, U.S. Department
of Commerce, Washington, D.C. 20233.
3-18
-------
CHAPTER 4
CONTINGENT VALUATION DESIGN AND RESULTS:
OPTION PRICE AND USER VALUES
4.1 INTRODUCTION
Application of the contingent valuation approach, also referred to as the
direct survey approach in environmental economics, asks individuals their
dollar valuation of a nonmarket "commodity"—i.e., some good or service not
traded in an actual market.* In environmental applications, the analyst must
create a hypothetical market by describing how individuals would pay for
specific improvements in environmental quality. For this benefits study of
the Monongahela River basin, the contingent valuation design used a household
survey to ask individuals' valuation in terms they could understand—terms
that translate the water quality improvements into additional activities, such
as swimming and recreation fishing, that individuals could undertake along
the Monongahela River.
Contingent valuation offers the analyst considerable flexibility in design-
ing the "commodity" to be valued in the hypothetical market. At the same
time, however, it requires that he take considerable care in designing the mar-
ket so it is both credible and understandable to the respondent. Indeed,
research suggests that contingent valuation results may be sensitive to the
question formats used to elicit an individual's valuation, the mechanism used
to obtain the hypothetical payments (payment vehicle—e.g., user fee or utility
bill increase), and the interviewers used to conduct the survey. To give use-
ful results, the survey design must successfully surmount these influences.
The contingent valuation design for estimating the recreation and related
benefits of improved water quality in the Monongahela River used research
methods in fields ranging from survey and sample design to resource econom-
ics. This chapter traces the origins of the design, describes the survey
questionnaire, characterizes the survey respondents, and presents the results
on option price and user value for the water quality improvements.
Section 4.2 reviews survey design issues, paying close attention to poten-
tial biases in contingent valuation research, and Section 4.3 describes major
components of the survey questionnaire, including the design for determining
*The interpretation of the valuation requested of respondents will depend
upon the nature of the question. For example, whether a willingness-to-pay
or willingness-to-sell measure is elicited will depend on the property rights
and nature of the change proposed in the question.
4-1
-------
differences in techniques to elicit option price responses, the selection of a
payment vehicle, and the design of tests for achieving plausible results.
Section 4.4 characterizes the survey respondents and the main groups of
interest among them (users and nonusers of the river and people who refused
to pay any amount for improved water quality), Section 4.5 describes the esti-
mated values for option price and the statistical analyses of these estimates,
and Section 4.6 provides the same information for user values. Section 4.7
summarizes the chapter's main findings.
4.2 A REVIEW OF DESIGN ISSUES IN CONTINGENT VALUATION SURVEYS
In constructing a hypothetical market, the contingent valuation approach
defines the commodity to be valued, specifies how the exchange would occur,
and describes the other structural elements of the market. Brookshire,
Cummings, et al. [1982] have labeled this process as "framing the question,"
or as simply setting the context presented to respondents as part of the con-
tingent valuation experiment. As with almost any type of experimental design,
the context can influence the outcome. For example, within the range of
different contingent valuation contexts, an individual might participate directly
in a bidding procedure to elicit willingness to pay for the hypothetical com-
modity, might directly reveal this value (with or without the aid of some type
of payment card), or simply might evaluate (rank) various outcomes of the
hypothetical market, as in the case of the contingent ranking format.
Partially because of this range of contexts, the various attempts to
classify the methods for implementing the contingent valuation approach—and
their design features—have created considerable confusion. Therefore, to
consider the context of the contingent valuation approach used for the
Monongahela River basin, this section is organized according to the approach's
potential biases. These biases are not neatly compartmentalized; rather, they
are overlapping and in some cases interrelated. (Indeed, one analyst's
strategic bias is another's hypothetical bias.) At the risk of blurring the
boundaries between compartments, the section notes the most important of
these interrelationships. The boundaries themselves may, in large part, be a
question of judgment.
4.2.1 Hypothetical Bias
Hypothetical bias in contingent valuation surveys is the bias attributable
to the use of a hypothetical, not an actual, market situation, and it arises
when individuals cannot or will not consider the questions in a manner that
corresponds to how they would treat the actual situation. Consequently, we
can expect that they provide inaccurate answers to the contingent valuation
questions about it. Mitchell and Carson [1981] argue that hypothetical bias
may increase respondents' uncertainty and ambivalence about the hypothetical
experiment or induce them to provide answers that they perceive are socially
desirable. In general, hypothetical bias may result in respondents rejecting
or refusing to participate in the contingent valuation experiment, but the net
effect is to increase the statistical variance and to lessen the reliability of the
estimated willingness-to-pay amounts.
4-2
-------
The empirical evidence on hypothetical bias is somewhat mixed, with some
studies hindered by it and others showing no evidence. To test for several
biases, Bohm [1971] designed an experiment that compared alternative bidding
and payment schemes for the valuation of public television. Several alterna-
tives were provided to respondents, and, in some cases, the respondents were
actually given money to spend on several alternatives to public television.
Bohm compared results from the group that answered hypothetical willingness-
to-pay questions with those from a group that actually had to pay for public
television. The willingness-to-pay bids from respondents who had to pay for
public television were less, and significantly different, than those from
respondents who were simply asked how much they were willing to pay.
These results imply that hypothetical and strategic behavior were present in
the contingent valuation approach.
Mitchell and Carson [1981] question Bohm's [1971] conclusion on hypo-
thetical bias based on a reinterpretation of his statistical evidence. Bohm's
results showed that only one group out of six had different mean values when
structured across different types of information and market actuality. The
group that did exhibit higher willingness-to-pay amounts was also the group
that had higher incomes, which, Mitchell and Carson argue, may account for
the size of its mean willingness-to-pay bid. This same group also had one
outlier that raised the mean bid considerably. If the outlier is removed, the
mean payment is reduced to a level at which it is no longer a statistically sig-
nificant difference in the means.
Bishop and Heberlein [1979] designed a mail survey that compared hypo-
thetical willingness-to-pay amounts and actual willingness to sell. In this
study respondents were mailed checks in randomly selected amounts and re-
quested to sell a hunting license they had previously purchased. The authors
found that the amounts the respondents were willing to accept for their hunt-
ing licenses when presented with an actual check were considerably less than
the willingness-to-pay amounts they gave in the hypothetical bidding game
portion of the experiment. However, the results of the hypothetical and simu-
lated market experiment suggested that the hypothetical market underestimated
willingness to pay relative to the actual estimates from the simulated market.
The Bishop-Heberlein findings suggest hypothetical bias may be a significant
problem in contingent valuation survey design, but the implications of their
research may be limited by their experimental design.
Significantly, the results of several studies have indicated that hypothet-
ical bias may contribute to the considerable variability in contingent valuation
estimates of willingness to pay. For example, the Brookshire, Ives, and
Schulze [1976] and Brookshire et al. [1979] air quality studies explain less
than 10 percent of their bid variation by either socioeconomic variables or
changes in the level of the environmental good that the survey was designed
to measure.
While not invalidating the approach as a means of measuring consumers'
willingness to pay, the potential for hypothetical bias in contingent valuation
surveys indicates the need for considerable attention in the instrument design
phase to provide a credible survey questionnaire. The respondent must be
4-3
-------
able to perceive the experiment as a realistic approach to measuring the good
under consideration. Aizen and Fishbien [1977] have shown that the more
closely a hypothetical experiment corresponds with actual situations, the
greater the chance of reducing hypothetical bias. Mitchell and Carson [1981]
argue that reducing hypothetical bias in a contingent valuation survey instru-
ment does not necessarily lead to increased probability of incurring strategic
bias (where a respondent attempts to influence results) or other types of
biases. Rather, they suggest that a hypothetical experiment in which the
market realism is high and consequence realism is low will reduce or minimize
each type of bias. That is, respondents will perceive that a hypothetical situ-
ation closely corresponds to a real market situation (high market realism), but
they will not perceive the nature of the consequences of the hypothetical
experiment (to themselves) to the extent that they will attempt to influence
the outcome (low consequence realism).
The Mitchell and Carson position differs considerably from that of Schulze
et al. [1981], who argue that the potential for strategic bias increases when
hypothetical bias is reduced. Mitchell and Carson present a viable alternative
to the Schulze position in showing that both biases can be overcome in survey
design. Specifically, Mitchell and Carson were able to explain a considerably
larger percentage of the variation in willingness to pay than could authors of
most earlier contingent valuation studies and did not find evidence of strategic
behavior on the part of respondents. Furthermore, the Mitchell and Carson
results are particularly encouraging because their hypothetical market design
offered national water quality as a product, an unconventional situation that
should be particularly sensitive to hypothetical bias.
4.2.2 Strategic Bias
The concern for strategic bias is usually attributed to Samuelson [1954],
who suggested that any attempt to value public goods will be plagued by in-
centives on the part of individuals or respondents to behave strategically.
Samuelson argued that, if individuals perceive they will be able to obtain a
public good and enjoy its consumption, they may indeed try to obtain this
public good by not revealing their true preferences. The thrust of the
Samuelson argument for questionnaire design is that, depending on how
respondents perceive the consequences of the hypothetical experiment, they
may behave strategically. For example, an environmentalist who thinks his
bid might affect some environmental policy may bid higher than his true
willingness to pay in order to increase the average bid, provided he knows
he will not have to pay based on these bids. Alternatively, if an individual
believes his payment will be based on responses given to the questions, there
will be incentives to conceal true preferences provided the individual is
reasonably sure the good will be provided.
The empirical evidence on strategic behavior in contingent valuation sur-
veys has generally found that strategic behavior is not a major problem for
interpreting willingness-to-pay amounts. For example, Brookshire, Ives, and
Schulze [1976] and Rowe, d'Arge, and Brookshire [1980] attempted to design
experiments that would indicate the existence of strategic bias. In these
experiments, respondents were asked to reveal their willingness to pay for
4-4
-------
changes in a public good, which, if provided, would in turn require them to
pay their share of the mean of all bids. Brookshire et al. [1979] show that,
for respondents to engage in strategic behavior in the type of situation used
in the Brookshire and the Rowe, d'Arge, and Brookshire studies, they would
have to know not only the amounts that other individuals had bid, but also
the number of bidders who had already been asked and their mean bid. Both
studies concluded that strategic bias was not evident in the sample data gener-
ated when respondents were told they would have to pay the mean of the
sample. The Brookshire test for strategic bias examined the distribution of
the bids, arguing that strategic bias leads to a bimodal distribution in which
the means for environmentalists are concentrated in the high values of the
distribution while the means for nonenvironmentalists fall primarily at the other
extreme. The Rowe, d'Arge, and Brookshire test involved a more rigorous
statistical analysis but found no support for strategic bias after problem bids
were eliminated. This study also provided one group of respondents with
information on the sample mean bid after it had made its bid and allowed it to
change on the basis of this new information. The authors found that only
one respondent desired to change an overall bid. The complexity of the sur-
vey questionnaire used in the Rowe study, as well as the methods used to
screen observations omitting some bids from the sample, limits the generality
of the study results. A study by Brookshire, Ives, and Schulze [1976] also
found no evidence of strategic bias in an examination of the distribution of
willingness-to-pay amounts.
Mitchell and Carson [1981] argue that the distribution test used to indi-
cate strategic bias in these earlier studies is inappropriate because it is
impossible for most willingness-to-pay distributions to have standard normal
distributions. They argue that the likely distribution is a lognormal one, as
shown in their empirical results. Unfortunately, there are two problems with
the -Mitchell and Carson results on strategic bias. First, their sample was
subsegmented into groups by income levels, which could have influenced the
hypothesized relationship between willingness to pay and income. Second,
Mitchell and Carson's results were limited by a substantial number of zero bid-
ders and protest bidders who, given the limitations of the experimental design,
prevented them from eliciting additional information on true preferences.
A forthcoming report by Cronin [1982] on willingness to pay for improved
water quality in the Potomac River suggests the existence of strategic bias.
The design of this study partitioned respondents into groups based on whether
they would actually have to pay their bid through increased local taxes based
on the mean bid or would have to pay very little because the Federal govern-
ment would pay for most of it. A comparison across the two groups showed
statistically significant differences in the mean willingness-to-pay amounts that
are consistent with the presence of strategic bias. Some caution is needed in
interpreting the Cronin finding because of a poorly designed survey question-
naire and specification problems in the willingness-to-pay equation.
Based on the evidence that currently exists, strategic bias is not the
pervasive problem that researchers originally feared. However, it may be a
problem if the questionnaire design does not provide a low-degree-of-conse-
quence realism. Mitchell and Carson [1981] conclude that effectively designed
survey questionnaires can achieve the required degree of realism.
4-5
-------
4.2.3 Payment Vehicle Bias
Payment vehicle bias occurs when a respondent is influenced by the
method of payment selected for the contingent valuation study. A number of
different payment methods comprise the range of payment vehicles: user fees,
increases in utility bills, and higher consumer prices and taxes. To be effec-
tive, a payment vehicle must be realistic and familiar to respondents so they
consider it plausible and realize the implications of the implied payment fre-
quency for their total willingness to pay in a given time period. The ideal
payment vehicle would combine believability with a wide range of alternative
payment amounts.
The contingent valuation literature indicates very little about the influence
of payment vehicle bias. In the only study that systematically examined this
bias, Rowe, d'Arge, and Brookshire [1980] found that the type of payment
vehicle—utility bill or payroll deduction—had a significant effect on willingness
to pay. One likely consequence of a particular payment vehicle is that it may
condition respondents to a range of values their responses are expected to
take. For instance, when a user fee is selected as the payment vehicle, it is
quite possible that the respondent will think in terms of a usual range for user
fees. Thus, payment vehicle bias may actually show up as starting point bias,
discussed below. On the other hand, general resentment of taxes could lead
to "pure" payment vehicle bias, in which the respondent rejects the payment
vehicle itself.
4.2.4 Starting Point Bias
The contingent valuation literature has devoted more attention to the
question of starting point bias—the influence of the starting points used in
iterative bidding (or any contingent valuation procedure that uses starting
point "keys," such as the Mitchell-Carson [1981] payment card)—than it has
to the other biases. In an evaluation of willingness to pay for air quality in
the Farmington, New Mexico, area, for example, Rowe, d'Arge, and Brookshire
[1980] found strong evidence of the effects of starting points, with a respond-
ent's bid for improvements in visibility increasing by $0.60 for every $1.00
increase in the starting point.
Brookshire et al. [1979] also found starting point bias in some of their
alternative bidding situations. However, their starting point bias tests are
difficult to interpret because their study had very small sample sizes across
the alternative starting points, ranging from 2 to 16 respondents. Combined
with the substantial standard deviation for the mean responses, these small
sample sizes make it difficult to reject the null hypothesis that starting point
has no effect. Mitchell and Carson [1981] argue that the small sample size
may have had a greater impact on the study's inability to detect starting point
bias in the Brookshire et al. [1979] study than the researchers realized. In
addition, Mitchell and Carson [unpublished 1982] have also suggested that the
Greenley, Walsh, and Young [1981] study was also hindered by starting point
bias. The payment vehicles chosen by Greenley, Walsh, and Young inadvert-
ently set two different starting points for the bidding process.
4-6
-------
Several other studies—including those by Brookshire and Randall [1978],
Thayer and Schulze [1977], Randall et al. [1978], and Thayer [1981]--have
also tested for starting point bias in various degrees. These studies found
no evidence of influence on willingness to pay that could be attributed to dif-
ferent starting points. Unfortunately, the research design of some of these
studies was inadequate to sufficiently test for starting point bias. The
Randall study was not able to differentiate mean bids by starting points, and
several of the other studies tested starting points whose relative amounts were
too close to provide conclusive results.
In summary, the literature on starting point bias indicates that, when a
bidding game is used to elicit willingness to pay, the results can be influenced
by the starting point used in the bidding process, suggesting that tests for
starting point bias should be included in the research design. The Thayer
[1981] study provides both a simple test for the existence of starting point
bias and an adjustment for willingness-to-pay bids if starting point bias
exists. However, the assumptions implicit in Thayer's test may limit its prac-
tical application, since it assumes the respondent has a nonstochastic honest
bid.
4.2.5 Information Bias
Information bias is the influence on an individual's valuation that is
attributable to the amount of information given to respondents in the survey
questionnaire. The literature provides very little evidence on the extent of
information bias. Careful questionnaire design and thorough interviewer train-
ing to provide consistent and equal information to each respondent should
minimize this bias.*
4.2.6 Interviewer Bias
Interviewer bias is attributable to the effect of using different interview-
ers to elicit individuals' valuations. This bias can stem from one interviewer
being more effective than another, either in administering a bidding game or
in establishing rapport with the respondent. In his seminal research on
wilderness experiments in the Maine woods, Davis [1963] established a high
level of rapport with the respondents but performed all of the interviews him-
self. A recent study by Cronin [1982] was able to test for the existence of
interviewer bias and indicates that willingness to pay can be influenced by
the interviewer. But the design of the test was not sufficiently robust for a
conclusive result. The prospects for interviewer bias can be minimized with
training sessions and by using experienced professional interviewers. None-
theless, even when training is used, the research should examine the influence
of using different interviewers because this may serve to identify other influ-
ences on the bids that were not previously recognized.
This is an example of a bias category that is not easily distinguished
from the problems associated with "framing" the experiment.
4-7
-------
Table 4-1. Summary of Biases in Contingent Valuation Experiments
Type of bias
Definition
Studies that have
tested for bias
Summary of
current results
General
Hypothetical
Strategic
Instrument
related
Starting
point
Vehicle
I nformation
Interviewer
Error introduced by posing hypothetical
conditions rather than actual condi-
tions to an individual; response may
not be a good guide to actual actions
individual would take
One known test--
Bishop-Heberlein
[1979], Bohm
[1971)
Attempt by respondents to influence out-
come of study by systematically over-
or under-bidding so action favors
their true interests; strategic
responses depend on how payment scheme
is defined and whether it is believed
At least eight tests
(see Schulze,
d'Arge, and
Brookshire [1981]
for summary;
Cronin [1982])
Contingent valuation experiments using
bidding game format have started with
suggested payment and use yes or no
responses to derive final willingness
to pay; suggestion may be perceived as
appropriate bid
At least five tests
(see Schulze,
d'Arge, and
Brookshire [1981]
and Rowe and
Chestnut [1981])
Characteristics of proposed mechanism
for obtaining respondent's willingness
to pay may influence responses
Effect of information provided to
respondent on costs of action under
study or other dimensions of problem
may affect responses '
Responses vary systematically according
to interviewer
At least four tests
(see Schulze,
d'Arge, and
Brookshire [1981]
and Mitchell
and Carson [1981])
At least four tests
(see Schulze,
d'Arge, and
Brookshire [1981]
and Mitchell and
Carson [1981])
Two tests—
Desvousges, Smith,
and McGivney
[1982] and
Cronin [1982])
Some indication that
hypothetical nature
of question did
influence responses,
but could not dis-
tinguish this effect
from instrument-
related biases
Very little evidence
of strategic bias
except for Cronin
[1982]
Some differences in
opinion over impor-
tance of starting
point bias; Mitchell-
Carson [1981] feel
starting point bias
is important, and
Desvousges, Smith,
and McGivney [1982]
provide some support;
Schulze, d'Arge, and
Brookshire [1981]
feel it is more limited
Some evidence of
effects in at
least two studies
Limited evidence of
effects
No evidence of bias
Bias present
aThe definitions and results summarized in this table are based on Schulze, d'Arge, and Brookshire [1981],
Rowe and Chestnut [1981], and Mitchell and Carson [1981].
4-8
-------
4.2.7 Summary and Implications for Contingent Valuation Research Design
Table 4-1 summarizes the relevant research on potential biases in contin-
gent valuation studies discussed above. Based on this information, the
Monongahela River contingent valuation study was designed to test for starting
point bias. In addition, after the surveys were completed, the statistical
analysis examined the prospects for interviewer bias. The structure of the
survey attempted to control for information, vehicle, hypothetical and strategic
biases in the survey questionnaire.
4.3 QUESTIONNAIRE DESIGN
Questionnaire design is the most critical task in a contingent valuation
study. This section describes the questionnaire used to estimate the recrea-
tion and related benefits of water quality improvements for the Monongahela
River in Pennsylvania. Specifically, building on the sampling plan and survey
procedures discussion in Chapter 3 and on the contingent valuation survey
biases discussion in Section 4.2, this section explains the treatment of poten-
tial biases either as an element in the questionnaire design or as an objective
in the analysis of the resulting data.
4.3.1 Questionnaire Design: Part A
A key ingredient in successful contingent valuation surveys is establish-
ing credibility for the survey objectives (see Appendix D for a complete copy
of the questionnaire). The first component of the questionnaire has to achieve
this objective without biasing or offending the respondent. Part A in the
Monongahela River questionnaire attempted to achieve these goals by inquiring
about recreation activities the respondent had engaged in during the last year.
The first two questions dealt with boat ownership to determine if the respond-
ent had easy access to a boat for recreation purposes through either owner-
ship or "borrowing" rights. Ditton and Goodale [1973] found boat ownership
to be a significant factor in recreation attitudes and activities in Green Bay,
Wisconsin. This suggested a question that was unlikely to offend any
respondent. v
Following the boat ownership question, the interviewer presented the list
of outdoor recreation activities shown in Figure 4-1 and asked if the respond-
ent had participated in any of the activities within the past 12 months. The
list contains a wide range of activities, including those usually associated with
water recreation--boating, fishing, and swimming--and those that occur near
water—picnicking, biking, and sightseeing. The list is a subset of the activ-
ities listed in the 1977 Federal Estate Survey data base used in estimating the
travel cost model in Chapter 7. This activity matching was an attempt to pro-
vide additional compatibility between the methods.
A "no" answer to the participation question on the Monongahela question-
naire moved the respondent into the benefits section, while a "yes" response
initiated the site/activity matrix, illustrated in Figure 4-2. The interviewer
used the site/activity matrix to record the sites visited, the number of visits,
and the activities in which the respondent participated. The interviewer
provided the respondent with two addi'tional visual aids to facilitate this
4-9
-------
On or In
Water "
Near
Water"
-01 Canoeing, kayaking, or river running
02 Other boating
03 Sailing
04 Water skiing
OS Fishing
L 06 Swimming outdoors or sunbathing
r 07 Camping in a developed area
08 Picnicking
09 Walking to observe nature or bird watching;
wildlife or bird photography
10 Other walking for pleasure or jogging
11 Bicycling
12 Horseback riding
13 Hunting
14 Hiking or backpacking
15 Attending outdoor sports events (do not
include professional football or baseball)
16 Other outdoor sports or games
17 Driving vehicles or motorcycles off-road
18 Driving for pleasure
_19 Sightseeing at historical sites or natural wonders
Figure 4-1. Activity card.
Site Hues
Not Lifted
/
Site
Codec
No. of
Viiiti
-
CANOEING, KAYAKING, ETC.
01
01
01
01
01
01
01
01
01
01
01
01
01
01
01
01
OTIER BOATING
02
02
02
02
02
02
02
02
02
02
02
02
02
02
02
02
SAILING
03
03
03
03
03
03
03
03
03
03
03
03
03
03
03
03
WATER SKIING
04
04
04
04
04
04
04
04
04
04
04
04
04
04
04
04
FISHING
OS
05
OS
05
OS
OS
OS
05
OS
OS
OS
OS
OS
05
OS
OS
SWWMING, SUNBATHING
06
06
06
06
06
06
06
06
06
06
06
06
06
06
06
06
CAWING
07
07
07
07
07
07
07
07
07
07
07
07
07
07
07
07
PICNICKING
OS
08
08
08
08
08
08
08
08
OS
08
OS
08
08
08
08
1
g
aa
09
09
09
09
09
09
09
•09
09
09
09
09
09
09
09
0-
OTHER WALKING/ JOGGING
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
BICYCLING
11
11
11
11
11
11
11
11
11
11
11
11
11
11
11
11
WRSEBACK RIDING
12
12
12
12
12
12
12
12
12
12
12
12
12
12
12
12
g
13
13
13
13
13
13
13
13
13
13
13
13
13
13
13
13
HIKING OR BACKPACKING
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
5
IS
IS
IS
IS
15
IS
IS
IS
IS
IS
IS
15
IS
IS
IS
IS
OTHER OUTDOOR SPORTS
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
OFF-ROAD DRIVING/RIDING
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
PLEASURE DRIVING
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
siarrsEEiNG
19
19
19
19
»
19
19
19
19
19
19
19
19
19
19
19
Figure 4-2. Site activity matrix.
4-10
-------
discussion—a colored pictorial map of the area shown in Figure 4-3 and a list
of recreation sites (also shown on the map) displayed in Figure 4-4. The re-
spondent described the information requested for these sites or any other sites
visited. The data collected in Part A completed a recreation profile of the
respondent that could be used in the analysis phase and established a rapport
with him without influencing the main objective—benefit estimation. Part A
also reinforced the idea that a wide range of recreation site services is influ-
enced by water quality.
4.3.2 Benefits Measures: Part B
Part B of the Monongahela River questionnaire established the hypothetical
market by describing its institutional arrangements. In other words, this part
described the hypothetical market, the commodity to be valued, the payment
vehicle, and enacted the valuation experiment. The first section introduced
the setting for the hypothetical market:
The next group of questions is about the quality of water in the
Monongahela River. Congress passed water pollution control laws in
1972 and in 1977 to improve the nation's water quality. The States
of Pennsylvania and West Virginia have also been involved in water
quality improvement programs of their own. These programs have
resulted in cleaner rivers that are better places for fishing, boating,
and other outdoor activities which people take part in near water.
We all pay for these water quality improvement programs both as
taxpayers and as consumers.
In this study we are concerned with the water quality of only the
Monongahela River. Keep in mind that people take part in all of
the activities on Card 1 (Figure 4-1) both on and near the water.
Following the introduction, the interviewer handed the respondent the key
visual aid for the hypothetical market—the Resources for the Future (RFF)
water quality ladder developed by Mitchell and Vaughan at RFF and used by
Mitchell and Carson [1981] in their contingent valuation study of national water
quality (see Figure 4-5). Appendix E provides details on its construction.
The ladder's major attribute is that it easily establishes linkages between recre-
ation activities and water quality based on an index of technical water quality
measures and informed judgment. This type of linkage illustrates a crucial
distinction between the contingent valuation method and indirect techniques for
measuring the benefits of water quality. Specifically, rather than observing
the actual behavior of recreationalists, who demand different site services de-
pending on the level of water quality, it directly introduces the relationship
between activities and different water quality levels into the hypothetical
market.
After showing the key visual aid, the interviewer read the following text*
to describe the ladder and establish the desired linkages:
The words in all capitals are instructions for the interviewers only and
were not read to the respondent. They are included in the discussion for
completeness.
4-11
-------
Figure 4-3. Map of Monongaheia River
and other recreation sites.
Allegheny River:
01 Near Kittanning
02 Near Oakmont
03 Where Beaver River and Ohio River meet
04 Crooked Creek Park
OS Loyalhanna Lake
06 Keystone Dam
07 Lake Arthur in Moraine State Park
08 Ohiopyle State Park
09 North Park Lake (Near Alliwn Park)
10 Racoon Creek State Park
11 Youghiogheny River Lake Reservoir
12 Cheat River Lake
13 Ryerson Station
14 Yellow Creek
Monongaheia River Area:
15 Pittsburgh (The Point, Smithsfield Bridge, Braddock)
16 Where Monongaheia and Youghiogheny meet near McKeesport
17 Elrama
18 The Town of Monongaheia
19 Donora and Webster
20 Near Charleroi (Lock and Dam #4)
21 In the California-Brownsville Area
22 Maxwell Lock and Dam
23 Ten Mile Creek
24 Grays Landing—Greensboro (Lock and Dam #7)
25 Point Marion—Cheat River Area (Lock and Dam #8)
26 Morgantown
27 Hildebrand
28 Opekiska
29 Fairmont
Figure 4-4. Recreation sites.
4-12
-------
BEST POSSIBLE
IWATEH QUALITY
—10
WORST POSSIBLE
WATER QUALITY
SAFE TO DRINK
SAFE FOR SWIMMING
SAME FISH LIKE BASS
CAN LIVE IN IT
OKAY FOR BOATING
Figure 4-5. Water quality ladder.
Generally, the better the water quality, the better suited the water
is for recreation activities and the more likely people will take part
in outdoor recreation activities on or near the water. Here is a
picture of a ladder that shows various levels of water quality.
GIVE RESPONDENT CARD 4, "WATER QUALITY LADDER."
The top of the ladder stands for the best possible quality of water.
The bottom of the ladder stands for the worst possible water qual-
ity. On the ladder you can see the different levels of the quality
of the water. For example: (POINT TO EACH LEVEL—E, D, C,
B, A—AS YOU READ THE STATEMENTS BELOW.)
Level E (POINTING) is so polluted that it has oil, raw
sewage and other things like trash in it; it has no plant
or animal life and smells bad.
Water at Level D is okay for boating but not fishing or
swimming.
4-13
-------
Level C shows where the water is clean enough so that
gamefish like bass can live in it.
Level B shows where the water is clean enough so that
people can swim in it safely.
And at Level A, the quality of the water is so good that
it would be possible to drink directly from it if you
wanted to.
Following this description, the interviewer asked the respondent to use the
ladder to rate the water quality in the Monongahela River on a scale of 0 to
10 and to indicate whether the ranking was for a particular site, and, if so,
to name it.
Question B-2 introduced the respondent to a key element in the hypothet-
ical market: the distinction between user, option, and existence values.
Specifically, the interviewer gave the respondent the value card shown in Fig-
ure 4-6 and described each type of value. An attitudinal question punctuated
the descriptions of each type of value by inquiring how important the factors
of actual use, potential use, and no use were in valuing water quality. The
attitudinal responses to these questions—displayed on a five-point scale rang-
ing from very important to not important at all—reinforced the concepts, pro-
vided a break in the discussion, and presented an additional check for the
consistency in responses. The textual explanations for the three types of
values are:
Why We Might Value Clean Water in the Monongahela River
I. Use
Swimming Hiking
Fishing Sitting by the shore
Boating Hunting
Picnicking Driving vehicles off road
Birdwatching Jogging
II. Might Use
To have clean water in the river to use if you should decide in the future that
you want to use it.
III. Just Because It's There
Preserve for future generations.
Satisfaction from knowing that there is a clean river.
Satisfaction from knowing that others can enjoy the river for recreation.
Figure 4-6. Value card.
4-14
-------
Another important purpose of this study is to learn how much the
quality of water of the Monongahela River is worth to the people
who live in the river basin. In answering this question, there are
three ways of thinking about water quality that might influence
your decision. GIVE RESPONDENT CARD 5, "VALUE CARD." The
three ways are shown on this card.
One, you might think about how much water quality is worth to you
because you use the river for recreation. POINT TO PART I OF
VALUE CARD AND GIVE RESPONDENT TIME TO READ THAT
PART.
How important a factor is your actual use of the river in making a
decision about how much clean water is worth to you? CIRCLE
NUMBER.
VERY IMPORTANT 01
SOMEWHAT IMPORTANT 02
NEITHER IMPORTANT NOR
UNIMPORTANT 03
NOT VERY IMPORTANT 04
NOT IMPORTANT AT ALL 05
Another way you might think about how much clean water is worth
to you is that it is worth something to you to know that a clean
water river is being maintained for your use if you should decide,
in the future, that you want to use it. POINT TO PART II OF
VALUE CARD AND GIVE RESPONDENT TIME TO READ THAT
PART. For example, you might buy an advance ticket for the
Steelers or Pirates just to be able to go to a home game if you later
decide you want to go. Likewise, you might pay some amount each
year to have a clean water river available to use if you should de-
cide to use it.
In deciding how much clean water is worth to you, how important a
factor is knowing that a clean water river Is being maintained for
your use, if you should decide to use it? CIRCLE NUMBER.
VERY IMPORTANT. 01
SOMEWHAT IMPORTANT 02
NEITHER IMPORTANT NOR
UNIMPORTANT 03
NOT VERY IMPORTANT 04
NOT IMPORTANT AT ALL 05
4-15
-------
A third thing you might think about in deciding how much clean
water is worth to you is the satisfaction of knowing that a clean
water river is there. POINT TO PART III OF VALUE CARD AND
GIVE RESPONDENT TIME TO READ THAT PART. For example, you
might be willing to pay something to maintain a public park even
though you know you won't use it. The same thing could be true
for clean water in the Monongahela; that is, you might pay some-
thing just for the satisfaction of knowing that it is clean and that
others can use it.
In deciding how much clean water is worth to you, how important is
knowing that a clean water river is being maintained? CIRCLE
NUMBER.
VERY IMPORTANT 01
SOMEWHAT IMPORTANT 02
NEITHER IMPORTANT NOR
UNIMPORTANT. 03
NOT VERY IMPORTANT 04
NOT IMPORTANT AT ALL 05
The first paragraph of Question B-3, which introduces the payment vehicle to
the respondent, is presented below:
Now, we would like you to think about the relationship between
improving the quality of water in the Monongahela River and what
we all have to pay each year as taxpayers and as consumers. We
all pay directly through our tax dollars each year for cleaning up
all rivers. We also pay indirectly each year through higher prices
for the products we buy because it costs companies money to clean
up water they use in making their products. Thus, each year, we
are paying directly and indirectly for improvements in the water
quality of the Monongahela River.
I want to ask you a few questions about what amount of money you
would be willing to pay each year for different levels of water qual-
ity in the Monongahela River. Please keep in mind that the amounts
you would pay each year would be paid in the form of taxes or in
the form of higher prices for the products that companies sell.
This payment vehicle was selected because it corresponds with how people
actually pay for water quality, connotes no implicit starting point, and pro-
vides a vehicle that will bias the responses downward, if in any direction,
because of public attitudes toward increased taxes and higher prices.
The introduction continues with a reference to the value card (see Fig-
ure 4-6) and requests that initial amounts be based on actual use and poten-
tial future use—user and option values but not existence values. The present
overall level of water quality is desqribed as Level D, where it is clean enough
for boating.
4-16
-------
Question B-3 embodies the comparison of the alternative contingent valua-
tion methodologies. Specifically, by dividing the sample of 397 households into
fourths and using a different color survey instrument for each quarter, Ques-
tion B-3 compares the direct question method of eliciting willingness-to-pay
amounts, both with and without a payment card (illustrated in Figure 4-7)r to
the iterative bidding games with $25 and $125 starting points. Thus, the
questionnaire design provides an explicit test for starting point bias within
the iterative bidding game, as well as a test for differences between direct
questions and bidding games.
0
25
50
75
100
125
150
175
200
225
250
275
300
325
350
375
400
425
450
475
500
525
550
575
600
625
650
675
700
725
750
775
Figure 4-7. Payment card.
The payment card used in the direct question method was simply an array
of numbers representing annual amounts from $0 to $775 per year. This is in
contrast with the Mitchell and Carson [1981] payment card, which showed
amounts individuals paid for various public goods adjusted to correspond with
the respondent's income level. Mitchell and Carson split their sample to test
for the effect of the different types of public goods provided, but the sample
size in the Monongahela study was much smaller and already partitioned into
four groups, so no anchoring amounts were listed on the payment card.
Mitchell and Carson found no effect from the anchoring amounts, but this
result may have been hampered by their adjustment of the amounts to corres-
pond to the respondent's income level.
The hypothetical market queried the respondent for willingness-to-pay
amounts for three water quality levels:
Avoiding a decrease in water quality in the Monongahela River
from D, beatable, to E, not suitable even for boating.
Raising the water quality from D, beatable, to C, where game-
fish could survive.
Raising the water quality from C, fishable, to B, where people
could swim in the water.
Table 4-2 summarizes the formats for eliciting the option prices in the
contingent valuation questionnaire. (For details on question procedures, see
Appendix D, which contains a complete, copy of the survey questionnaire.)
4-17
-------
Table 4-2. Summary of Option Price Question Formats by
Interview Type
Interview type
Question format
Iterative bidding, $25
Iterative bidding, $125
Direct question
Payment card
To you (and your family), would it be worth $25
each year in higher taxes and prices for products
that companies sell to keep the water quality in
the Monongahela River from slipping back from
Level D to Level E?
To you (and your family), would it be worth $125
each year in higher taxes and prices for products
that companies sell to keep the water quality in
the Monongahela River from slipping back from
Level D to Level E?
What is the most it is worth to you (and your
family) on a yearly basis to keep the water qual-
ity in the Monongahela River from slipping back
from Level D to Level E, where it is not even
clean enough for boating?
What is the most it is worth to you (and your
family) on a yearly basis to keep the water qual-
ity in the Monongahela River from slipping back
from Level D to Level E, where it is not even
clean enough for boating?
The process for the direct question is very simple, with the interviewer asking
the respondent for an amount for each level and stressing that additional
amounts are being requested. The water quality ladder and the value card
are in front of the respondent while the market process is initiated. The same
procedure was used in the payment card format, with the only difference being
that the payment card was given to the respondent.
Table 4-2 also summarizes the procedure for the bidding games with start-
ing points. A similar procedure was used for both bidding games, the only
difference being the starting points used. In the bidding game, the inter-
viewer initiated the market process at the starting point and increased or de-
creased the requested amount until the respondent's maximum value was ob-
tained. This was repeated for each of the water quality levels, with emphasis
given to the additional nature of the amounts for the higher levels of water
quality.
To conclude this part of the hypothetical market, the interviewer asked
any respondent who gave a zero amount why that amount was given, as shown
in the question below. The purpose of this question was to distinguish be-
tween a true zero amount and a zero that essentially represented a protest
against either the experiment or some part of it, such as the Davment
vehicle. 7
4-18
-------
We have found in studies of this type that people have a lot of dif-
ferent reasons for answering as they do. Some people felt they did
not have enough information to give a dollar amount, some did not
want to put dollar values on environmental quality, and some objected
to the way the question was presented. Others gave a zero dollar
amount because that was what if was worth to them.
Which of these reasons best describes why you answered the way
you did? REPEAT REASONS IF NECESSARY AND CIRCLE NUMBER.
NOT ENOUGH INFORMATION 01
DID NOT WANT TO PLACE DOLLAR VALUE .... 02
OBJECTED TO WAY QUESTION WAS PRESENTED. . 03
THAT IS WHAT IT IS WORTH 04
OTHER (SPECIFY) 05
The next section of the questionnaire attempted to break down the option
price into its individual components of user and option values. The questions
and results for option value are described in detail in the following chapter,
so no additional discussion is provided in this chapter.
Part B contained two additional plausibility/consistency check questions
that asked what effect improved water quality in the Monongahela River would
have on visits to substitute sites and the Monongahela River sites. The an-
swers to these questions were structured by choices ranging from a change
(either increase or decrease) of more than five visits to no change or "don't
know."*
The last question in Part B asked the respondent to perform a contin-
gent ranking as specified by the text from the questionnaire. Figure 4-8 de-
picts one of the four combinations that the respondent was asked to rank.
This particular card shows the combination of the lowest level of water quality
and the lowest payment. Payment amounts of $50, $100, and $175 were paired
with beatable, fishable, and swimmable levels of water quality, respectively.
The survey design asked all respondents to rank the cards after participating
in one of the other valuation exercises. This design is a compromise resulting
from the limited resources available for sampling respondents and the objective
to compare as many methods as possible. A complete comparison would have
required an additional segmentation of the limited sample. Chapter 6 discusses
the theory and results from the contingent ranking experiment.
"These questions were suggested by the Office of Management and Budget
(OMB) in its review of the survey questionnaire.
4-19
-------
$100
BEST POSSIBLE
WATER QUALITY
"~ II ——
WORST POSSIBLE
WATER QUALITY
WATER QUALITY LADDER
SAFE TO OHINK , M
SAFE f OR 3'.V!W.USi:
GAME FISH LIKE SASS
CAN LIVE IN IT
E
'JKAY f I5K I
w^
H- i
Figure 4-8. Rank order card.
4.4 PROFILES OF SURVEY RESPONDENTS
Respondents in a contingent valuation survey should represent the popu-
lation of interest to provide plausible results. This section profiles the sample
respondents from the Monongahela River basin area and compares these pro-
files with Census data for the area as a check for representativeness. Users,
nonusers, zero bidders, and protest bidders are also profiled to assess the
role of socioeconomic and attitudinal characteristics in influencing any of these
groups.
Table 4-3 presents the characteristics of key groups of respondents in
the Monongahela survey. These data are for the 301 completed questionnaires
that provided valid responses. Two questionnaires were eliminated because
the respondents were unable to complete the session. One person was 97
years old and had difficulty seeing the cards; the other had trouble hearing
the interviewer.
4-20
-------
Table 4-3. Characteristics of Key Respondent Groups
User
Nenuser
Zero
Nonzero
Characteristic
Standard
devi-
ation N
Standard
devl-
Mean atlon N
Standard
devl-
Meah atlon N
Standard
devi-
Mean atlon N
Protest Bids8
Total
Standard
devl-
Mean atlon N
Standard
devl-
Mean atlon N
Five
county
region
In 1980 Sample
•U
1*yes, 0*no for ownerahlp or
use of a boat
1«yes, 0*no for participation In
any outdoor recreation In the
last year
Numerical rating of the
Monongahela River:
0=lowest, 10*highest
1*yes, 0=no If rating Is for a
particular site
Length of residence
Years of education
Race (1 If white)
Income
Age
Sex (1 If male)
0.23 0.43 94 0.12 0.32 207 0.11 0.32 108 0.18 0.39 193 0.15 0.37 58 0.16 0.36 301
0.95 0.23 94 0.38 0.49 207 0.38 0.49 108 0.88 0.48 193 0.50 0.50 58 0.56 0.50 301
3.87 1.98 89 3.77 2.01 132 3.51 1.76 61 3.92 2.07 160 3.63 1.68 38 3.81 1.99 221
0.34 0.48 94 0.08 0.27 207 0.07 0.26 108 0.21 0.41 193 0.10 0.31 58 0.16 0.37 301
6.83 0.95 94 6.80 1.02 207 6.82 0.95 108 6.80 1.02 193 6.74 1.18 58 6.81 1.00 301
13.06 1.96 86 12.61 2.12 177 12.38 2.20 86 12.93 1.99 177 12.77 1.73 47 12.75 2.07 263 10.96b 12.75
0.88 0.32 94 0.91 0.29 206 0.94 0.23 107 0.88 0.33 193 0.93 0.26 57 0.90 0.30 300 .92 .90
20,833 13,482 87 18,887 13,022 173 17,577 11,500 87 20,534 13,879 173 19,895 11,484 48 19,538 13,184 2€0 19,987b 19,538
38.93 16.20 94 51.87 17.85 207 54.55 16.91 108 44.06 18.07 193 52.60 17.27 58 47.82 18.34 301 45.6 47.8
.31 .46 94 0.39 0.49 207 0.35 0.48 108 0.37 0.48 193 0.44 0.50 58 0.36 0.46 301 .47 .36
SOURCE: U.S. Bureau of the Census. 1980 Census of the Population end Housing. Washington, D.C. 1982.
'Protest bios ere zero bids for reasons other than "all they could afford" or "that Is what It Is worth."
bStetewlde statistics.
-------
To develop a reasonably clear snapshot of the respondent group important
for the analysis of survey results, no adjustments for outliers are included in
the profile information. The first two columns of Table 4-3 compare users
and nonusers of the Monongahela River. The users are broadly defined based
on all respondents who reported a user value or visited one of the 13 Monon-
gahela River sites. This broader definition of user can be contrasted with a
narrow definition that includes only those respondents who visited a site.
The broader definition is used throughout this report because it allows for
the inclusion of some users who may have been prevented from visiting a
Monongahela site within the 12 months between November 1981 and November
1982 for medical or other personal reasons but still had some user value for
the services of the Monongahela. Tests indicated that the differences be-
tween the user definitions were insignificant. This broad definition explains
why a few Monongahela River users had not participated in an outdoor recrea-
tion activity in the second row of Table 4-3.
Results of t-tests for differences between the means of users and non-
users (shown in Appendix C) highlight some important distinctions that con-
tinue throughout the survey results. Users of the Monongahela River are
younger, are more likely to own a boat, and are more likely to have rated a
particular Monongahela River site than their nonuser counterparts. The water
quality ratings place the Monongahela above beatable, but a full point below
fishable, on the Water Quality Ladder (see Figure 4-5); however, the ratings
are not different between the two groups. There are no differences in educa-
tion, income, race, sex, or length of residence between users and nonusers.*
For these two groups t-tests for differences in means between zero and
nonzero bidders and a logit analysis comprise the analysis. Based on these
results, nonzero bidders were on average younger than zero bidders, earned
higher annual family incomes, were more likely to have rated the Monongahela
at a particular site, and have participated in outdoor recreation during the last
year. These results are consistent with the findings of Mitchell and Carson
[1981]. In addition, no significant differences existed between the groups in
terms of sex, education, water quality rating for the river, boat ownership,
and length of residence in the area. The protest bidders who rejected some
aspect of the contingent valuation approach had higher incomes and were more
likely to have participated in outdoor recreation in the last year than were
those with valid zero bids.
The questionnaire design also provided the respondent's reason for giv-
ing a zero bid. These responses are shown in Table 4-4 for the four elicita-
tion methods. The direct question method without the payment card yielded
most of the respondents who could not place a dollar value on water quality,
The percentage of woman respondents (64 percent) in the sample is
somewhat higher than in other studies—a somewhat surprising result since the
random procedure used to select the respondents should have given a more
even distribution. The respondent was asked to respond for the household,
which should reduce any potential bias.
4-22
-------
Table 4-4. Reasons for Zero Bids by Elicitation Method
Reason for zero bid
Payment
card
Direct
question
$25
iterative
bidding
was presented
That is what it is worth 12
Other 1
All they could afford 1
Government waste or 2
misuse of tax dollars
Industry pollutes so let 3
them clean it up
Taxes are too high already 2
Desire no increase in taxes 1
for something that does
not affect respondent
Total 27
10
5
3
0
3
0
33
7
5
1
2
1
0
22
$125
iterative
bidding
11
5
5
1
2
0
26
Total
Not enough information
Cannot place dollar value
Objected to way question
1
4
0
1
9
0
0
2
1
2
0
0
4
15
1
40
16
10
5
8
8
1
108
which roughly indicates the value of either the payment card or the starting
value in the bidding process. Approximately 40 percent of the respondents
bid zero because that is what they felt the water quality is worth. Some evi-
dence of the consistency in the response is indicated by the 10 respondents
who bid zero because that is all they could afford. These respondents tended
to be elderly persons living on limited incomes.
Table 4-5 shows the attitudinal information broken down for user, non-
user, and zero bids. These responses on the importance of water quality
were elicited during the discussion of the value card (see Figure 4-6) and
prior to the elicitation of the willingness-to-pay amounts. These responses
are very consistent with the earlier characteristics of the groups. Users and
nonzero bidders were much more likely to have given very or somewhat impor-
tant responses to the questions than were nonusers and zero bidders.
Table 4-6 completes the profiles of the three groups by highlighting the
respondents' willingness to identify themselves by certain labels. Several
interesting features are apparent from these attitudinal responses. The users
and nonzero bidders were much more likely to identify themselves as outdoors
persons than were nonusers and zero bidders. However, the differences be-
tween the groups is much smaller for the environmentalist label, with 26 per-
4-23
-------
Table 4-5. Degree of Importance.of Water Quality by Key Respondent Groups
K
Degree of Importance
of water quality
For own recreation
Very important
Somewhat Important
Neither important nor
unimportant
Not very Important
Not important at all
Total
For possible future use
Very important
Somewhat important
Neither important nor
unimportant
Not very important
Not important at all
Total
Even if never use river
Very important
Somewhat Important
Neither important nor
unimportant
Not very important
Not important at all
Total
User
Frequency %
47
28
4
10
5
94
49
34
5
3
3
93
49
29
7
6
2
93
50.0
29.7
4.3
10.6
5.3
52.1
36.2
5.3
3.2
3.2
52.7
31.2
7.5
6.5
2.2
Nonuser
Frequency %
48
36
33
46
43
206
70
53
26
33
25
207
74
70
21
27
15
207
23.3
17.5
16.0
22.3
20.9
33.8
25.6
12.6
15.9
12.1
35.7
33.8
10.1
13.0
7.2
Zero
bids
Frequency %
20
14
21
25
28
108
27
21
18
23
19
108
27
33
15
22
11
108
18.5
13.0
19.4
23.1
25.9
25.0
19.4
16.7
21.3
17.6
25.0
30.6
13.9
20.4
10.2
Nonzero bids
Frequency %
75
50
16
31
20
192
92
66
13
13
9
193
96
66
13
11
6
192
39.1
26.0
18.3
16.1
10.4
47.7
34.2
6.7
6.7
4.7
50.0
34.4
6.8
5.7
3.1
Protest bids3
Frequency %
16
9
14
11
8
58
16
14
13
9
6
58
19
18
9
9
3
58
27.6
15.5
24.1
19.0
13.8
27.6
24.1
22.4
15.5
10.3
32.8
31.0
15.5
15.5
5.2
Total
Frequency %
95
64
37
56
48
300
119
87
31
36
28
301
123
99
28
33
17
300
31.7
21.3
12.3
18.7
16.0
39.5
28.9
10.3
12.0
9.3
41.0
33.0
9.3
11.0
5.7
aProtest bids are zero bids for reasons other than "all they could afford" or "that is what It is worth.
-------
Table 4-6. Respondent Attitudes About Self by Key Respondent Groups
User
Attitude Frequency %
An outdoors person
A lot.
Somewhat
A little
Not at all
No opinion
Total
An environmentalist
A lot
Somewhat
A little
Not at all
No opinion
Total
Against nuclear power electric
plants
A lot
Somewhat
A little
Not at all
No opinion
Total
Concerned about water pollution
A lot
Somewhat
A little
Not at all
No opinion
Total
Willing to pay the cost required
to control water pollution
A lot
Somewhat
A little
Not at all
No opinion
Total
42
24
19
9
0
94
26
27
30
11
0
94
27
12
15
31
9
94
45
29
15
4
0
93
19
42
21
10
1
93
44.7
25.5
20.2
9.6
0
27.7
28.7
31.9
11.7
0
28.7
12.8
16.0
33.0
9.6
48.4
31.2
16.1
4.3
0
20.4
45.2
22.6
10.8
1.1
Nonuser
Frequency .%
50
56
38
63
0
207
38
56
51
59
2
206
45
19
23
79
40
206
87
71
29
17
3
207
31
59
50
58
8
206
24.2
27.1
18.4
30.4
0
18.4
27.2
24.8
28.6
1.0
21.8
19.2
11.2
38.4
19.4
42.0
34.3
14.0
8.2
1.4
15.0
28.6
24.3
28.2
3.9
Zero
bids
Frequency %
29
23
21
35
0
108
28
14
24
39
2
107
26
13
8
37
24
108
41
31
18
16
2
108
8
18
15
58
8
107
26.9
21.3
19.4
32.4
0
26.2
13.1
22.4
36.4
1.9
24.1
12.0
7.4
34.3
22.2
38.0
28.7
16.7
14.8
1.8
7.5
16.8
14.0
54.2
7.5
Nonzero bids
Frequency %
63
57
36
37
0
193
36
69
57
31
0
193
46
18
30
73
25
192
91
69
26
5
1
192
42
83
56
10
1
192
32.6
29.5
18.7
19.2
0
18.7
35.8
29.5
16.1
0
24.0
19.4
15.6
38.0
13.0
47.4
35.9
13.5
2.6
0.5
21.9
43.2
29.2
5.2
0.5
Protest bids3
Frequency %
20
12
9
17
0
58
20
10
10
15
2
57
15
9
3
20
11
58
28
17
8
5
0
58
6
10
9
28
4
34.5
21.0
15.5
29.3
0
35.1
17.5
17.5
26.3
3.5
25.9
15.5
5.2
34.5
19.0
48.3
29.3
13.8
8.6
0
10.5
17.5
15.8
49.1
7.0
Total
Frequency %
92
80
57
72
0
301
64
83
81
70
2
300
72
31
38
110
49
300
132
100
44
21
3
300
50
101
71
68
9
299
30.6
26.6
18.9
23.9
0
21.3
27.7
27.0
23.3
0.7
24.0
10.3
12.7
36.7
16.3
44.0
33.3
14.7
7.0
1.0
16.7
33.8
23.7
22.7
3.0
*Protest bids are zero bids for reasons other than "all they could afford" or "that is what it is worth."
-------
cent of the zero bidders indicating the closest identity with the label. This
is even more evident when only the protest zero bids are examined. Thirty-
five percent gave the strongest response, which is consistent with the fre-
quency responses shown in Table 4-4 for the reasons why people bid zero.
The most dramatic differences between respondents are evident in the willing-
ness to pay the cost required to control water pollution. Only 24 percent of
the zero bidders were willing to identify with this descriptive statement. This
consistency across different attitude responses suggests that the respondents
correctly perceived the contingent valuation experiment and gave careful re-
sponses that would not have been given if hypothetical bias were present. It
is also suggestive of the importance of attitudinal questions in contingent
valuation studies both for analysis purposes and as consistency checks.
Table 4-7. Logit Estimation of Zero Bids'
Independent variable
Coefficient
t-ratio
Derivative of the
probability
evaluated
at the mean
Constant
Sex
Age
Education
Income
Version B
Version C
Version D
Willing to pay cost of
water pollution (1 if
very much or somewhat)
Interviewer #1
Interviewer #2
Interviewer #3
Interviewer #5
Interviewer #7
Interviewer #8
Interviewer #9
-0.435
-0.522
0.036
-0.108
6.9 x 10~9
-0.319
-1.728
-0.665
-1.622
-0.625
1.095
-0.683
-1.158
-1.519
0.192
1.099
-0.251
-0.924
2.703
-0.867
0.326
-0.506
-2.407
-1.099
-3.185C
-0.627
1.318
-0.807
-0.913
-1.175
0.215
0.843
-0.042
0.003
-0.009
0.59 x 10~6
-0.026
-0.113
-0.050
-0.169
-0.044
0.128
-0.050
-0.072
-0.082
0.017
0.141
Note: Log of likelihood function = 65.511. Estimated marginal probabilities
for mean value of dependent variables: Probability = 1, 0.095; proba-
bility = 0, 0.905.
aThe dependent variable is equal to 1 if the individual bid zero dollars and
zero otherwise. All protest bids were eliminated.
The t-ratio is the ratio of the estimated parameter to the estimated standard
error. Given the assumptions of the estimates are maintained, the maximum
likelihood, logit parameter estimates are asymptotically normal. We have used
a t-distribution in judging the significance of these parameter estimates.
Significant at the 5-percent level.
4-26
-------
Additional insight into zero bidders issues can be obtained from a logit
analysis of valid zero bids (see Amemiya [1981]). To perform this analysis
for the Monongahela study, the dependent variable was set equal to 1 if a
nonprotest zero bid was given and equal to zero if a positive bid was given.
Consequently, protest bids were eliminated from the analysis. For consist-
ency, the explanatory variables used are the same as in the option price
regression (as discussed in Section 4.5). The binary variable to denote
Monongahela users and several interviewer dummies were eliminated due to a
lack of variation.
The results of the logit analysis of zero bidders are shown in Table 4-7.
This model requires a cautious interpretation of the estimated coefficients. In
the logit procedure, the expected change in the probability of bidding zero is
derived from the estimated equation where the probability of bidding zero
depends on the value of the independent variables.
The results were encouraging, with no evidence of interviewers signifi-
cantly affecting the odds of bidding zero. The performance of other variables
is consistent with previous results and a priori reasoning. Increases in age
significantly affected the likelihood of bidding zero. Each year's increase,
evaluated at the mean, is expected to change the probability of bidding zero
by 0.003. The results also indicate a relationship between zero bids and
questionnaire version. When the respondent was presented with the $25 bid-
ding game rather than the payment card, the probability of bidding zero de-
creased by 0.113. Also, the attitude toward cost was consistent, because
those respondents who stated a willingness to pay a portion of cleanup cost
had a lower probability of bidding zero.
The logit model was also used to explain why individuals protested the
option price question. As shown in Appendix C, the results are very weak,
with only the attitude toward cost variable significant and all other analysis
variables insignificant.
4.5 OPTION PRICE RESULTS
The central element in a contingent valuation study is the valuation
responses revealed in the hypothetical market situation. Much of the analysis
in the early contingent valuation experiments focused on the fitting of a bid
function to the willingness-to-pay bids. In this section, a linear approximation
is used in a regression analysis to fit the bid function. However, the basic
emphasis of the regression analysis is to organize the information presented
and not to estimate the bid function.*
*The willingness-to-pay data contain no negative bids which implies that
they are truncated at zero. This can lead to biased parameter estimates with
regression analysis, depending upon the distribution of bids. Since the sam-
ple excludes protest bidders, all responses should fall in the positive domain.
Negative responses would be inconsistent with the group being described by
the model. The difficulties posed by truncation could be handled in a variety
of ways including: transforming the dependent variable (i.e., using the log
4-27
-------
Specifically, this section summarizes the analytical basis of the option
price and user amounts, the statistical procedures employed to analyze these
estimates, the comparison of estimates between elicitation methods, and the
results on starting point and interviewer bias. In addition, it also compares
results with those from previous studies.
The amounts provided by the respondents represent their option prices
rather than user willingness to pay, as measured in many previous contingent
valuation studies. That is, the option price includes both the expected con-
sumer surplus that respondents anticipate from future use of the site's ser-
vices as well as a premium—the option value—that they are willing to pay to
obtain these site services should they decide to use them. The premium can
be attributed to uncertainty either in the respondents' future demand for the
site and/or uncertainty in the supply of the site's services at given water
quality levels. Chapter 5 explores these issues in more detail, but it is
important to understand this distinction to correctly interpret the results.
As discussed in Chapter 2, the option price amounts are based on the
Hicksian surplus measures, with the equivalent surplus measure used for the
loss of the recreation services of the Monongahela River (Level D to Level E)
and the compensating surplus measures used in measuring the option price for
the improvements to fishable and swimmable water. The use of these meas-
ures corresponds to the existing property rights for the overall level of
Monongahela recreation services, with the river currently supporting boating
activities. It is important to note that several sections of the Monongahela
have considerably higher water quality and are capable of supporting sport
fishing due to the Influence of tributaries. However, the boatable designation
is a reasonable description of the overall water quality level.
Determining the treatment of outlying responses is an important step in a
contingent valuation study. Randall, Hoehn, and Tolley [1981] suggest that,
once the outliers are determined and removed, the contingent valuation method
will provide a "core" of responses useful for analysis. In general, previous
efforts have used subjective judgment in making this determination, with little
or no discussion provided. For example, Rowe, d'Arge, and Brookshire
[1980] follow the procedure mentioned in Randall, Ives, and Eastman [1974] of
eliminating bids greater than 10 standard deviations from the mean. In neither
case is much discussion provided on the judgments made in selecting this pro-
cedure. While the role of* judgment will almost always loom large in these
decisions, it is difficult to evaluate and transfer the methods used to evaluate
the contingent valuation results unless a more systematic basis for the judg-
ment is detailed.
of the bids, if the zero bidders were dropped) and using an alternative esti-
mator. For the purposes of the present analysis, these models are intended
to be used only as a basis for judging the factors likely to influence bids and
not necessarily to estimate the magnitude of their impact. Past evidence on
the bias of ordinary least squares in presence of truncation effects indicates
that it did not greatly affect these judgmental evaluations of specific variables.
4-28
-------
Our approach relies on more formal use of statistical indexes of the influ-
ence of particular observations on a model's estimated parameters. Belsley,
Kuh, and Welsch [1980] suggest a number of statistical procedures that can
be used in prescreening data for outliers. The Monongahela study used a
procedure that follows their discussion to identify outlier candidates. The
Belsley-Kuh-Welsch statistic (DFBETA) measures the effect of each individual
observation on each of the estimated coefficients in a regression model. It is
estimated by Equation (4.1):
(XTX)1
DFBETA = b - b(i
i-n.
where
b = the estimated coefficient with all observations included
b(i) = the estimated coefficient with one less observation
h. = x. (XTX)~1x.T
6j = the ordinary least-squares residuals.
This statistic is not a formal statistical test. It is merely an index of the
extent of influence of particular observations. It implicitly assumes that option
prices can be related to economic characteristics. In this application, the sta-
tistics presented in the first column of Table 4-8 are expressed as percentage
changes in the income coefficient of the final regression model discussed later
in this chapter. The effect of income was selected because this variable is
the only variable we know, based on economic theory, that should influence
option price bids. Moreover, the relationship between option price and user
value can be expected to be influenced by the role of income in an individual's
indirect utility function. These changes represent approximations of elastici-
ties described in Belsley, Kuh, and Welsch [1980].
Rather than employ one of the arbitrary statistical criteria suggested in
Belsley, Kuh, and Welsch, the procedure was supplemented in this study with
a judgment that (±) 30 percent was the cutoff point for outliers. An element
of judgment is also required in selecting the regression model from which the
Belsley-Kuh-Welsch statistic is calculated. After comparing models, the judg-
ment was made to select the general model presented later in Table 4-11. How-
ever, in comparing the results between the models, the 16 outliers determined
by the same cutoff point for another regression model (see Appendix G) were
all included in the 32 outliers profiled in Table 4-8.
The results in Table 4-8 are striking in terms of the differences from
the Randall, Ives, and Eastman [1974] criteria. Many of the outliers are small
or zero bids that would have been retained in their procedure. In addition,
the consistency in the characterization of the outliers is informative. For the
respondents classified as outliers, 63 percent earned annual incomes of $2,500
4-29
-------
Table 4-8. Profile of Outliers
Belsley-
Kuh-Welsch
statistic
-2.1 3. 12
-155.99
-100.04
-79.83
-66.19
-63.25
-62.95
-56.70
-54.98
-49.68
-44.62
-43.80
-43.16
-37.34
-36.46
-36.03
-31.40
-30.43
31.24
33.98
35.39
37.77
41.78
47.15
52.23
52.86
58.18
65.70
69.15
79.58
82.52
112.04
Option
loss of
Version
$125 bidding game
$125 bidding game
direct question
$125 bidding game
$125 bidding game
$25 bidding game
payment card
$25 bidding game
direct question
payment card
$125 bidding game
$25 bidding game
$125 bidding game
$25 bidding game
$25 bidding game
$25 bidding game
direct question
$125 bidding game
direct question
$125 bidding game
$125 bidding game
payment card
payment card
$125 bidding game
$125 bidding game
payment card
$125 bidding game
$125 bidding game
direct question
$125 bidding game
payment card
payment card
price: avoid
site (D to E)
($/yr)
$125
$125
$200
500
$125
25
450
60
0
50
155
5
155
5
25
0
200
200
5
0
0
75
25
5
0
0
0
0
10
55
0
0
Option price:
improve water
quality to swimmable
($/yr)
$260
200
200
500
220
5
200
85
10
250
250
5
250
5
0
0
300
285
3
0
0
10
10
130
30
0
0
10
20
0
0
25
I ncome
$/yr
2,500
2,500
7,500
22,500
7,500
2,500
17,500
2,500
2,500
7,500
12,500
2,500
12,500
2,500
2,500
2,500
27,500
22,500
7,500
12,500
2,500
2,500
2,500
2,500
7,500
2,500
2,500
2,500
2,500
2,500
2,500
2,500
Age
(yr)
25
20
67
39
43
70
37
23
82
40
57
69
44
62
46
76
21
66
34
38
78
59
72
61
50
43
79
66
33
71
53
26
Sex
Male
Female
Male
Male
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Male
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Education
(yr)
12
12
12
14
10
10
12
12
10
14
12
10
10
10
10
16
12
12
12
12
0
12
12
12
12
10
10
12
12
10
12
12
User of
Monongahela Boat
site ownership
No
Yes
No
No
Yes
No
Yes
No
No
Yes
No
No
No
No
No
No
Yes
Yes
No
No
No
Yes
No
Yes
Yes
No
No
No
Yes
No
No
Yes
No
No
No
Yes
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
Yes
-------
a year or less, and 78 percent earned less than $7,500 a year. Female
respondents comprised 80 percent of the outliers, while only 4 respondents
had more than a high school degree. The last element of interest is that 14
of the 32 outliers had received the $125 starting point bidding game—twice as
many as the next version (the payment card). This last feature confounds
the interpretation of starting point bias presented later in this and the follow-
ing chapter.
In summary, the Belsley-Kuh-Welsch [1980] procedure is a systematic
approach for identifying outlying bids within contingent valuation studies. It
does not replace the need for judgment but gives a basis for making the judg-
ments .
The results presented in this chapter are all based on two edits of the
301 completed survey questionnaires. The first edit removed the protest bids
from the calculation of means and the regressions. Protest zeros were
respondents who bid zero for reasons other than "that is all they could afford"
or "that is what it was worth." This removal is consistent with practices of
Randall, Ives, and Eastman [1974] and Rowe, d'Arge, and Brookshire [1980].
The second edit removed the outliers following the Belsley-Kuh-Welsch
[1980] procedure. Appendix C presents the estimated means for both the full
sample and the sample with only the protest bids excluded. Calculated
t-statistics revealed no statistically significant differences between the means
estimated from the full sample and those estimated with the protest bids
excluded. The effects of omitting the outlier observations are discussed at
the appropriate points in this and in the following chapter.
The salient questions to be answered from the survey results center on
the comparison of the alternative methods used to elicit the option price
amounts, while the plausibility of the results is substantiated by testing for
potential biases in the responses. Table 4-9 presents the estimated means
grouped by questionnaire version, with distinctions made between users and
nonusers. The mean values are provided for the loss of the recreation ser-
vices of the site (avoiding a decrease from Level D to Level E on the water
quality ladder in Figure 4-5), for an improvement in water quality from boat-
able to fishable (Level D to Level C), and for an improvement in water quality
from fishable to swimmable (Level C to Level B). Combined option prices are
presented for the improvements in the level of water quality and for the im-
provements plus the loss of the services of the site.
One inference that can be drawn from Table 4-9 is that the option prices
are sizable for the Monongahela River but are of the same order of magnitude
regardless of the method used to elicit the amount. Option price amounts com-
bined for all levels range from a mean of $54 per year for the bidding game
with a $25 starting bid to $118 for the bidding game with a $125 starting bid.
Mean bids for the combined amounts for the payment card and direct question
equaled $94 and $56, respectively. The range of mean option price amounts
is even narrower when only the bids for improvements are considered, varying
from $25 per year to $60 per year for the two bidding games.
4-31
-------
Table 4-9. Estimated Option Price for Changes in Water Quality:
Effects of Instrument and Type of Respondent--
Protest Bids and Outliers Excluded
User
Nonuser
Combined
Change in
water quality
n
X
n
n
1. Iterative bidding framework—starting point = $25 (Version C)
D to E avoid 27.4 16.7 19 29.7 35.7 39 29.0
D to C 18.9 16.3 19 14.5 15.2 39 15.9
C to B 11.8 14.5 19 7.2 11.6 39 8.7
D to Ba 32.1 27.1 19 21.7 24.0 39 25.1
Combined: all levels 59.5 38.1 19 51.4 53.1 39 54.1
2. Iterative bidding framework—starting point = $125 (Version D)
D to E (avoid) 94.7 66.0 16 38.8 51.3
D to C 58.1 51.9 16 26.3 45.4
C to B 33.1 48.4 16 11.6 33.1
D to B 99.7 87.9 16 40.5 69.0
Combined: all levels 194.4 136.5 16 79.2 102.5
3. Direct question framework (Version B)
D to E (avoid) 45.3 65.2 17 14.2 27.1
D to C 31.3 44.2 17 10.8 21.6
C to B 20.2 35.5 17 8.5 21.9
D to B 52.9 72.5 17 20.3 41.4
Combined: all levels 98.2 103.5 17 34.5 66.4
32
32
32
32
32
34
34
34
34
34
57.4
36.9
18.8
60.2
117.6
24.5
17.6
12.4
31.2
55.7
30.6
15.5
12.7
25.3
48.5
62.0
49.5
39.7
80.0
126.0
45.4
32.1
27.4
55.2
85.2
4. Direct question framework: payment card (Version A)
58
58
58
58
58
48
48
48
48
48
51
51
51
51
51
D to E (avoid)
D to C
C to B
D to B
Combined: all levels
46.
45.
22.
71.
117.
8
3
9
2
9
42.2
71.4
48.7
117.7
117.0
17
17
17
17
17
53.0
21.9
7.7
29.9
82.8
76.3
33.8
20.0
47.5
104.7
37
37
37
37
37
51.
29.
12.
42.
93.
0
3
5
9
9
67.1
49.3
32.2
78.1
108.9
54
54
54
54
54
D to B are the combined amounts for improvements only.
The results of the test for differences in means between methods for both
users and nonusers are shown in Table 4-10. These results show that the
differences do arise between the means in the bidding games, suggesting there
may be a bias attributable to the difference in the starting points. The com-
bined and user means are statistically different at the 5-percent level of sig-
nificance for users and for the combined groups. However, the evidence is
not completely conclusive because the differences in nonuser means are not
significant. In addition, the regression results shown in Table 4-11 do not
conclusively show a starting point bias problem. The regression model esti-
mated without the outliers shows no statistically significant difference between
the iterative bidding games. If the outliers are not removed, the model sug-
gests starting point bias, as indicated in Appendix C. Thus, in the regression
4-32
-------
Table 4-10. Student t-Test Results for Option Price--
Protests Bids and Outliers Excluded
.a
Means combined
User
Nonuser Combined
Payment card vs.
D to E
E to B
direct question
2.806
2.300
2.353
1.991
Payment card vs. $25 iterative bidding
D to E
D to C
E to B 2.061
Payment card vs. $125 iterative bidding
D to E -2.499
Direct question vs. $25 iterative bidding
D to E
Direct question vs. $125 iterative bidding
D to E -2.161
D to C
E to B -2.289
D to B
$25 iterative bidding vs. $125 iterative bidding
2.
1.
263
954
2.530
-2.074
-2.453
-2.117
-3.020
-2.308
-2.8786
-2.109
D to E
D to C
E to B
D to B
-4.294
-3.119
-4.131
-3.183
-3.072
-3.046
-3.539
-3.159
Only cases with statistically significant differences in the means at the 0.05
significance level are reported.
analysis, differences attributable to starting point cannot be distinguished from
the influence of the outlier observations.
Some additional insights into differences in the elicitation method can be
developed from the results in Tables 4-10 and 4-11. The mean option-price
for users of the Monongahela is significantly higher when the bidding game
with the $125 starting point is used to elicit option price compared to either
direct question technique. The differences are present for the aggregate op-
tion price and for the loss of site services, but no differences are detected
for the incremental improvements to fishable and swimmable water quality
levels.
The regression results from Table 4-11 are generally consistent with the
means tests. Using the dummy variable technique to compare the payment
card with the other three versions shows option price is significantly higher
for the payment card than for the direct question and the $25 bidding game,
while no differences exist between the payment card results and those for the
4-33
-------
Table 4-11. Regression Results for Option Price Estimates--
Protest Bids and Outliers Excluded
Independent variables
Intercept
Sex (1 if male)
Age
Education
1 ncome
Direct question
Iterative bidding game ($25)
Iterative bidding game ($125)
User (1 if user)
Willing to pay cost of
water pollution
(1 if very much or somewhat)
Interviewer #1
Interviewer #2
Interviewer #3
Interviewer #4
Interviewer #5
Interviewer #6
Interviewer #7
Interviewer #8
Interviewer #9
R*
F
Degrees of freedom
D to E (avoid)
-34.512
(-0.973)
8.451
(0.916)
-0.292
(-1.094)
5.294
(2.071)°
0.0006
(1.652)
-32.311 .
(-2.771)"
-20.623
(1.852)
1.7522
(1.421)
8.840
(0.919)
17.001
(1.788)
14.211
(0.750)
1.723
(0.099)
-22.833
(-1.344)
-28.125
(-0.860)
6.932
(0.404)
47.012
(0.887)
27.670
(1.425)
14.022
(0.801)
17.874
(0.454)
0.334
3.78
136
Water
D to C
-29.307
(-1.098)
-0.672
(-0.097)
0.290
(-1.440)
2.901
(1.508)
0.0003
(1.151)
-14.372
(-1.638)
-12.572
(-1.500)
6.639
(0.716)
8.083
(1.117)
21.960 .
(3.06B)D
7.090
(0.497)
12.242
(0.938)
21.141
(1.653)
3.050
(0.124)
4.996
(0.387)
95.513 .
(2.394)D
2.470
(0.169)
29.961 .
(2.274)b
39.586
(1.336)
0.284
3.00
136
quality changes
C to B
-5.430
(-0.257)
-1.657
(-0.302)
-0.265
(1.668)
-5.27
(0.347)
0.0003
(1.260)
-3.500
(0.505)
-5.657
(-.854)
0.739
(0.101)
6.839.
(1.96)D
10.023
(1.772)
11.334
(1.006)
16.849
(1.634)
17.578
(1.740)
20.605
(1.059)
2.191
(0.215)
66.288 .
(2.102)°
4.130
(0.357)
19.871
(1.908)
-7.935
(-0.339)
0.166
1.50
136
Total, all levels
-56.653
(-0.916)
6.484
(0.403)
-0.854
(-1.834)
8.066
(1.810)
0.0012
(1.832)
-50.734 b
(-2.495)°
-39.566
(-2.037)°
31.089
(1.446)
26.026
(1.552)
51.326
(3.095)°
26.509
(0.802)
24.719
(0.817)
9.292
(0.314)
-12.334
(-0.216)
11.435
(0.382)
198.450
(2.146)°
39.645
(1.170)
58.063
(1.902)
37.330
(0.544)
0.366
4.36
136
Total improve-
ments only
-22.141
(-.517)
1.967
(-0.177)
-0.562
(1.743)
2.773
(0.899)
0.0006
(1.278)
-18.423
(-1.309)
-18.943
(1.409)
13.568
(0.912)
17.187
(1.481)
34.326
(2.990)°
12.298
(0.538)
22.996
(1.099)
32.125
(1.567)
15.791
(0.400)
4.503
(0.217)
151.439 .
(2.366)°
11.975
(0.511)
44.041.
(2.08)°
19.456
(0.409)
0.269
0.278
136
aNumbers in parentheses are.asymptotic t-ratios for the null hypothesis of no association.
"Significant at the 0.05 level.
bidding game with the $125 starting point. The differences are significant
only for the loss of site services and for the combined option price. When
other influences are held constant in the regression analysis, respondents who
received the payment card expressed aggregate option prices approximately
$40 to $50 higher than those expressed by respondents in the $25 starting
point bidding game and the direct question. It is possible to conclude that
there are significant differences between methods but that all methods estimate
option price at the same order of magnitude. The differences cannot be
detected among the bids for improvements in water quality levels, possibly
because the effects of the methods are limited to the initial amounts given.
This may minimize the effect of the question format when incremental amounts
are elicited. This conclusion should be viewed with some caution since the
differences between methods could be difficult to detect simply because the
4-34
-------
number of bids for the improvements is too small to offset the variation in the
amounts expressed. The consistency in the results from the various tests,
however, is particularly encouraging as a plausibility check against the influ-
ence of hypothetical bias in the contingent valuation design.
An examination of the regression results for option price combined over
all water quality levels reinforces the plausibility of the results. The coeffi-
cients of the socioeconomic variables all have the expected signs, and the co-
efficients for age, education level, and income are significant at either the
0.05 level or very close to it. The results indicate a strong role for respond-
ent attitude toward paying the cost of water pollution. Persons who identified
themselves as either very much or somewhat willing to pay for water pollution
control were willing to spend $50 more per year than persons who were not
willing to pay the cost, with all other things held constant. This consistency
of attitudes, combined with the performance of the socioeconomic variables and
the ability of the model to explain almost 37 percent of the variation in option
price, builds a strong case against the influence of hypothetical bias in the
contingent valuation design.
The regression results in Table 4-11 also shed some light on the question
of a bias in the willingness to pay that could be attributable to differences in
interviewers. Using the dummy variable technique, the results indicate that
the influence of interviewer bias is limited. Only for two interviewers are the
coefficients statistically significant at the 0.05 level for some levels of water
quality. One of the cases involved an interviewer who conducted only two
interviews before being removed from the interviewing team. This interviewer
did not take part in the training session and also conducted interviews only
in the Latrobe area, which is a considerable distance from the Monongahela
River. The second interviewer also conducted interviews in the Latrobe area
and in one area very close to the river. These cases may simply reflect the
model's inability to differentiate between an interviewer effect and some omitted
variables. Thus, the effect of the interviewer is quite small and reinforces
the importance of the training sessions that were conducted in Pittsburgh prior
to the survey.*
Table 4-12 presents the results of student t-tests for differences in means
between users of the Monongahela River and nonusers broken down by the
technique used to elicit option price. The results show that users who re-
ceived either the direct question or the $125 starting point bidding game ex-
pressed bids that were higher than those of nonusers. There were no statis-
tically significant differences in means for either the payment card or the $25
starting point. This suggests that users have somewhat higher option prices,
*To conclusively design a test for interviewer bias would require that
interviewers be randomly assigned to different areas in the survey. The prac-
tical issue is that this could have a significant impact on data collection costs
because of interviewers having to cover a substantial part of the survey
area. In the Monongahela survey, interviewers were assigned areas based on
the lowest travel costs to obtain the interview.
4-35
-------
Table 4-12. Student t-Test Results for Option Price-
Protest Bids and Outliers Excluded
Means compared
User vs. nonuser Means compared User vs. nonuser
Payment
card (A)
D to E
D to C
C to B
D to B
E to B
Direct
question (B)
D to E
D to C
C to B
D to B
E to B
-0.313
1.645
1.322
1.103
1.847
2.414a
2.234a
1.454
2.669a
2.049a
$25 iterative
bidding (C)
D to E
D to C
C to B
D to B
E to B
$125 iterative
bidding (D)
D to E
D to C
C to B
D to B
E to B
-0.275
1.026
1.322
0.591
1.488
3.2313
2.186a
1.819
3.279a
2.555a
'Denotes significance at the 0.05 level.
but this difference is not pervasive. Thus, a survey of only the users of
Monongahela River would have substantially underestimated the recreation and
related benefits of water quality improvements. The full extent of these in-
trinsic benefits is developed in the following chapter.
4.6 USER VALUE RESULTS
Table 4-13 shows estimated user values, which resulted from respondents
referring to the value card (see Figure 4-6) and breaking out the user value
component of the option price. These values are comparable to those estimated
in most of the previous contingent valuation efforts and are compared with
the benefits estimated with the travel cost method in Chapter 8.
User value means are presented for users only and the means calculated
for all respondents. Tests to determine whether the user values are statis-
tically different from zero, shown in Appendix C, indicated that the user
values for the D to E levels and combined over all levels are statistically dif-
ferent from zero at the 0.05 level of significance. The user values for im-
provements in water quality are only different from zero for the $25 bidding
game and not for any other methods. Additional tests for differences in user
values between methods, also contained in Appendix C, showed that means
from the $25 bidding games were statistically different (lower) than those esti-
mated with the $125 bidding game, but only for Levels D to E and the user
values for all combined water quality levels. The differences for the user
4-36
-------
Table 4-13. Estimated User Values--Protest Bids
and Outliers Excluded
User only Combined
s n X s n
Iterative bidding framework
$25 starting point (C)
D to E 6.59 12.59 19 2.16 7.73 58
D to C 4.21 7.68 19 1.38 4.76 58
C to B 5.00 7.99 19 1.64 5.08 58
D to B 10.53 14.43 19 3.45 9.52 58
Combined: all levels 17.11 25.13 19 5.60 16.28 58
Iterative bidding framework
$125 starting point (D)
D to E 36.25 58.98 16 12.08 37.52 48
D to C 20.31 42.67 16 6.77 25.98 48
C to B 20.00 42.82 16 6.66 25.99 48
D to B 48.75 87.87 16 16.25 54.81 48
Combined: all levels 138.11 85.00 16 28.33 87.90 48
Direct question (B)
D to E 19.71 37.85 17 6.57 23.38 51
D to C 21.18 42.22 17 7.06 25.93 51
C to B 10.00 29.10 17 3.33 17.14 51
D to B 31.18 64.63 17 10.39 39.46 51
Combined: all levels 50.88 77.46 17 16.96 50.07 51
Direct question
D to E
D to C
C to B
D to B
Combined: all
: payment card (A)
levels
19.
30.
19.
51.
70.
71
88
71
18
88
34.
74.
49.
122.
127.
30
57
42
88
61
17
17
17
17
17
6.20
9.72
6.20
16.11
22.31
20
43
28
71
77
.99
.45
.68
.65
.59
54
54
54
54
54
values combined for all respondents were the same as those for users, except
for the comparison of bidding games, where the difference was significant only
for the Level D to E change.
Table 4-14 presents the results for the regression models with the user
values as the dependent variables. The models generally have less explana-
tory power than the option price models but do show some limited ability to
explain variations in user value. Age and respondent attitude toward paying
the cost of water pollution are the key variables in the model, and both have
the expected signs.
4-37
-------
Table 4-14.
Regression Results for User Value Estimates of Water Quality
Changes—Protest Bids and Outliers Excluded
Water quality changes
Independent variable
Intercept
Sex (1 if male)
(1 if male)
Age
Education
Income
Direct question
Iterative bidding game ($25)
Iterative bidding game ($125)
Willing to pay cost of
water pollution
(1 if very much or somewhat)
Interviewer #1
Interviewer #2
Interviewer #3
Interviewer #4
Interviewer #5
Interviewer #6
Interviewer #7
Interviewer #8
Interviewer #9
R2
F
Degrees of freedom
D to E (avoid)
10.372
(0.551)
1.070
(0.218)
-0.236
(-1.761)
0.193
(0.142)
0.00001
(0.073)
-2.842
(-0.456)
-4.769
(-0.803)
6.665
(1.014)
9.931 .
(1.988)D
-1.585
(-0.157)
4.626
(0.500)
-3.479
(-0.395)
-9.651
(-0.553)
-5.724
(-0.624)
-6.266
(-0.221)
12.634
(1.225)
-5.509
(-0.589)
-18.707
(-0.889)
0.13
1.22
137
D to C
1.529
(0.070)
-1.625
(-0.285)
-0.264
(-1.690)
0.156
(0.098)
0.0002
(0.740)
-5.766
(-0.796)
-10.724
(-1.554)
-8.540
(-1.119)
10.828
(1.866)
4.020
(0.343)
13.666
(1.270)
27.836 .
(2.721)°
7.079
(0.349)
1.410
(0.132)
19.835
(0.602)
4.664
(0.389)
11.417
(1.050)
-3.159
(-0.129)
0.14
1.34
137
C to B
-2.143
(-0.138)
-0.107
(-0.026)
-0.201
(-1.817)
0.464
(0.412)
0.00003
(0.167)
-4.300
(-0.836)
-5.072
(-1.035)
-3.006
(-0.554)
8.116 b
(1.969)D
3.029
(0.364)
11.118
(1.455)
19.108 .
(2.630)D
2.996
(0.208)
-0.087
(-0.012)
11.477
(0.491)
1.177
(0.138)
3.960
(0.513)
-3.381
(-0.195)
0.14
1.28
137
Total com-
bined all levels
6.686
(0.180)
0.121
(0.013)
-0.507
(-1.918)
-0.063
(-0.023)
0.0002
(0.607)
-11.536
(-0.940)
-15.588
(-1.333)
-7.103
(-0.549)
19.654
(1.997)D
8.758
(0.441)
25.736
(1.411)
47.530
(2.740)
9.987
(0.290)
3.474
(0.192)
27.795
(0.498)
16.328
(0.803)
15.851
(0.860)
-8.995
(-0.217)
0.14
1.34
137
Total improve-
ments only
17.058
(0.363)
1.191
(0.097)
-0.743
(-2.220)b
0.130
(0.038)
0.0003
(0.508)
-14.378
(-0.925)
-20.358
(-1.374)
-0.438
(-0.027)
29.586
(2.374)D
7.172
(0.285)
30.362
(1.314)
44.051 .
(2.005)D
0.336
(0.008)
-2.250
(-0.098)
21.529
(0.305)
28.962
(1.125)
10.342
(0.528)
-27.702
(-0.528)
0.15
1.44
137
Numbers in parentheses are asymptotic t-ratios for the null hypothesis of no association.
Significant at the 0.05 level.
4.7 SUMMARY
The contingent valuation estimates of the option price for quality
improvements are consistently plausible throughout the various analytical
considerations. The empirical results indicate that the methods used to elicit
the bid do have a statistically significant effect on the estimates of an individ-
ual's valuation. Payment cards and the bidding game with a $125 starting
point produced higher willingness-to-pay estimates than either the direct ques-
tion or the bidding game with a $25 starting point. There is some evidence
4-38
-------
of a starting point bias in the bidding game, but the statistical analyses are
not conclusive. The results comparing bidding games with nonbidding games
indicated no differences when these combined comparisons are made. In terms
of future contingent valuation experiments, the results imply that using bid-
ding games to elicit willingness to pay requires a range of starting points to
test for starting point bias. No statistical or analytical differences are appar-
ent when nonbidding games are employed to elicit willingness to pay.
For the continued use of the contingent valuation method to estimate
benefits of water quality improvements, the general prognosis from the results
of the Monongahela River case study is a good one. The empirical models per-
formed reasonably well in explaining variations in willingness to pay, with little
indication that individual interviewers influenced the results. The consistently
plausible signs and magnitudes of key economic variables suggest that the
respondents perceived the realism of the survey and did not experience prob-
lems with the hypothetical nature. Moreover, the results came from a random
sample of households from an area whose socioeconomic profile is not ideally
suited for a contingent valuation survey: The respondents were older, less
educated, and poorer than in previous contingent valuation studies.
4-39
-------
CHAPTER 5
CONTINGENT VALUATION DESIGN AND RESULTS:
OPTION AND EXISTENCE VALUES
5.1 INTRODUCTION
Over a decade ago, Krutilla [1967] emphasized the importance of nonuser
benefits to the process of efficiently allocating natural environments. In his
development of the special problems associated with valuing the services of
natural environments, Krutilla identified several types of nonuser values. The
objective of this chapter is to present survey results that attempt to measure
directly two of the sources of benefits Krutilla identified—option value and
existence value. It should be acknowledged at the outset that the first of
these, option value, has received the greatest attention in the literature and
is regarded as one of the most important components of nonuser values. As a
consequence, the majority of this chapter is devoted to the theoretical and
empirical problems associated with modeling and measuring option value.
The simplest approach to defining option value is to use an example.
Consider an individual who is uncertain whether he will visit a recreation
site on the Monongahela River in the future. Also, suppose this person is
uncertain whether the facility will be available in the future should he decide
to use it. This uncertainty over availability may arise because the individual
either does not know whether the facility will permit any use or does not know
the types of uses that will be permitted. (For example, a river may not per-
mit any use, or it simply may not be available for swimming. Of course, the
inability to support recreational swimming does not preclude the provision of
sport fishing and boating services.) What is at issue is uncertainty over the
character of the supply. This uncertainty can involve the all-or-none case, a
concept conventionally used in the theoretical literature, or simply a change
in the types of uses that can be supported in the future. Given these condi-
tions, a rational individual may be willing to pay some amount for the right to
use the facility's services in the future. This payment can be interpreted as
a means of insuring access to the site's services. Of course, it does not elim-
inate the individual's uncertainty over whether he will actually decide to use
the site's services.
In all discussions of option value, the payment is assumed to be constant
regardless of whether or not the individual visits the site. The payment is
usually described as the option price. The option value is defined as the dif-
ference between this payment and the individual's expected consumer surplus
from having access to the site's services. In the extreme case, where the
choice is use or no use, the expected consumer surplus is the weighted sum
5-1
-------
(by the relevant probabilities) of the consumer surplus associated with access
and use of the site plus that of access and no use. Of course, it must be
recognized that this discussion assumes that markets do not exist for contin-
gent claims that could handle the prospects for a future demand. Thus, there
is no alternative mechanism (other than purchasing the option) available to
the individual for diversifying the risk he experiences.
Researchers have generally agreed that this description of behavior is
plausible. The literature, however, includes a wide array of arguments con-
cerning the relationship between the maximum willingness to pay for the op-
tion and the exoected consumer surplus. For example, Cicchetti and Freeman
[1971] observed that option value existed as a direct result of risk-averse be-
havior and was therefore positive. By contrast, using a similar framework,
Schmalensee [1972] concluded that option value may be positive or negative
depending on the vantage point selected for evaluating the individual's choices.
Subsequent contributions questioned Schmalensee's definition of risk aversion
(Bohm [1975]); introduced time specifically into the analysis (Arrow and Fisher
[1974]; Henry [1974]; and Conrad [1980]); and, more specifically, considered
the mechanisms available to the individual for diversifying risk (Graham
[1981]). The result has been a large and often confusing literature.
Understanding the past contributions in this area requires a clear de-
scription of three aspects of the role of uncertainty in each model. This
characterization of uncertainty is most easily summarized by posing three ques-
tions:
What is the source of the uncertainty in the individual's deci-
sion problem?
How will the uncertainty in this decision problem ultimately be
resolved?
Is it possible to amend the decision process to accommodate new
information that may resolve some of the uncertainty?
Each of the past analyses of option value provides implicit answers to these
questions. Moreover, the answers help explain why these analyses yield such
diverse conclusions.
Two recent papers have provided the elements necessary to integrate a
significant portion of the literature. The first of these is a review article by
Bishop [1982] that provides an excellent summary of past contributions and
extends earlier work by amending Schmalensee's framework to delete the indi-
vidual's demand uncertainty and to explicitly include supply uncertainty. In
the second paper, Graham [1981] seeks to define the appropriate measure of
benefits for benefit-cost analyses in the presence of uncertainty. He con-
cludes, as Bohm [1975] did earlier, that option price and not expected con-
sumer surplus is the appropriate valuation measure. Unfortunately, his evalu-
ation of the problem tends to focus on cases where individuals face specific
risks and have access to ideal markets in which to diversify these risks. For
these cases, he quite correctly concludes option value is largely irrelevant.
5-2
-------
Of course, most resource and environmental problems do not "fit" these
assumptions. Nonetheless, his framework and evaluation of the case of collec-
tive risk provide another important insight into the appropriate treatment of
option value.
Section 5.2 of this chapter reviews the modeling of uncertainty and,
specifically, the use of a contingent claims framework. This review is neces-
sary to understand the implications of alternative definitions of risk aversion.
With this background it is possible in Section 5.3 to describe the "timeless"
analyses of option value and to relate them to the recent contributions of
Bishop [1982] and Graham [1981]. Section 5.4 briefly discusses the relation-
ship between option value and quasi-option value introduced by Arrow and
Fisher [1974].
Section 5.5 discusses three recent attempts to empirically estimate non-
user values—the Greenley, Walsh, and Young [1981] estimates of option value
from potential water quality degradation in the South Platte River basin in
Colorado; Mitchell and Carson's [1981] estimates of the total "intrinsic" values
for improvements in national water quality; and the Schulze et al. [1981] anal-
ysis of visibility benefits for national parks in the Southwest.
Sections 5.6 through 5.8 describe the survey results for the Mononga-
hela River basin. Section 5.6 describes the questions used to estimate option
value and to determine its sensitivity to the character of the supply uncer-
tainty. The survey has been structured so that it is possible to distinguish
the estimates according to the question used, the level of supply uncertainty,
and the character of the respondents. Respondents are grouped according to
whether they have used the river for recreation purposes. Section 5.7 pre-
sents a summary of the empirical results and an evaluation of the effects of
the questioning mode (as well as of the starting point for the iterative bid-
ding scheme) used for the estimates. In addition to measuring option value,
attempts were made to measure existence values independently. Section 5.8
discusses these efforts. Section 5.9 presents a summary of the primary find-
ings of this research.
5.2 CONTINGENT CLAIMS MARKETS AND THE MODELING OF
UNCERTAINTY*
The traditional approach to dealing with production and exchange deci-
sions under uncertainty involves a definition of new commodities that specifies
not only their physical characteristics, location, and date of availability, but
also a particular state of the world that must be realized if the stipulated
transaction is to take place. In terms of the example used in Section 5.1,
one state of the world permits access to the Monongahela River recreation site
and another does not. In this framework, uncertainty has the effect of
expanding the commodity set available to the individual. For example, if, in
The theoretical analysis in this chapter is an expanded version of. that
reported in Smith [1983].
5-3
-------
the absence of uncertainty, there are N commodities, and if uncertainty intro-
duces K states of nature, a contingent claims model redefines the commodity
set to be N-K contingent claims. Each is a claim to a good contingent upon
the state of nature. In this framework, the model is describing how an indi-
vidual's plans for activities are made rather than the actual activities them-
selves. These plans involve the selection of claims to goods, should the state
of the world be realized. Thus, the individual must allocate his budget opti-
mally among these claims before the state of the world is known.
Of course, defining optimality in this framework requires consideration of
the rule that aggregates these claims. Because each of these new commodities
involves both a good and a state of world, each outcome needs an associated
probability. This permits the use of expected utility—justified in the early
work of von Neumann and Morgenstern [1947]—as the rule for aggregating
the values associated with these claims. That is, given the four postulates of
rational choice, the utility of any set of contingent claims (e.g., a commodity
considered over all states of nature) can be derived as the expected utility.*
The most important of these postulates for understanding the literature on op-
tion value is the uniqueness postulate, which requires the expected utility of
a set of claims to be independent of the "state labeling" of the commodities
involved in these claims. That is, these commodities could be rearranged over
all states of nature without changing the expected utility as long as each com-
modity is realized with the same probability.
Most analyses of option value drop this postulate by assuming that the
individual has a different utility function depending on whether the services
of a recreation site are demanded or not demanded. The presence of a posi-
tive level of demand for the site is not simply a reflection of a higher income
or a lower price. With a given income and prices of substitute goods, conven-
tional statements of an individual's demand function often assume that there is
a price at which the services of a site will not be demanded. With a state-
dependent demand specification it is unlikely that the reasons why the site
will not be demanded can be fully explained. Rather, this specification is
used simply to reflect a different set of preferences that depend on the exist-
ence of demand for the site. To emphasize this assumption, the following
review summarizes the difference between the consumer's allocation decisions
(among contingent claims) and the definition of risk aversion under the two
frameworks—one that maintains the uniqueness postulate and one that does not.
*The four postulates are: (1) ordering and preference direction—larger
incomes are preferred to smaller incomes; (2) certainty equivalence—there is
an amount, the certainty equivalent, that is intermediate in size to the largest
and smallest consequences of a given prospect; (3) independence—a cjaim,
designated as Z, can be substituted for its preference equivalent, say Z, in
any prospect into which Z enters and vice versa; and (4) uniqueness—the
certainty equivalent of a prospect depends only on the magnitudes of the prob-
abilities and incomes, not on their state designations. See Hirshleifer {1970,
pp. 219-20] or Malinvaud [1972, pp. 285-90] for further discussion. Cook
and Graham [1977] provide additional perspective for irreplaceable goods.
5-4
-------
Consider the case of two contingent commodities (or claims), Xt and X2,
corresponding to States 1 and 2 and having probabilities of p and (1-p)/ re-
spectively. If the prices of these claims are rt and r2, and if utility is de-
pendent on the amount of X., such as u(X.), the individual's objective func-
tion, when the uniqueness postulate is satisfied, can be written as Equation
(5.1):
V = piXXj + (1-p)M(X2) , (5-1)
where V is the expected utility. If the initial endowment of claims is
X2), the budget constraint limiting the individual's choices would be:
+ r2X2 = rjXj + r2X2 . (5.2)
Maximizing Equation (5.1) subject to Equation (5.2) and solving the first-
order conditions yields the familiar equality of relative prices and probability-
weighted marginal utilities, as in Equation (5.3)*:
pu'(X,) _ r^ (5.3)
(1-p)u'(X2) ~ r2 '
This result is usually specialized further by consideration of a "fair"
gamble case (i.e., where p dXj + (1 - p)dX2 = 0). This case implies the
equality of the probability ratio and the price ratio for the two contingent
claims (i.e., p/(1 - p) = rt/r2).t Using this condition, Equation (5.3) can
be rewritten as:
u'(X2) • '
The optimal allocation calls for equal claims in Xt and X2, as given by the
point R in Figure 5-1. Thus, the selection in this case will fall along the cer-
tainty locus (both income and utility) — the 45° line in Figure 5-1.
The traditional definition of risk aversion for this framework maintains
that risk-averse individuals require better than "fair" gambles before they
will select these alternatives over a certain claim with the same expected in-
come. Under the assumption of uniqueness there are two further implications
The second-order conditions are d2X2/dXt2 > 0. This can be shown,
given uniqueness, to be implied by the assumption of concavity of u(.). That
is: d2X2/dX!2 = 8/aXi (dX2/dXx) + 8/3X2 (dX./dXt) [dXz/dXj, where
dX2/dX± = -[p/(1-p)]-[u'(Xl) / u'(X2)] hence d^Xj, / dXj2 = p u"(X±) /
(1-p) u'(X2) - p2(u'(X1))2u"(X2) / (1-p)2(u'(X2))3. Concavity of u(.) implies
that u"(.) < 0, and thus dX22/dXj2 is positive, because p, (1-p), u1^), and
u'(X2) are all positive.
fThis conclusion is derived by recognizing the implications of a constant
initial budget and the "fair" gamble for selections of contingent claims: A
constant budget implies r1dX1 + r2dX2 = 0; a fair gamble implies pdXt +
(1-p)dX2 = 0; thus, a fair gamble implies -dX2/dXi = p/(1-p) = rt/r2.
5-5
-------
X2
Income Certainty Locui
Utility Certainty Locus
(a) "fi/ij -p/1-p
Figure 5-1. Optimal allocation of choice with contingent claims.
associated with risk-averse behavior. They are important because they pro-
vide the means for explaining the divergence between Schmalensee [1972] and
Bohm [1975] in their respective interpretations of the appropriate definition of
risk aversion. To understand these divergent interpretations, imagine a risk-
averse individual subject to the choice of X with certainty versus the_ prospect
of XJL with probability p and X2 with probability (1 - p). Assume X = pXj +
(1 -p)X2. Then a risk-averse individual's choice would be consistent with a
utility function that ranks these prospects as follows:
u(X) > pu(Xj)
- p)u(X2)
(5.5)
Equation (5.5) will be realized if u(.) is concave. Thus, the concavity of
u(.) is usually taken to imply risk aversion. In this study's analysis of "fair"
gambles, as given in Equation (5.4), the risk-averse individual's choices can
also be characterized as implying an allocation of resources among claims such
as u'(Xj) = u'(X2). All individuals will allocate their resources among claims
to States 1 and 2 so that these marginal utilities are equalized in the case of
"fair" prices. Since risk aversion is defined by the concavity of u(.), the
behavioral responses of a risk-averse individual will be determined by how he
responds to a change in p. However, once the assumption of uniqueness is
relaxed and state-specific utility functions are permitted, the condition for
fair gambles implies only that the marginal utilities will be equalized and not
that either the total utilities or the total monetary claims in each state will be
equalized. Without uniqueness there will be a distinction between the locus of
equal consumption (or income) over states (i.e., the 45° line defined as the
income and utility "certainty" locus under the assumption of uniqueness) and
the utility certainty locus, where ut(Xj) = u2(X2), as illustrated in Figure 5-2.
Moreover, the optimal allocation will not necessarily lie on the utility certainty
locus as it did under the assumption of uniqueness. Schmalensee [1972] mis-
5-6
-------
X2
Utility
Certainty
Locus
Income Certainty
Locus
tan («)» r1/r2 - p/1-p
Figure 5-2. Optimal allocation of choices of contingent claims
without uniqueness.
interpreted this possibility as an indication that concavity was an inappropriate
definition of risk aversion and selected the equality of marginal utilities as
the characteristic necessary to define risk-averse behavior in the case of
state-dependent utility functions. In summary, the contingent claims model
provides an analytical vehicle that will aid in deciphering the misunderstand-
ings of option value that have developed in the research literature.
5.3 OPTION VALUE: THE "TIMELESS" ANALYSES
The first analytical evaluations of option value employed a "timeless"
framework with the only source of uncertainty associated with the state of the
individual's preference structure (see Cicchetti and Freeman [1971], Bohm
[1975], and Schmalensee [1972, 1975]). To simplify the explanation of these
analyses, assume that individual preferences can be described by just two
states: State 1, which demands the services of the asset with u1(.)/ and
State 2, which does not demand the services of the asset with u2(.)- Each
state's utility function will have two arguments—income, Y, and a variable in-
dicating access to the asset's services, with d implying the services are avail-
able and d implying they are not. This argument can proceed using the com-
pensating variation definitions of consumer surplus, option price, and option
value, but comparable arguments can be developed using equivalent variation.
5-7
-------
Equations (5.6) and (5.7) define consumer surplus for the i state
(SC.) and option price (OP), respectively:
u,(Y. - SC,, d) = u.(Y.,d), 1=1,2 (5.6)
2 2
Z 71. u.(Y. - OP, d) = Z n. u.(Y., d) (5.7)
where
Uj(Y, d) = individual utility for State i with income Y. and with access to
the services of the asset
n. = probability of utility State i (n2 = 1 - 7^).
Substituting Equation (5.7) in Equation (5.6) and rearranging terms gives:
2
Z n. [u^Y. - OP, d) - Uj(Yj - SC-, d)] = 0 . (5.8)
Schmalensee [1972] proposed using concavity of the state-specific utility func-
tions to expand Equation (5.8). That is, the inequalities given in Equations
(5.9) and (5.10) hold for concave Uj(.):*
u,(Yj - OP, d) - u.(Yj - SC., d) > (SC. - OP) [au./BY. (Y, - OP, d)] (5.9)
u.(Y. - OP, d) - u^Yj - SC-, d) < (SC. - OP) [3u./9Y. (Y. - SC., d)].(5.10)
Substituting each into Equation (5.8) and rearranging terms gives inequalities
for option price involving Bohm's [1975] weighted expected consumer surplus
terms as Equations (5.11) and (5.12):
2 2
OP > Z n. SC. [3U./3Y. (Y. - OP, d)] / Z n.[au./aY. (Y. - OP, d)] . (5.11)
j=1 » i i i i .=1 i i i i
2 2
OP < Z n. SC. [3U./9Y. (Y. - SC., d)] / Z n.[au./3Y. (Y, - SC., d)]. (5.12)
- j=1 i i i i i i j=1 i i i i i
Because option value (OV) is defined as the difference between the
option price (OP) and the expected consumer surplus (SC)—i.e., OV = OP
*ln the analysis that follows, the point of evaluation of the partial deriva-
tives will be important to the interpretation given to each relationship. There-
fore, [3u./8Y (a,b)] will refer to the partial derivative of u.(.) with respect to
Y evaluated at the point (a, b).
5-8
-------
2 \
Z Ttj SCj 1 --these inequalities offer the potential for determining the sign
of the option value if it is possible to relate the weighted consumer surplus to
the expected value of the consumer surplus.* Schmalensee's definition of risk
aversion as equality of the marginal utilities of income across states (i.e.,
3u1/3Y1 = 8u2/3Y2) provides the ability to make this association by making the
weights in Equations (5.11) or (5.12) unity. That is, depending upon whether
the equality is realized at Y. - OP or Y. - SC., option price will be greater or
less than expected consume/- surplus. Thus, Schmalensee concludes that the
sign of option value depends on the point of evaluation.
As observed earlier, Bohm has correctly observed that this judgment is
misleading for at least two reasons. First, the interesting expression is Equa-
tion (5.11) because the point of evaluation of the marginal utilities correctly
assigns to the individual the relevant income/access conditions. This expres-
sion describes the relationship between option price and expected consumer
*To illustrate this point let
(Yj-OP, d)
I n. ^-(Y.-OP, d)
i=1 ' 8Yi '
(Y2-OP, d)
w i ^
W2 = "2 §uT~
I n. ^(Y.-OP, d)
This specification will imply Wi + w2 = 1. Consequently, Equation (5.11) can
be rewritten as
2
OP > Z w.SC.
To compare the specification with the expected consumer surplus
requires some knowledge about the relationship between w. and n.. For exam-
a,, a,. i i
pie, if it is assumed that (Yt - OP; d) = (Y2 - OP, d) (the marginal
utilities of income are equal in each period), then w. = n. and Equation (5.11)
allows option value to be signed.
5-9
-------
surplus when the individual's income is reduced by the option price. As Bohm
suggested:
We are asking him how much he can abstain from today in terms of an
OP and enter the future state, whatever it may be, with a disposable
income of Y-OP without being worse off. He will be at Y-SC. only
if he does not pay an option price—and that is another story. We
do not ask him about the maximum amount he is; willing to pay pro-
vided he does not pay that amount. (Bohm [1975], p. 735)
The second consideration involves the Schmalensee definition of risk
aversion. The previous section noted that the conventional definition of risk
aversion, with the uniqueness assumption for state utility functions, simul-
taneously implies that:
The utility function must be concave to admit such a response
to a "fair" gamble.
In response to a fair gamble the risk-averse individual will
always select a point where marginal utilities of income are
equal.
This latter point is a result of optimizing behavior in the presence of a fair
gamble and concavity of the utility functions. Once the uniqueness assump-
tion is relaxed and state-specific utility functions are permitted, the only
plausible definition for risk aversion is by the concavity of the state-specific
utility functions. Thus, when the correct point of evaluation (i.e., the in-
equality given in Equation [5.11]) and the appropriate definition of risk aver-
sion are used, the sign of option value cannot be established. It may be
positive, negative, or zero depending upon the relationship between the mar-
ginal utilities of income at each state.
Given these conclusions, how do Cicchetti and Freeman [1971] establish,
apparently unambiguously, a positive sign for option value while Bohm does
not? To answer this question, return to the example of a "fair" gamble with
state-specific utility functions that was given in Figure 5-2. Schmalensee in-
correctly interpreted this divergence to indicate the inadequacy of u.(.)'s con-
cavity as the sole basis for defining risk-averse behavior. However, Cic-
chetti and Freeman apparently intended to focus on a comparison along the
utility certainty locus.* As Anderson [1981] has recently observed, they as-
*Cicchetti and Freeman seem to have wanted to use the utility certainty
locus to make the state-specific actions commensurate. This can be seen in
their proposal that:
To make the choice problem solvable, there must be some way
of making the utilities of the two alternative mappings commens-
urable. We have proceeded as follows to derive a rule for com-
paring the utilities from the two alternative mappings. For any
level of disposable income, if the individual did not demand the
5-10
-------
sumed that the individual's income was equal across all states and that, when
income was equal, total utilities in each state were also equal at the preferred
price vector. In the present analysis, this would correspond to equal utility
for conditions of access to the resource [i.e., u.(Y., d) = U;(YJ/ d) for Yj =
Y.]. Unfortunately, the Cicchetti-Freeman analysis did not correctly de-
scribe an individual's choices of Y and the services of the asset. While they
proposed to consider a discrete choice similar to the d versus d description,
they represented the services as continuously available, designed by X.
Figures 5-3 and 5-4 reproduce the Cicchetti-Freeman figures (III and IV)
for the analysis. If Figure 5-4 is interpreted as an illustration of the "no-
demand" case, the assertion that u8 = us at Y0 is incorrect. If the relevant
budget constraint, Blf is considered, the individual will not choose to consume
the same level of Y. In the "no-demand" case (i.e., u8), the selected income
will be Y0, but the "demand" case will be Y5 in Figure 5-3. Similar arguments
can be developed for the assumption that Uj = u6 at Y0 - OPj, which indicates
that the construction of Figure 5-4 is incorrect. To adequately deal with the
equivalence of state-specific utility functions at equal income levels, a graphical
analysis must be in terms of indirect utility functions as described by Bishop
[1982]. In this case the ambiguity in the sign of option value is clearly
demonstrated.
In Graham's [1981] recent attempt to use the Schmalensee framework to
comment on the appropriate treatment of option value, he argues that the
reasonableness of using option price for benefit-cost analyses will depend on
the nature of the problem under study. More specifically, Graham concluded
that:
Option price is the appropriate benefit measure for project anal-
ysis when one can assume the individuals affected are similar
and they all experience the same risk.
Expected willingness to pay will be the appropriate measure for
those cases with similar individuals but with risks specific to
each.
These conclusions are derived using a generalization of the option price defini-
tion (Equation (5.7)). To understand them, Graham's arguments must be con-
sidered in detail. For the case of individual risks, he assumes that payments
may be state specific. This is equivalent to the assumption that a complete
set of markets for contingent claims exists. Under these assumptions, the
definition of option price in Equation (5.7) would be replaced by Equation
(5.13):
good, he would choose a consumption point on the Y axis and
experience a certain level of utility; if he were to demand the
good (assuming that it is available), he would choose a tangency
point on the budget line associated with that point, and exper-
ience a given level of utility. We assume that the alternative
outcomes have the same utility. (Cicchetti and Freeman [1971],
p. 534)
5-11
-------
Yft
OP*
»S
Figure 5-3. Option value in Cicchetti-Freeman's analysis.
- f(Y)
Figure 5-4. Option value in Cicchetti-Freeman's analysis with "no demand."
5-12
-------
2
Z
n u (Y - P , d) = Z 7i u (Y , d)
(5.13)
Graham defines this relationship as the willingness-to-pay locus. The special
case of Px = P2 = OP would yield the conventional definition of the option
price. The locus also includes the point where P. = SC. (by construction),
as well as the fair-bet and the utility certainty points, as illustrated in Fig-
ure 5-5.
To illustrate some of the points on the locus, assume TT1 corresponds to
the individual's budget constraint where the prices of claims in States d and
d correspond to the probabilities of each state. F will then designate the
fair-bet point. When payments are constant, regardless of the state of na-
ture, as with point P, the locus describes the willingness to pay under insti-
tutional conditions consistent with an option price, OP. Point S corresponds
to the coordinates of the consumer surpluses for each state. To calculate the
expected value of the consumer surplus, the budget constraint through S
parallel to TT1 is used (to reflect the state probabilities). The intersection
of this new budget line, RR1, with the 45° line defines the expected consumer
surplus. For this example, option value is positive.
$
In
Statt
d
E(S) OP
Sin
Statt d
Figure 5-5. Option value with contingent claims in Graham's analysis.
5-13
-------
Aggregating the willingness-to-pay loci across individuals, Graham ar-
gued that:
Justification of the project hinges upon the question of whether or
not contingent prices exist at which aggregate willingness to pay in
each state exceeds the corresponding resource cost of the project.
Should such prices exist, that point from an individual's locus
which has the greatest value at these prices is the one relevant for
cost-benefit analysis, and the corresponding value at these prices
is the appropriate measure of benefit. (Graham [1981], p. 719)
To apply this approach in particular examples requires that one distinguish:
(1) the benefits realized as a result of moving from an initial distribution of
income to another that assures an efficient distribution of risk and (2) the
benefits resulting from the project itself.
Graham's conclusions are based on two rather special cases. The first
of these avoids the issue of an inefficient distribution of risk by assuming
that individuals are alike and that they face identical risks. The second case
also skirts this issue by assuming the existence of either complete contingent
claims markets or an ideal, state-dependent tax collection scheme (tied to the
project under evaluation). In either case, an efficient distribution of risk
will be realized. Of course, neither of these sets of assumptions is plausible
in most applications, where some attempt must be made to include a measure
of the value of an option to use the services of an environmental resource.
Consequently, as Graham acknowledges, one is left with option price as the
"best" basis for measuring benefits. Thus, for practical purposes, Graham's
analysis has strengthened Bohm's conclusion: Option price is the relevant
focus for applied welfare economics.
Given these conclusions, why worry about the sign and magnitude of op-
tion value? One pragmatic reason arises with the difficulty in measuring each
individual's option price. If it is possible for wide classes of assets and their
associated prospective users to demonstrate that the corresponding option
values of the assets would be positive, one would be safe in assuming that
measures of the expected user benefits (i.e., as derived from an "ideal" con-
sumer surplus calculation) would understate the total benefits provided by the
asset.*
5.4 THE TIME-SEQUENCED ANALYSES
Time-sequenced evaluations of option value offer more specific answers to
the three questions raised at the outset. That is, these analyses provide an
explicit statement of the relationship between decisions over time. In gene-
ral, the uncertainty is supply related. It is resolved with the passage of
time, and decisions cannot be altered. The first of these models was devel-
*This argument ignores the potential role of existence values as de-
scribed by Krutilla [1967] and more recently discussed by Freeman [1981].
5-14
-------
oped by Arrow and Fisher [1974], whose framework introduced a time-
sequencing of decisions and, as a result, assumed there was a resolution of
the uncertainty facing the decision process with the passage of time. Their
model considered decisions to develop or preserve a fixed amount of land.
Decisions to develop some fraction (or all) of the land were irreversible.
Therefore, any information acquired with the passage of time could affect only
the decisions made on the remaining stock of preserved land.
Arrow and Fisher's quasi-option value can be interpreted as the ex-
pected value of the information obtained through delay, as has been sug-
gested by Conrad [1980] and, indeed, acknowledged earlier by Krutilla and
Fisher [1975] in their overall evaluation of special problems associated with
allocation decisions involving unique natural environments. For example,
Krutilla and Fisher observed that:
The key new element in Arrow and Fisher is a Bayesian information
structure. The passage of time results in new information about
the benefits of alternative uses of an environment, which can in
turn be taken into account if a decision to devote it to development
is deferred. Since development is not reversible, once a decision
to develop is made, it cannot be affected by the presence of new
information which suggests that it would be a mistake in the future.
The main result of the analysis is then that there is an option
value, or quasi-option value, to refraining from development—even
on the assumption that there is no risk aversion, and only expected
values matter. (Krutilla and Fisher [1975], pp. 70-71)
Conrad also suggested that option value could be interpreted as the ex-
pected value of perfect information. In so doing, he implicitly maintains that
over time one progressively learns of and resolves the uncertainty. However,
his conclusion is correct only if it is regarded as the only appropriate trans-
lation of the "timeless" analysis of option values into a time-sequenced deci-
sion process. Henry [1974] has drawn a similar conclusion in his evaluation
of the importance of this transition, noting that:
The relationship so established between risk aversion and option-
price appears rather obvious when it is viewed as being en-
countered in a 'timeless world' where I [the individual] has one and
only one decision to take; iri a world of this type any decision is
just as irreversible as any other [emphasis added] and it is impos-
sible to introduce Krutilla's option value which is nothing but a risk
premium in favour of 'irreplaceable assets'. Krutilla's idea can only
be examined in a 'sequential world' where | [the individual] really
has a succession of decisions to take. (Henry [1974], p. 92)
Thus, if it is assumed that uncertainty is resolved over time, that the
asset under consideration is in some respect irreplaceable, and that the deci-
sions are made sequentially with the benefit of the acquired information, there
is clearly a positive option value. If, on the other hand, the resolution of
the uncertainty is not allowed as a part of a set of decisions, option value
will be a reflection of risk aversion, and its sign will depend on the nature of
the state-specific utility functions.
5-15
-------
This distinction has important implications for any attempts to develop
estimates of option price. If a direct survey or contingent valuation method
is used to obtain these estimates, the results will be based on hypothetical
conditions in which it is unlikely that respondents can be given a means of
obtaining information and reacting to it. That is, as a practical matter, it is
probably safe to assume that questions designed to elicit an individual's op-
tion price will not be posed in a way that identifies mechanisms through which
the individual can obtain information and alter decisions based on it. Thus,
the timeless analyses are more likely to be the relevant models for understand-
ing the empirical measurement of option value. However, this judgment does
not imply that a careful description of the source of uncertainty and the
means through which it is resolved can be ignored in question design.
Rather, it simply recognizes that formulating questions that acknowledge the
prospects for learning and that offer mechanisms for enhancing learning would
likely increase the complexity of the instrument to a point where it was not
usable.
Together with extensions of it in Smith [1983], this analysis suggests
that supply uncertainty can be important to the sign of option value in a time-
less framework. Accordingly, supply uncertainty should be acknowledged and
explicitly identified in questionnaires designed to measure option price.
5.5 RECENT ESTIMATES OF NONUSER VALUES
There appears to have been only one published study estimating option
values. This study by Greenley, Walsh, and Young [1981] attempts to meas-
ure the option value for the recreational use of preserved water quality in
the South Platte River basin in Colorado. These authors used two payment
vehicles—an increment to the sales tax and an increase in the monthly water-
sewer fee—in a survey of a random sample of 202 residents of Denver and
Fort Collins. Their study attempted to estimate specific components of the
benefits of maintaining water quality, including option, user, existence, and
bequest values. Their paper focuses on the results of the question for op-
tion value. Two aspects of their option value question are important. First,
it seems to be eliciting an option price, not option value, and specifies a res-
olution of the supply uncertainty associated with the preservation of water
quality. And, second, the question treats the two payment vehicles differ-
ently. The question is reproduced below:
Given your chances of future recreational use, would you be willing
to pay an additional cents on the dollar in present sales taxes
every year to postpone mining development? This postponement
would permit information to become available enabling you to make a
decision with near certainty in the future as to which option (re-
creational use or mining development) would be most beneficial to
you. Would it be reasonable to add to your water bill every
month for this postponement? (Greenley, Walsh, and Young [1981],
p. 666, emphasis added)
5-16
-------
As discussed earlier, option value is the difference between an individ-
ual's option price and his expected consumer surplus. It would seem that this
question is soliciting the option price. Unfortunately, the authors interpreted
the responses as measures of the option value and asked a separate question
intended to obtain user values. The Greenley, Walsh, and Young results with
the sales tax payment vehicle indicate an average option value of approximately
$23.00 per year (with the water fee payment vehicle, it was $8.90).*
The interpretation of these results has been somewhat controversial.
Both questions used by Greenley, Walsh, and Young seem to be asking for an
option price--the first under a timeless interpretation and the second under a
time-sequenced format. Greenley, Walsh, and Young interpret one as a meas-
ure of expected consumer surplus and the other as option value. Mitchell and
Carson [1981] appear to have been the first to question the interpretation of
the Greenley, Walsh, and Young questions. While Mitchell and Carson did not
relate their criticisms to the two conceptions of option value, they did argue
that both questions measure option price. Moreover, they suggested that the
Greenley, Walsh, and Young results indicate the possibility of a starting point
bias, based on the differences in designated starting points used for each
payment vehicle. In a recent unpublished response to the Mitchell-Carson
comments, Greenley, Walsh, and Young [1983] argue that the interviewing
process itself prevented interpretation of the questions as requesting option
price. They observe that:
Some confusion may arise when expected consumer surplus and op-
tion value questions are taken out of the context in which they are
used because they often take the same general form as questions
asking for option price. . . . The important distinction in this case
[their study] is that a population of users was first asked to esti-
mate their expected consumer surplus, and in addition, a separate
estimate of option value. They were informed that these are sepa-
rate and distinct values, and provided the opportunity to adjust
values previously reported. The respondents provided well-focused
estimates for each question. We conclude that the procedures
employed in our study capture, reasonably accurately, the values
necessary to assess the recreational benefits of improved water
quality." (Greenley, Walsh, and Young [1983]).
While this may be the case, no explanation is offered of why the house-
holds adjust their two bids. If each is measuring what the authors intended,
there would be no basis for adjustment. Equally important, one can judge
the responses to a contingent valuation experiment based only on the questions
posed. If they are not clearly connected to the concept desired, there is
reason to question whether informal discussions between the interviewer and
respondent will assure understanding. Finally, our evaluation of the questions
(in contrast to Mitchell and Carson) leads to the conclusion that two different
concepts of option price are in fact asked.
*lt should be noted that these summary statistics include all zero
bids--both the "true" zero bids and the zero bids of those individuals who
refused to participate in the bidding game.
5-17
-------
Of course, in fairness to all participants in the exchange, there is no
complete record of exactly what the interviewers discussed with survey re-
spondents. Greenley, Walsh, and Young's [1983] recent notes on the Mitchell-
Carson critique suggest that they were aware of the potential ambiguity in
their questions. What is at issue is not only how successful the interviewers
were in overcoming it but also that the terms of the contingent market may
differ for each respondent (because of the interviewer effect) making the
results problematical.
The second empirical study focusing on user and nonuser values was con-
ducted by Mitchell and Carson [1981]. It sought to measure each individual's
willingness to pay for cleaning up a\± rivers and lakes in the United States to
a particular level. Since individuals were not classified according to whether
or not they used these water resources, the responses must be assumed to
include both use and nonuse values.* Indeed, Mitchell and Carson argue that
it is beyond the capability of many respondents to reliably determine separate
values for subcategories of water quality benefits. Their survey was based
on a national probability sample of 1,576 individuals and was conducted as part
of an opinion poll soliciting these individuals' responses to other questions
associated with environmental attitudes. This study introduced the water
quality ladder used in the survey conducted for the present study. In addi-
tion, it assumed that the household payment vehicle was through higher prices
and taxes (the same vehicle used in this survey). Four versions of an
anchored payment card were used, rather than an iterative bidding framework.
They were differentiated according to the range of values reported on the
cards and by the anchor points reported. The cards were distinguished by
income class so that the anchored values on the card corresponded to the
average of the actual payments made by members of each income group. The
four sets of anchor points used in this study were:
Version
Average household expenditures (through taxes) to the space
program, highways, public education, and defense.
B Same four public goods as in Version A plus police and fire
protection.
C The same four public goods as in Version A, but amounts in-
creased by 25 percent for each income group over the levels
used with Version A.
D The same four public goods and amounts as in Version A plus
the estimated amount for water pollution control.
*Since individuals do not conceive of using aM rivers and lakes in the
United States, it must be assumed that only a subset of these can be consid-
ered a part of the set actually used or planned for future use. To the extent
that individuals express a willingness-to-pay bid for improved water quality
at all water bodies, they are expressing expected user values, any option
values (associated with uncertain future use), and existence values.
5-18
-------
Table 5-1. Summary of Mitchell-Carson Estimated Mear>a
Annual Willingness to Pay by Version and Water Quality
Water quality Version of payment card
category A (274) B (255) C (244)
Boatable $168 $133 $161
Fishable $214 $180 $198
Swimmable $247 $212 $222
aThis table was summarized from Mitchell and Carson's [1981] Table 5-1, p.
5-3. The numbers in parentheses are the numbers of respondents providing
values to the water quality questions for each version in 1980 dollars.
For three of the four versions of the payment card, Table 5-1 reports
the mean estimates for beatable, fishable, and swimmable water qualities.*
While this study provided detailed analysis of potential survey biases, its ques-
tions relate to an abstract conception of the impacts of a water quality im-
provement for the individual. That is, while the water quality is described
as improving to levels defined by the activities—swimmable, fishable, and
boatable--the quality of the water already available to the individual is un-
known. If the water bodies available to the individual have quality levels that
permit the full range of his desired uses, the responses might be expected to
reflect an existence value for all other sites. By contrast, if the available
sites for water-based recreation do not permit all or some subset of these activ-
ities, the responses may reflect user values. Without knowledge of these site-
specific features, Mitchell and Carson must average heterogeneous responses.
That is, ideally, the responses based on user values and those associated with
nonuser values should be distinguished. Moreover, the analysis should control
the influence of the differential availability to individuals of sites with the
desired water quality. The Mitchell-Carson method implicitly assumes all indiv-
iduals will benefit equally from the uniform improvement of the water quality
at all sites. This may not be correct. The benefit realized by each individual
will depend on his access to sites with the desired water quality before the
change.
Mitchell and Carson estimate the nonuser benefits of water quality im-
provements by assuming that the willingness-to-pay responses of surveyed
The effects of knowing what was actually paid for water quality control
(i.e., version D) were also reported by the authors. Forty-seven percent of
the 354 respondents to version D said they were willing to pay the amount
shown on the card that they were told would raise water quality to fishable in
the next few years. For further details on these results, see Mitchell and
Carson [1981, pp. 5-6 to 5-7]. The figures are not reported here since they
reflect only that some people were'willing to pay at least these amounts.
5-19
-------
individuals who did not engage in in-stream recreation will be "almost purely
intrinsic in nature." Even if this reasoning is correct, it does not imply that
nonuser willingness to pay will be a reasonable estimate of option value. It
may include existence values as well. Nonetheless, based on this logic, 39
percent of the respondents with willingess-to-pay data reported they had no
in-stream use of freshwater in the past 2 years. The nonusers mean bid for
fishable water was $111. The mean bid by users for the same water quality
change was $237. Hence, by these estimates, intrinsic values were judged to
be approximately 45 percent of total willingness to pay of users.
Rae [1981a, 1981 b] has also reported estimates of option price for "clear"
visibility conditions for future visits of current users in two separate onsite
surveys in 1981 at the Mesa Verda National Park and Great Smoky National
Park. His analysis was conducted along with a contingent ranking evaluation
of the benefits of improving visibility conditions (see Chapter 6 for a more
complete summary). A payment card was used as the instrument, and
respondents were asked how much they would pay for an insurance policy to
guarantee clear visibility conditions for all visits to the park. Prices on the
card ranged from 0 to $10 in increments of $0.25. The average bid was $4.17
for Mesa Verda respondents and $5.96 for Great Smoky respondents (estimates
in 1981 dollars). Rae interprets this as a present value option price, and uses
estimates of current user values for visibility improvements derived from the
contingent ranking framework to estimate option value.
To make Rae's interpretation requires assumptions concerning the indi-
vidual's rate of time preference and probabilities of future visits. Rae uses
different assumptions in estimating option value in the two studies. For the
Mesa Verda case, he assumed a zero discount rate and one future visit while,
with the Great Smoky case, he postulated an 8 percent discount rate and a
0.77 probability of one return visit after 5 years. The expected user values
estimated for the two cases were $3.00 and $5.00, respectively. Both sets of
assumptions assure a positive estimate of the option value.
In order to evaluate these estimates, the Rae methodology for estimating
user values with the contingent ranking framework must Be considered. In
the next chapter we will discuss, in detail, the use of the contingent ranking
approach for benefit measurement. Equally important, the formulation of the
question for option price is somewhat vague in its specification of the terms
of payment for the insurance. It has been interpreted as a one-time payment
in the analysis. Given that all the other components of the survey related to
fees associated with use, this distinction may not have been appreciated by
the survey respondents.
Finally, the estimation of option value requires assumptions on the time
horizon, future level of use, future probabilities of each level of use, and the
individual rate of time preference. Rae's example calculation was intended to
illustrate the required calculations. Unfortunately, there is little basis for
assuming values for each of these variables for his survey respondents.
5-20
-------
The last empirical effort at measuring nonuser benefits for an environ-
mental amenity is the Schulze et al. [1981] analysis of visibility at four national
parks. This survey was structured to distinguish users from nonusers of the
Grand Canyon. Each group was asked different questions. The users were
asked about the effects of visibility on their user values, while the nonusers
were asked about preservation values. The questions related to visibility at
four national parks, to the overall region, and to an evaluation of the willing-
ness to pay to avoid a visible plume. The respondents were drawn from four
cities: Los Angeles, Albuquerque, Denver, and Chicago. Questionnaires for
users employed a park fee as the payment vehicle, while nonusers were queried
about their willingness to pay for preservation values through electric utility
increases.
Their results suggest a substantial preservation value (in 1980 $) ranging
from $3.72 (the average value for preserving visibility at the Grand Canyon
by Denver respondents) to $9.06 (the average for Chicago respondents) per
month. These are substantially greater than the estimated user values, which
ranged from $0.99 to $5.40 per visit for a comparable visibility scenario. If
it is appropriate to compare these results across different individuals (i.e.,
implicitly assuming users would also have a preservation value), the estimated
preservation values for preserving visibility conditions at unique natural en-
vironments, such as the Grand Canyon, may be much greater than the user
values for the same visibility conditions. Unfortunately, the study does not
attempt to divide the preservation benefit into an estimate of option price and
an estimate of existence value. Thus, it is not directly comparable to either of
the two studies discussed earlier in this section. Furthermore, the choice of
two different payment vehicles may have introduced a starting point bias prob-
lem similar to that in the South Platte River study.
Thus, in summary, all past efforts at measuring nonuser values have met
with only limited success. There has been controversy over whether option
values were measured or it has not been possible to distinguish option price
from other components of intrinsic values.
5.6 MEASURING OPTION VALUE: SURVEY DESIGN
As noted in Chapter 1, an important component of the Monongahela sur-
vey was the measurement of option price and user values. In addition, the
question design permitted the implications of supply uncertainty for the esti-
mates of option value to be examined. Since Chapter 3 described the sample
survey design and Chapter 4 provided a summary of the features of the final
sample, these will not be repeated here. Rather, this section will review the
background information provided to each respondent and the form of the ques-
tions used to derive estimates of the option value associated with various water
quality changes in the Monongahela survey.
As noted, the payment vehicle was described to be the taxes paid directly
and the higher prices paid indirectly for improved water quality. This ap-
proach follows the format used by Mitchell and Carson [1981] with several im-
portant additions. Each interviewer was trained to explain carefully the mech-
anisms that underlie the payment vehicles. The objective of these explanations
5-21
-------
was to ensure that respondents understood the nature of the payment vehicle
and recognized that similar types of payments take place in practice as a result
of government and private sector decisions. Each respondent was shown a map
of the area highlighting the locations of recreation sites along the river. This
map is reproduced as Figure 4-3. Before proceeding to the questions, the
interviewer described the reasons why one might be interested in water quality
for the Monongahela River. Using a value card (i.e., Figure 4-6), actual use,
potential future use, and existence values were each identified as separate
reasons for interest in the river water quality. Each was acknowledged to be
a potential motivation for valuing water quality in the Monongahela River. The
value card was explained at the outset of the interview and then shown again
to each sample respondent as the questions designed to separate option price,
expected consumer surplus, and existence values were asked. Thus, the value
card translated the theoretical relationships relating option value, user value,
and existence value into a format that linked them to respondents' experiences.
There are at least two ways to ask questions designed to measure the
option values associated with water quality. The first of these involves pro-
posing to respondents counterfactual situations that describe, in hypothetical
terms, the probabilities and levels of use of the resource with different speci-
fied water quality levels. Each respondent is asked to value these plans. A
second approach relies on the interviewer's ability to explain to the respond-
ent why he might value water quality at a site, identifying the relationships
between those reasons and a benefits taxonomy that isolates option value.
With this explanation, the individual is then asked to bid in a way that sepa-
rates the individual components of the values.
These methods contrast with a third approach employed by Mitchell and
Carson, where a classification of individuals (as users or nonusers) assisted
in decomposing benefits. That • is, their classification, together with the
assumption that nonusers were always nonusers and therefore could not have
user values, allowed'the willingness-to-pay estimates from nonusers to be
interpreted as indicative of the intrinsic benefits held by users.
In the absence of the assumption that individuals are comparable (except
in the decision between use or nonuse), the first two approaches to partition-
ing the benefits of a water quality improvement face problems. The first one
attempts to "second guess" plausible demand conditions in its specification of
the probabilities and levels of use that might be associated with a water quality
level. Such specifications may actually bear little resemblance to what an
individual would select. Thus, this approach was not used in this analysis.
The second approach relies on individuals' ability to "divide the benefits
pie" consistently. Clearly, the estimates in this study depend not only upon
how well each individual understood the concepts on the value card, but also
upon how well he was able to (1) use them in classifying the contributions
made to overall option price by expected user benefits and option values and
(2) separate existence values as a distinct motive for valuing water quality
improvements.
5-22
-------
The survey questions elicited an option price—the individual's willing-
ness to pay for the water quality change due to actual and potential use of
the river. Following this question, the interviewer asked each person what
amount of the option price was associated with actual use. This response has
been interpreted as an estimate of the individual's expected consumer surplus.
Thus, the difference between the reported option price and the value associ-
ated with use corresponds to this study's estimate of option value.
The questionnaire design allowed evaluation of two further issues in the
measurement of option value: (1) the amount of the water quality change and
(2) the mode of questioning. The design considered three levels of change in
water quality as reproduced in the water quality ladder shown in Figure 4-5.
The first question considered the willingness to pay to avoid having the water
quality deteriorate from its current level, Level D, acceptable for boating, to
Level E, at which no recreation activities would be possible. Individuals were
also asked their willingness to pay for improvements from Level D to Level C,
acceptable for sport fishing, and improvements from Level C to Level B,
acceptable for swimming. As noted in the previous chapter, the water quality
levels were defined based on Resources for the Future's water quality index
(see Mitchell and Carson [1981]).
The second aspect of the questionnaire design involved the mechanism
used to elicit the willingness-to-pay response. To investigate the effects of
different questioning methods, the sample was divided into approximately four
equal parts, each using a different questioning method—two different iterative
bidding game procedures, a direct question procedure, and a procedure using
a direct question with a payment card. Iterative bidding games, practiced in
most early contingent valuation experiments (see Schulze, d'Arge, and Brook-
shire [1981] for a review), involve a sequential process in which an inter-
viewer proposes a value (the starting point) to the respondent and asks
whether it would be acceptable as a bid for the conditions described in the
question. Based on the response, the interviewer raises or lowers the bid by
a fixed amount until there is no change in the bid with repetition of the proc-
ess. Two subsets of the sample used bidding game procedures; the first used
a $25 starting point and a $5 increment, and the second used a $125 starting
point and a $10 increment.
The third procedure used to elicit individual willingness to pay was a
direct question with no suggestion of an amount. In the last component of
the sample, respondents were asked to look at a payment card (see Figure
4-7) arraying alternative dollar values and to select one or any other amount
as their willingness to pay. This last procedure is comparable to the Mitchell-
Carson [1981] approach, with one modification. The values on the card were
not identified as the individual's current spending on specific public sector
activities. This practice of anchoring the values was not used because it was
felt it would create the possibility of biased responses.
Each subsequent question for user values, supply uncertainty, and exist-
ence values repeated the amount given for willingness to pay and then asked
the respondent to indicate what portion of the reported willingness to pay is
5-23
-------
Table 5-2. Summary of Willingness-to-Pay Questions
by Type of Interview
Type of interview
Question format
Iterative bidding $25
Iterative bidding $125
Direct question
Payment card
To you (and your family), would it be worth $25
each year In higher taxes and prices for products
that companies sell to keep the water quality in
the Monongahela River from slipping back from
Level D to Level E?
To you (and your family), would it be worth $125
each year in higher taxes and prices for products
that companies sell to keep the water quality in
the Monongahela River from slipping back from
Level D to Level E?
What Ms the most it is worth to you (and your
family) on a yearly basis to keep the water qual-
ity in the Monongahela River from slipping back
from Level D to Level E, where it is not even
clean enough for boating?
What is the most it is worth to you (and your
family) on a yearly basis to keep the water qual-
ity in the Monongahela River from slipping back
from Level O to Level E, where it is not even
clean enough for boating?
Table 5-3. Summary of User, Supply Uncertainty,
and Existence Value Questions
Type of response
Question format
User value
Supply uncertainty
Existence value
In answering the next question(s), keep in mind
your actual and possible future use of the Monon-
gahela. You told me In the last section that it
was worth $(AMOUNT) to keep the water quality
from slipping from Level D to Level E. How much
of this amount was based on your actual use of
the river?
If the water pollution laws were relaxed to the
point that the water quality would decrease to
Level E and the area would be closed 1/4 of the
weekends of the year for activities on or in the
water but would remain open for activities near
the water, how much would you change this
(READ TOTAL $ AMOUNT) to keep the area open
all weekends for all activities?
What is the most that you (and your family)
would be willing to pay each year in the form of
higher taxes and prices for the goods you buy
for keeping the river at Level D where it is okay
for boating, even if you would never use the
river?
Suppose the change could not be reversed for an
even longer period of time than your lifetime.
How much more than (READ AMOUNT FROM a.)
would you (and your family) be willing to pay per
year to keep the river at Level D, even if you
would never use the river?
5-24
-------
associated with each of the components of value or complications to the choice
process. Table 5-2 reports the form of the willingness-to-pay questions used
for the case of preventing deterioration from water quality Level D to Level E
for each mode.
The questions used to measure the values associated with use, supply
uncertainty, and existence values did not change with the type of interview
and samples and are reported in Table 5-3. The examples correspond to the
scenario used for the willingness-to-pay questions in Table 5-2. The re-
sponses to these questions form the basis for the results reported in the next
section of this chapter.
5.7 SURVEY RESULTS—OPTION VALUE
The results for the empirical estimates of option value are divided into
two parts. The first considers the conventional treatment of option value as
a response to demand uncertainty. The second considers the sensitivity of
these findings to changes in the conditions of access to the Monongahela River
by varying the proposed likelihood of being able to use the site.
5.7.1 Option Value—Demand Uncertainty
Table 5-4 presents a summary of the sample mean estimate of option
value for each water quality change based on each of the four types of inter-
view frameworks. The estimates for each water quality change are the incre-
ments to the reported willingness to pay to prevent the water quality from
deteriorating to the level given as E. Thus, each respondent was asked if he
would be willing to pay more than the amount recorded for avoiding a move-
ment from D to E. When an affirmative answer was given, the interviewer
proceeded with the increments from D to C and from C to B. Since
some individuals were unwilling to pay for further improvements, the "no" re-
sponses to subsequent improvements were treated as zeros in constructing the
means.
Analysis of the survey responses revealed that two definitions of "users"
were possible. The first of these would classify individuals according to
whether they reported a user value or indicated that they had used the river
for recreation activities in the previous year. This definition is the focus
of attention in this chapter and is designated as the "broad definition" of
users. The second defines users as only those individuals who indicated that
they had used the Monongahela sites. This narrow definition focuses on a
subset of the users under the first definition. Appendix C reports a sample
of the results under the narrow definition.
The analysis performed for this study has considered both the sample
means and linear regression models to summarize the survey results. Table
5-4 provides estimates for option value for different levels of water quality
change according to the survey instrument used. Informal review of these
estimates seems to indicate that the question format influences the magnitude
of the estimates. Following the practices described in Chapter 4, these esti-
mates are based on a restricted sampfe: Observations identified as either
5-25
-------
Table 5-4. Estimated Option Values for Water Quality Change:
Effects of Instrument and Type of Respondent--
Protest Bids and Outliers Excluded
Type of respondent
Change in — ^£ Nonuser
water quality X s n X s n
1. Iterative Bidding Framework, Starting Point = $25
D to E (avoid) 20.79 16.61 19 29.74 36.69 39
D to C 14.74 13.99 19 14.49 15.17 39
C to B 6.84 10.70 19 7.18 11.63 39
D to B 21.58 22.05 19 21.67 24.04 39
2. Iterative Bidding Framework, Starting Point = $125
D to E (avoid) 58.44 66.60 16 38.75 51.32 32
D to C 37.81 49.13 16 26.25 45.38 32
C to B 13.13 32.65 16 11.56 33.06 32
D to B 50.94 71.44 16 40.47 69.02 32
3. Direct Question Framework
D to E (avoid) 25.59 43.04 17 14.18 27.12 34
D to C 10.12 24.45 17 10.82 21.56 34
C to B 10.18 24.49 17 8.47 21.87 34
D to B 21.77 48.57 17 20.32 41.45 34
4. Payment Card
D
D
C
D
to E
to C
to B
to B
(avoid)
27.06
14.41
3.26
20.00
33
20
8
25
.12
.38
.28
.06
17
17
17
17
52.
21.
7.
29.
97
89
70
87
76.
33.
19.
47.
31
80
99
54
37
37
37
37
These results are based on the broad definition of users.
protest bids or as rejecting or misunderstanding the contingent valuation ex-
periment were deleted. The latter were initially identified as outlying obser-
vations using regression diagnostics (see Belsley, Kuh, and Welsch [1980]).
This statistical identification was followed by an evaluation of the features of
the observations that made them distinct (see Table 4-8 and Section 4.5 for
further discussion). To consider this issue, as well as the potential effects
of being a user of the river, several null hypotheses have been chosen for
testing using a student t-test for the difference of sample means. Equation
(5.14) below provides the test-statistic used for these tests:
5-26
-------
t = - (5.14)
Jt
x, -
(ni - 1) Si2 + (n2 - 1
(nt + n2 - 2)
*9
) s22 r,
n
h + n2
1 * n2
where
X. = sample mean for the ith grouping of individuals (e.g., users,
1 nonusers, respondents with a particular question format, etc.),
s. = sample standard deviation for ith grouping of individuals
n. = sample size for the ith grouping of individuals.
All combinations of questioning format for each type of improvement in
water quality were compared for users and nonusers. Overall, there were
only a few cases where the estimated means were significantly different. As
a rule these cases were associated with comparisons of the iterative bidding
framework under the two starting points. Thus, there is some evidence of
starting point bias with this approach to soliciting an individual's valuation of
water quality. Indeed, these results for starting point bias would be
strengthened if the observations that were deleted as invalid (from the diag-
nostic analysis) were included in the sample. In several cases it was not
possible to distinguish the effect of the higher starting point (i.e., $125) as
an explanation of the observation's role as an outlier from another character-
istic of the survey respondent involved (see Chapter 4). Table 5-5 summar-
izes the cases where statistically significant differences in the mean values for
option value were found.
Table 5-5. Student t-Test Results for Question Format3
t- Ratios
Means compared User Nonuser
Direct question vs. iterative bidding with -2.069 -2.452
$125 starting point D to C
Iterative bidding with $25 starting point vs. -2.384
iterative bidding with $125 starting point D
to E (avoid)
Iterative bidding with $25 starting point vs. -1.960
iterative bidding with $125 starting point D
to C
Direct question vs. iterative bidding with — -2.035
$125 starting point D to E
Direct question vs. iterative bidding with -- -2.758
$125 starting point D to B
aThis table reports only the cases where statistically significant differences in
the means were found at the 0.05 significance level.
5-27
-------
The responses of users and nonusers were also compared for each type
of question and level of water quality change. Based on observation of values
in Table 5-4, none of these cases indicated a significant difference in the
means. Thus, despite the appearance of rather large differences for a few
cases (e.g., payment card with Level D to Level E), the estimated means are
not significantly different.
Table 5-6 reports the findings of a sample of the linear regression models
considered in attempting to explain the determinants of the option value esti-
mates using the survey respondents' economic and demographic characteristics.
These models should not be interpreted as estimates of a behavioral model.
Rather, they were estimated as summaries of the survey data in an attempt to
Table 5-6. Regression Results for Option Value Estimates--
Protest Bids and Outliers Excluded3
Water quality changes
Independent variables
Intercept
Sex (1 if male)
Age
User (1 if user)
Education
Income
Direct question
Iterative bidding
game ($25)
Iterative bidding
game ($125)
Willing to pay cost of
water pollution (1 if
very much or some-
what)
D to E
(avoid)
-17.014
(-0.540)
4.121
(0.484)
-0.411
(-1.637)
-18.454
(-2.097)
4.830
(2.052)
0.0005
(1.384)
-26.128
(-2.356)
-12.681
(1.188)
14.638
(1.245)
16.069
(1.842)
D to C
-7.170
(-0.380)
-0.133
(-0.026)
-0.216
(-1.435)
-10.609
(-2.011)
2.084
(1.477)
0.00005
(0.210)
-7.472
(-1.124)
-0.274
(-0.043)
20.601
(2.923)
16.611
(3.176)
C to B
10.149
(0.692)
-2.332
(0.589)
-0.131
(-1.120)
-4.518
(-1.104)
-0.167
(-0.152)
0.0002
(1.035)
3.335
(0.646)
1.773
(0.357)
7.575
(1.385)
4.510
(1.111)
D to B
3.635
(0.126)
-3.301
(-0.424)
-0.350
(-1.523)
-15.761
(-1.958)
1.986
(0.922)
0.0002
(0.532)
-3.817
(-0.376)
0.339
(0.035)
29.627
(2.754)
23.229
(2.910)
R2
F
Degrees of freedom
0.212
4.34
155
0.208
4.23
155
0.053
0.90
155
0.170
3.30
155
aNumbers in parentheses are asymptotic t-ratios for the null hypothesis of no
association.
5-28
-------
improve the ability to describe the attributes of individual respondents that
seemed to influence the estimates of option value. Thus, while these results
have limited explanatory power, as measured by the R2 of each equation, they
do provide somewhat different insights into the role of the type of respondent
than those offered by the analysis of sample means. The independent variables
in the model included qualitative variables for sex, question format (with the
payment card as the omitted questioning mode), user, and the individual's ex-
pressed attitude toward paying for water quality improvements. The last of
these was coded as a 1 if the individual "strongly" or "somewhat" considered
himself a person willing to pay the cost required to control water pollution.
Otherwise, the variable was coded as zero (i.e., for individuals who had little
or no such feelings or had no opinion on the matter).*
After the survey respondents' characteristics were controlled, users
seemed to have lower option values than nonusers. No differences were found
using tests based on sample means. Since the tests for the equality of means
did not control for the respondents' characteristics, the difference in the two
conclusions is not surprising. The regression results add further support to
the conclusion for a starting point bias. Two of the four models in Table 5-6
indicate that the qualitative variable identifying the respondents who received
the iterative bidding questionnaire with a $125 starting point was significantly
different from zero. This implies that these responses are significantly differ-
ent than those received using the payment card. The two most consistent
determinants of the option value results in these models were the qualitative
variables for user and for the individual's willingness to pay the costs required
for water pollution control.
Overall, these results indicate that it is possible to estimate option value
for water quality changes. In general, the estimates are significantly different
from zero. The effects of payment vehicle suggest that there appears to be
a starting point bias with several estimates of option value for specific water
quality changes. Morever, with the ability to control for respondents' charac-
teristics, the iterative bidding approach with a $125 starting point was found to
increase option value estimates over the responses made using a payment card.
The results were not especially successful in isolating the effects of other
individual characteristics on the option value estimates. Only the variable in-
dicating the individual's attitude toward paying for water pollution control was
a consistent determinant of the option value estimates for the water quality
changes.
These estimates are all based on the assumption that access to the site is
guaranteed. Accordingly, the implications of supply uncertainty for the re-
spondents' option prices are considered next.
5.7.2 Option Value—Supply Uncertainty
Because the theoretical analysis of the sign of option value and the re-
sults in Smith [1983] suggest that individuals' assumptions regarding their
*A more detailed description of these variables is provided in Chapter 4.
5-29
-------
ability to gain access to the site—i.e., the degree of perceived supply uncer-
tainty—may be important to the magnitude of option value, several questions
were incorporated to attempt to measure its effects on individual's responses.
Table 5-3 reported the question used to gauge the effects of supply uncer-
tainty. Three variants of the question were posed, each of which referred to
the amount an individual would be willing to pay to prevent water quality in
the Monongahela River from deteriorating from boatable to unusable. Supply
uncertainty was introduced by suggesting that the water quality deterioration
would take place and that it would reduce the probability of having access to
the river's recreation sites. The first question postulated that activities on
or in the water would be precluded for one-fourth of the weekends in the
year. The respondent was informed that it would not be known in advance
which weekends would be involved. The fraction of weekends during which
the sites were closed was progressively increased through two more steps to
one-half and three-fourths of the weekends. Table 5-7 reports the estimated
mean adjustments to the original bids made by users and nonusers. That is,
each respondent was reminded of his bid to prevent water quality from
deteriorating from Level B to Level E and then asked how much this amount
would be altered to reflect the supply uncertainty.
These responses indicate that supply uncertainty clearly affects the option
prices bid by users. The means for users under each of the three conditions
of supply uncertainty are significantly different from zero at the 5-percent
level. These results suggest that the option price would be reduced if the
water quality level led to uncertain availability of the site. The mean adjust-
ments to the option prices reported by nonusers were not significantly different
from zero.
Table 5-7. Effects of Supply Uncertainty on Option Price"
Condition of water quality change
Avoid a certain change B to E
Experience water quality change
to E, lose 1/4 weekends
Experience water quality change to
E, lose 1/2 weekends
Experience water quality change to
E, lose 3/4 weekends
Summary
statistics
X
s
n
X
s
n
X
s
n
X
s
n
Userb
114.710
112.501
69
-14.552
52.328
67
-22.537
58.331
67
-26.866
68.500
67
Nonuser
61.817
85.40
142
-6.354
39.891
96
-5.833
43.996
96
-6.042
46.220
96
These results are based on a sample that deletes protest bids and the obser-
vation's identified as inconsistent with the contingent valuation framework.
The difference in the number of observations between the certain case and
the uncertain cases reflects missing observations.
5-30
-------
Table 5-8. Student t-Tests for the Effects of
Supply Uncertainty for Users
Means t-Ratio
Water quality reduces access for:
(1) 1/4 weekends vs. 1/2 weekends 0.834
(2) 1/4 weekends vs. 3/4 weekends 1.169
(3) 1/2 weekends vs. 3/4 weekends 0.394
Table 5-8 reports the results for tests of the differences in the mean
adjustments with progressive increases in the degree of supply uncertainty.
The results suggest that the mean adjustments are not significantly different
with increases in the uncertainty in the availability of the site.
In summary, these empirical findings confirms the theoretical arguments
developed earlier. Supply uncertainty can be expected to affect option value.
Avoiding supply uncertainty and the associated risk is further basis for a
positive option value.
5.8 EXISTENCE VALUE ESTIMATES
Since they were first introduced by Krutilla [1967], existence values
have been given little attention within conventional models of consumer be-
havior.* The recent experimental findings of Schulze et at. [1981], discussed
earlier in this chapter, have changed this perspective. Their estimates of
preservation values for the Grand Canyon's visibility conditions indicate that
the nonuser values for this unique natural environment are likely to be sever-
al times the magnitude of the user-associated benefits. While it is not unam-
biguously clear, preservation values can be expected to include option value,
existence value, and, perhaps, bequest values. Each of these motivations for
desiring the services of a unique natural environment was identified by Kru-
tilla as values that would not necessarily be reflected in the private market
transactions for the services of such resources.
As a result of these empirical findings, the attention given to modeling
and measuring existence values has increased. Freeman's [1981] recent notes
on the problems associated with defining and measuring existence values indi-
cate at least two interpretations of an individual's reasons for valuing the
existence of a resource. In the first note, Freeman designates a stewardship
value (or motive), where the level of use of a resource affects the value de-
nved. In this case, one's existence value would be reduced if the resource
,r°ta^'e e*ce.p*ion is Mi!ler and Menz's [1979] model for describing
allocation deasions involving wildlife preservation. These authors
introduce species stock terms into individuals' utility functions as a source of
value without requ.ring that these values arise from consumptive uses ^ How-
' ldentjfy the rati°nale ""*•"• Deification
5-31
-------
were not properly managed. Freeman's second proposed reason for existence
value stems from a form of vicarious consumption. An individual derives bene-
fit from the knowledge that other individuals can use a resource.
Freeman's analysis does not develop either of these frameworks in detail.
They were suggested only as prospective explanations for values due to the
existence of a resource and can be- interpreted as defining different forms of
consumption. Thus, they do not provide direct insight into how existence
values might be measured. However, Freeman does suggest that attempts to
measure existence value should carefully identify the likelihood of future use
of the site and elicit an individual's user and nonuser values. In effect, he
proposes that questions call for the sum of option price and existence value.
The design of the existence value questions for this survey attempted to
use these insights. The sources of site valuation (on the value card used in
the interviews) were separated into direct use, potential use, and existence
motives. After reviewing these motivations, the interviewer asked each re-
spondent how much he would be willing to pay to prevent the deterioration of
water quality from boatable conditions to an unusable state even though he
never would plan to use the river. Responses to these questions were re-
garded as tentative estimates of existence values. The situation is a difficult
one for the respondent to conceptualize. Water quality is to remain at a boat-
able level, but the individual nonetheless will not use the river.
Table 5-9 presents these results for users and nonusers with the sample
restricted to exclude protest bids and observations judged to be inconsistent
with the contingent valuation framework. Both estimates are significantly dif-
ferent from zero. Users do exhibit significantly different estimated existence
values from nonusers at the 5-percent level. These values are quite compar-
able to the estimates for the option price (aggregated over question mode), as
reported in Table 4-9 for avoiding the loss of use of the river. Indeed, there
is not a significant difference between the means for either users or nonusers.
This finding, together with the fact that many respondents repeated their
option price bids for the existence value question, suggests that these results
should be interpreted with caution. Until the theoretical issues associated with
describing the relationship between user and existence values is resolved, it
cannot be concluded that these estimates represent independent sources of
value for a water quality improvement.
Table 5-9. Estimated Existence Values
Mean (X)
Standard deviation(s)
n
User
65.985
92.824
66
Nonuser
42.115
64.023
139
5-32
-------
5.9 SUMMARY
This chapter has reviewed the theory underlying the definition of option
value, summarized the results of past efforts to measure option and other non-
user values, and presented the results of the Monongahela River survey that
relate to nonuser values.
The findings provide clear support for a positive, statistically significant,
and substantial option value for water quality improvements for the Monongahela
River. The estimated option values for loss of the use of the area in its cur-r
rent condition (i.e., providing boating recreation activities) range from ap-
proximately $21 to $58 for users (and $14 to $53 for nonusers). The option
price for users ranges from approximately $27 to $95. Thus, option value is
a substantial fraction of the option price of users and generally exceeds their
use values for a change in water quality. The Monongahela River is not a
unique recreation site. Thus, these estimates may well require reconsideration
of the conventional assumption that option value is small in comparison to use
value for natural environments without unique attributes. Of course, it should
also be acknowledged that the available estimates of option value are quite lim-
ited. Most can be criticized for problems in the research design, including
possible flaws in the survey. The design of the Monongahela River study
places heavy reliance on the use of a schematic classification of the sources of
an individual's valuation of the river (i.e., the value card) in eliciting a
division of user and nonuse benefits. Because this is the first application of
this device, it was not possible to evaluate its effectiveness.
Users appear to have a somewhat lower option value than nonusers for
most levels of change in water quality. For the most part, the respondents'
socioeconomic characteristics were not useful in explaining the variation in esti-
mated option values.
The limited analysis of the role of supply uncertainty for measures of
option value clearly suggests it is an important influence on users' option price
(and therefore on the derived option value). Assurance of supply is quite
important to our positive estimates for option value.
Finally, this survey provided the ability to estimate existence values.
While the findings suggest that these values are positive and statistically sig-
nificant, prudence requires they be interpreted cautiously. It is not clear that
respondents understood the distinction sought. Many bid the same amounts as
their earlier option prices for a comparable change in water quality.
5-33
-------
CHAPTER 6
CONTINGENT RANKING DESIGN AND RESULTS:
OPTION PRICES*
6.1 INTRODUCTION
The purpose of this chapter is to report a set of water quality benefit
estimates based on an analysis of the Monongahela survey respondents' rank-
ings of four hypothetical combinations of water quality levels and amounts paid
in the form of higher taxes and prices. The use of data including individ-
uals' rankings of goods or services described in terms of the features of each
of a set of possible alternatives together with an extension of the McFadden
[1974] random utility model was first proposed by Beggs, Cardell, and Haus-
man [1981] as a method for measuring the potential demand for new goods.
Rae [1981a, 1981 b] has subsequently used this approach as an alternative
means of estimating individuals' valuation of air quality improvements. The
implicit assumption of the contingent ranking approach is that individuals are
more likely to be capable of ordering hypothetical combinations of environ-
mental amenities and fees than to directly reveal their willingness to pay for
any specific change in these amenities. Unfortunately, past studies have
tended to adopt only one or the other of these two approaches, and there has
been little basts for comparing their respective estimates. As a result, the
survey instrument for the Monongahela study was designed explicitly to include
the use of contingent ranking as a method for measuring individuals' valuation
of water quality improvements. All survey respondents were asked to rank
four hypothetical combinations of water quality and payments to permit a com-
parison of contingent valuation and contingent ranking methods within the
context of a common application.
To understand the economic basis for modeling consumer behavior using
contingent rankings, the random utility model—widely applied to model con-
sumer behavior that involves discrete choices—must first be considered. Sec-
tion 6.2 provides some of this background by describing the features of the
random utility model, and Section 6.3 discusses two possible methods for im-
plementing the model. The first, an adaptation of the conditional logit model,
can be derived under the assumption that the errors associated with the ran-
dom utility function are additive and follow an extreme value distribution
(i.e., the Weibull distribution). The second, a normal counterpart to the
*Special acknowledgment is due Donald Waldman of the Department of Eco-
nomics, University of North Carolina at Chapel Hill, who helped develop the
maximum likelihood program for ordered logit analysis and provided a general
program for estimating the Keener-Waldman ordered normal estimates. He also
assisted in the estimation and discussed several aspects of these models with
the authors.
6-1
-------
ordered logit, was recently developed by Keener and Waldman [1981], who
used numerical procedures to approximate the likelihood function associated
with a random utility function having additive normal errors. With this back-
ground, Section 6.4 summarizes the results of Rae's survey applications of
the contingent ranking approach to benefit estimation for visibility change;
Section 6.5 discusses the question used for contingent ranking and the empir-
ical estimates of random utility models; and Section 6.6 considers some of the
theoretical issues associated with Rae's proposed approach for benefit estima-
tion with the model and reports the results derived by applying it directly
with the Monongahela survey data. Finally, Section 6.7 summarizes the chap-
ter and proposes an alternative application of the random utility model.
6.2 CONSUMER BEHAVIOR AND THE CONTINGENT RANKING FRAMEWORK
The conventional economic description of consumer behavior generally
maintains that each individual consumes some amount of every good or service
that enters his utility function. The objective of these models is to describe
the choices individuals make for marginal increments to their consumption
levels. That is, individuals are usually portrayed as adding to previous con-
sumption of goods or services from which they derive utility.* Of course,
many consumer choices involve major purchases. In the purchase of an auto-
mobile or a house, the selection of an occupation, or the choice of an appli-
ance, the consumer's decisions all require discrete choices. In these cases,
the commodity often is durable and provides a stream of services over some
time period or involves some commitment of the individual's time. Thus, the
assumption of continuous incremental adjustment in the levels of consumption
of each good or service that is implied in the conventional model of consumer
behavior is not plausible for describing individuals' choices when they involve
discrete selections.
Several types of modifications to conventional models have been proposed
to make them more amenable to explaining such discrete choice problems. One
involves an extension of the time horizon in the conventional model of consumer
behavior. For example, on any particular- day a commuter will select a travel
mode to reach his job. Viewed on a daily basis, modal choice is discrete since
fractions of the available travel modes cannot, as a rule, be consumed in a
single trip to the workplace. However, over the course of a month or a year,
the individual may well select a varied menu of transport modes. Thus, with
this adaptation of lengthening the time horizon, the conventional model of
consumer behavior may be more relevant to explaining these decisions.
A second proposed adaptation for dealing with discrete choices involves
modeling consumer decisions as service flows rather than as the choice of any
particular asset. For example, an individual purchases an auto for transporta-
tion services. These service decisions may be more amenable, under this
interpretation of conventional theory, to modeling than the discrete choices of
*Conventional models of consumer behavior assume positive levels of con-
sumption of all goods and services to avoid dealing with corner solutions.
6-2
-------
durable goods themselves. As a practical matter, however, most of the modi-
fications to the conventional theory have enjoyed limited success. Information
on the consumption rates for the services of durables is virtually nonexistent.
Forecasts of the rates of use of travel modes based on aggregate information
over long time spans cannot take account of the specific constraints facing
individuals in making these decisions and, as a result, may be inadequate for
many problems.
The random utility model has been proposed as one approach for dealing
with discrete consumer choices. It generally replaces the assumption of a com-
mon behavioral objective function across individuals with the assumption of a
distribution of objective functions. Attention is shifted from the intensive
choice margin and the associated incremental analysis to individual decision-
making at an extensive margin with discrete selections. As a result, random
utility models are often quite simple in their description of the choice proc-
ess. Individuals are assumed to have utility functions affected by (1) the
objects of choice and their features and (2) the characteristics of the individ-
uals making the decisions. The analyst is assumed to be capable of observing
the distribution of individuals and their respective choices but does so with-
out complete information. Thus, the observed behavior is assumed to be de-
scribed as a trial—the drawing of one individual from a population; the re-
cording of his attributes, the alternatives available, and their features; and
the making of a choice. Because there is a distribution of individuals, the
model describes the choice process using a conditional probability. Each alter-
native has some probability of being selected based on its characteristics, the
other alternatives available and their features, and the attributes of the indi-
vidual selected. Behavior is described by modeling these probabilities.
The random utility function provides the vehicle for modeling these condi-
tional probabilities. In a random utility framework, the individual is assumed
to select alternatives that provide the highest utility level. Thus, if Equation
(6.1) describes a random utility function, then individual j's probability of
selecting alternative k, given j's attributes, z., and in the presence of the
set of alternatives defined by A, is defined byjthe probability that j's utility
of k will exceed the utility of all other alternatives, as given in Equation (6.2)
below:
U(a, z) = V(a, z) + e(a, z) , (6.1)
where
U(a, z) = utility provided by an alternative's vector of characteristics,
a;
z = attributes of the individual;
V(a, z) = nonstochastic component of utility, describing what consti-
tutes representative tastes in the population; and
e(a, z) = stochastic effect reflecting the nondeterministic effects of
taste on decisionmaking for an individual with attributes, z,
facing an alternative with characteristics, a.
t
6-3
-------
Prob [ak 2j, A] = Prob [UR > Uj for all i f k] =
Prob [V(a. , z.) - V(a-, z.) > e(a;, z.) - e(a. , z.), for all i i- k]
k j i j i j K J
By making distributional assumptions to characterize the e's, the probability
statement in Equation (6.2) can be defined in terms of the characteristics of
the alternatives and the features of the individual. For example, assuming
that the e's are independently, identically distributed with the Weibull distri-
bution* allows the probability to be expressed as a logistic, as in Equation
(6.3):
exp(Vk) (6 3)
Prob[Uk > U; for i t k] = exp(vk) + exp (V ;)
Before the relationship of random utility functions to contingent ranking
is explained, several observations on the nature of these functions should be
noted. The description in Equation (6.1) is a conventional treatment (see
McFadden [1974] or [1981]) that is completely general. In this general de-
scription there is no explicit treatment of the constraints to choice, such as
an individual's income or market prices. To make these constraints clearer,
it is completely consistent with the random utility model to view V(-) as the
result of a constrained optimization process. Within such a framework, V(-)
would be an indirect utility function, reflecting an individual's attributes, the
characteristics of the choice alternatives (to the extent they are not reflected
in market prices), the individual's income, and the prices of the alternatives
available on organized markets.!
Thus, a random utility function framework does not imply that the con-
ventional economic view of the consumer behavior be ignored. Indeed, as
McFadden [1981] has suggested, V(-) can be regarded as an indirect utility
function, even in applications where it has been specified as linear in its
parameters. This interpretation is possible because any continuous function
can be approximated to any desired degree of accuracy with a linear specifi-
cation. The requirement that V(*) be homogeneous of degree zero in income
and prices can be met by requiring that the variables in the linear approxima-
tion (in parameters) be homogeneous of degree zero. (This requirement is
necessary for consumers to be free from "money illusion" and to respond only
to changes in relative prices and income.)
*The distribution function for the Weibull distribution is:
Prob(Z < t) = exp(exp (-(t-
The ordered logit is derived for a standardized form with a = 0 and 6 = 1.
This implies that variance of the errors will be 1.6449. See Chapter 20 of
Johnson and Kotz [1970] for more details.
tThis description admits the possibility of a model comparable to the he-
donic framework used in modeling property values (see Rosen [1974]) or,
more recently, adapted to a travel -cost recreation demand framework by
Brown and Mendelsohn [1980]
6-4
-------
Alternatively, it is possible to assume that the indirect utility function is
separable in all commodity prices but the ones of direct interest. Moreover, in
principle, these prices can be replaced by a price index that can be assumed
to normalize the incomes and the prices of goods and services of interest.
However, it should also be acknowledged that this approach imposes quite
restrictive assumptions on the structure of individual preferences.* The pri-
mary conclusion to be drawn from these general observations is that conven-
tional neoclassical models of consumer behavior can be used as an integral part
of random utility models when the utility functions are interpreted as indirect
functions describing the outcomes of households' optimizing decisions.
A second feature of the models used in the random utility framework stems
from the assumption of the independence of irrelevant alternatives. This as-
sumption is important to the structure of any model in the framework because
it implies that the odds of one alternative being chosen over a second alterna-
tive are not affected by any other alternatives. McFadden [1974] has conven-
iently summarized the implications of this assumption in discussing the limita-
.tions to the random utility model:
The primary limitation of the model is that the independence of ir-
relevant alternatives axiom is implausible for alternative sets contain-
ing choices that are close substitutes. . . . application of the model
should be limited to situations where the alternatives can plausibly
be assumed to be distinct and weighed independently in the eyes of
each decisionmaker. (McFadden [1974], p. 113)
With this background on the random utility model and its relationship to
the conventional model of consumer behavior, it is possible to consider the
contingent ranking methodology. The contingent ranking methodology main-
tains that individuals' valuation of environmental amenities, such as visibility
or improved water quality, can be described within a random utility framework.
Thus, an approach to estimating individuals' values for changes in these amen-
ities could be developed by estimating the deterministic component of the ran-
dom utility function--i.e., the V(-) in Equation (6.1). The process of collect-
ing the information necessary to derive these estimates involves presenting
individuals with a set of alternatives. Each alternative describes a specific
state of the world in that it characterizes the features of the environmental
resource and the cost to the individual of having access to the resource under
the specified conditions. Individuals are then asked to order the alternatives
from most to least preferred. If the determinants of V(-) are known and it
can be approximated using models that are linear in parameters, the ranking
of the alternatives provides sufficient information to estimate (relative to a
scale factor) the parameters of these models.
Applications of these principles have been used by Hausman and Wise
[1978]. The restrictive assumptions required are discussed in detail by Black-
orby, Primont, and Russell [1978]. Based on their analysis (especially in
Chapter 5), this approach—used by Hausman and Wise, for example—requires
separability in commodity prices (called indirect separability by Blackorby,
Primont, and Russell) and additive price aggregation. These assumptions imply
that the utility function will exhibit homothetic separability.
6-5
-------
The contingent ranking methodology provides an operational basis for
benefit measurement. However, several factors should be considered in using
this methodology to estimate benefits of environmental amenities. Consistent
benefit measurement requires recognition of the constraints on individual
choice. Thus, to define compensating variation or compensating surplus bene-
fit measures, V(O must be considered an indirect utility function. Moreover,
when individuals are asked to rank alternatives that involve levels of an en-
vironmental amenity and a fee, the role of the fee must be considered within
an optimizing model of consumer behavior. That is, the selection of the pay-
ment vehicle may have an important effect on the specification of the random
utility function. For example, if the fee included in each alternative is a user
charge associated with gaining access to the resource whose features are also
being described, the fee would be treated as a price per unit of use of the
resource. Therefore, it would enter the indirect utility function in a format
comparable to any other price. By contrast, if the fee is described as an
annual payment, regardless of how much the resource is used, it would be
expected to enter as an adjustment of income rather than as a price per unit
of use of the resource. The indirect utility function can be expected to be
homogeneous of degree zero in income and prices. While assumptions that can
simplify the form of the function and the number of distinct prices need to be
considered, they impose significant restrictions on the types of features of
demand relationships between the commodities consumed by the individual.
These issues are discussed in more detail below.
The required assumption of independence of irrelevant alternatives limits
the generality of the contingent ranking methodology for benefit estimation.
The definition of the alternatives presented to individuals in a contingent
ranking is largely arbitrary and is constructed to ensure a distinct ranking
of the combinations presented. Indeed, the literature to date has not explic-
itly considered the issues associated with experimental design in selecting the
alternatives used. While this problem does not arise in application of the
model to alternatives defined by what is available in the real world, it may
well be an important consideration when the alternatives are specified to repre-
sent feasible alternatives or defined to provide the "best" estimates of an in-
dividual's compensating surplus for a change in an environmental amenity.
The framework used for benefit estimation (and described later in this
chapter) implies that the level of environmental quality and proposed fee are
subject to continuous tradeoffs as each varies over predefined ranges. This
presumption is quite different from those cases for which McFadden [1981]
argued the random utility function is best suited. Thus, even a brief consid-
eration of the economic theory and assumptions underlying conventional formu-
lations of the random utility model indicates there may be problems with its
use in the contingent ranking methodology as a procedure for benefit estima-
tion. Equally important, economic theory offers some guidance in selecting
the most appropriate specification in empirical applications of the model.
6-6
-------
6.3 ESTIMATION OF RANDOM UTILITY MODELS WITH
ORDERED ALTERNATIVES
The random utility model can be estimated using the information provided
in contingent rankings with a maximum likelihood estimator. That is, once
the additive error associated with each individual's utility function is assumed
to follow a probability distribution, the decision rule given in Equation (6.2)
describing how each individual orders the available alternatives provides the
information necessary to describe the probability of a specific ordering of
alternatives. Of course, for some assumptions concerning the probability dis-
tribution for e(«)/ the form is simpler than it is for others. Nonetheless, in
principle, any assumed probability distribution provides the basis for de-
scribing this probability, which is the basic ingredient in the definition of the
likelihood function (i.e., the joint probability of observing all the orderings
given in a specific sample as a function of the parameters of the utility func-
tion). The criteria of maximum likelihood estimation can then be used to de-
rive estimates of the parameters (relative to a scale factor) of the determinis-
tic portion of the utility function.
The discussions to this point as well as the existing applications of the
contingent ranking methodology have assumed, for analytical convenience, that
the errors follow a Weibull distribution in deriving an ordered logit estimator
for the parameters of the function specified to represent (or to approximate)
V(-). Because the logic underlying this derivation has been outlined in
Beggs, Cardell, and Hausman [1981], some features of the estimator are simply
highlighted here as they relate to the logit model applied to problems involving
discrete choices (as given in Equation (6.3)) versus those based on an order-
ing of several alternatives.
A closed form expression for the probability of an ordering of the alter-
natives can be derived using the properties of the Weibull distribution. More
specifically, the conditional probability Prob(U. < t U. > U. , for j t k) differs
J ~ f *
only in its location parameter from the unconditional distribution, as illus-
trated for this two-alternative case in Equations (6.4a) and (6.4b):
Prob(U. < t) = exp (-exp(-(U.-V.))), unconditional distribution. (6.4a)
Vi Vk
Prob(U. < t U. > U. for j t k) = exp (-exp(-(U-log (e J+ e )))). (6.4b)
J J K
Beggs, Cardell, and Hausman [1981] have outlined how this result can be
used to derive the probability of an ordering of alternatives as given in Equa-
tion (6.5):
V,
U2 > U3 >...> Uh) =
e
k=1
i=k
where
H = the number of alternatives.
6-7
H V.
Z e
(6.5)
-------
Equation (6.5) describes for any individual the probability of an observed
ordering of alternatives. Under the assumption that each individual's deci-
sions on ordering the alternatives are independent of all others, the likelihood
function can be defined for a sample of T individuals as:
L =
T
n
H
n
k=1
v.,
e
H
I e
i=k
VJi
(6.6)
By specifying the determinants of V-k, the likelihood function, L, can be ex-
pressed in terms of unknown, estimatable parameters. Thus, for example, if
V.. is described by Equation (6.7), the likelihood function can, for a given
J*^
sample, be expressed in terms of the unobservable parameters, 0:*
Vjk = Zik
(6.7)
where
Zik =
vector (1*K) describing the individual's characteristics, attributes
of alternatives being ranked, and other variables as detailed by
economic model used to describe behavioral choice
P = vector Kxl of parameters to be estimated.
Substituting Equation (6.7) into Equation (6.6) and taking the logarithm yields
the log-likelihood function for the ordered logit estimator.! Maximum likeli-
hood estimation involves solving this function for the value of p, which maxi-
mizes the log-likelihood function. In most cases, this solution involves numer-
ical optimization procedures. Our analysis of the logit estimator used the
Davidon, Fletcher, and Powell [1963] (DFP) algorithm with numerical partial
derivations.
The second estimator for use with information from contingent ranking
was developed by Keener and Waldman [1981] and follows the same behavioral
model. In the Keener and Waldman framework, the errors associated with the
*See Section 6.3, above, for a description of the relationship between a
general form for the Weibull and the standardized form that underlies the
Beggs, Cardell, and Hausman [1981] derivations.
tThis estimator is actually the same method proposed by Cox [1972] for
dealing with duration problems. That is, Cox proposed a conditional likeli-
hood model based on ordering the variable of interest. His framework main-
tains a proportional hazard formulation of the problem. The two likelihood
functions will be identical in the absence of ties (i.e., Cox's analysis allows
for ties in the ordering of the dependent variable, while the ranked logit
does not).
6-8
-------
random utility function were assumed to follow independent normal distrib-
utions. The probability of an ordering of alternatives is described by
the multivariate, normal cumulative distribution function evaluated at
Zr(£+1)^"Zr(£)^' A = 1' 2'---' H"1' where r(£) is the index of the component
of the vector of utilities for a given individual with rank I. In general, the
solution to the likelihood function for the normal distribution would pose a dif-
ficult numerical integration problem. However, Keener and Waldman observe
that the error covariance matrix is tridiagonal and propose a computationally
tractable method of numerically evaluating the probabilities composing the like-
lihood function. Thus, the likelihood function for the ranked normal estima-
tor is derived by numerically integrating these functions to obtain the prob-
abilities of the orderings provided by each sample respondent. Numerical max-
imization of this function yields the Keener-Waldman estimates. The DFP algo-
rithm was also used to maximize the likelihood function associated with this
estimator. Because ranked logit is globably concave, most experience with the
method indicates it converges rapidly. Thus, estimation with the ordered logit
framework is comparatively inexpensive. By contrast, as the above description
implies, the maximum likelihood estimator based on the assumption of normality
can be an expensive approach. Consequently, the ranked logit method has
been used here to examine a wide array of alternative specifications for the
deterministic component of the random utility function and the ranked normal
for the subset of those models that were judged to be the "best."
6.4 PAST APPLICATIONS OF CONTINGENT RANKING
The use of contingent ranking procedures for benefit estimation with
environmental amenities has been a recent development. The applications
have been exclusively conducted by Douglas Rae of Charles River Associates
and have focused on valuing visibility changes. Our review considers two
unpublished reports (Rae [I98la, 1981b]) describing applications of the meth-
odology.* Because the studies were largely motivated by concern over the
benefits associated with defining alternative visibility standards for Class I
areas (as mandated under the 1977 Amendments to the Clean Air Act), the
surveys have been conducted at fairly unique recreational areas—the Mesa
Verde National Park and the Great Smoky National Park.
The experimental design used in the two surveys was quite similar. In
each case, a sample of users of a park' was asked to rank a set of alterna-
tives. The set was composed of two types of alternatives. One type speci-
fied combinations of conditions for the park where the survey was being con-
ducted. These conditions included different visibility conditions (using photo-
graphs to display an integral vista within the park), a recreational quality
measure (generally measured by waiting time at a key landmark or availability
of activities at a park service center), and a per vehicle entry fee. The sec-
ond type of alternative included other sites. The reports are not clear as to
*Since the draft version of this report was prepared, a third application
(Rae [1982]) to visibility changes in Cincinnati has been undertaken, but is
not considered in this review. Future references will use the author's name
[Rae] and will refer to these 1981 reports.
6-9
-------
whether comparable attributes were reported on the cards used to describe
these other sites or whether the evaluation of the characteristics of these
sites was left to the respondents. Table 6-1 describes the features and se-
lected results for each of these studies.
Each respondent was asked to provide two rankings. The information
detailed in Table 6-1 is based on the rankings for deterministic conditions.
That is, in the cases shown in Table 6-1, the alternatives were explained as
having constant visibility at the level prescribed. In addition to these rank-
ings, individuals in each study were asked about alternatives that included
deterministic and probabilistic descriptions of visibility conditions (i.e., three
probabilistic cases and four with constant visibility prescribed). The prob-
abilistic cases specified the percentage of summer daylight hours when one of
four conditions could be expected to prevail. Unfortunately, no attempt was
made to take account of the different probability structures used in describ-
ing the visibility conditions in the estimation of the random utility functions
from these rankings.
As Table 6-1 indicates, the empirical results from these studies are
mixed. The entry fee was found to be a significant determinant of the rank-
ing of alternatives in both studies. However, the qualitative variables for
visibility conditions were not significant determinants of utility. The Great
Smoky results were somewhat more definitive. They indicated that serious
impairments in visibility had a negative and significant impact on the level of
utility. However, at lower levels of impairment the results for some specifi-
cations of the model contradict £ priori expectations.
These studies are important because they demonstrate an alternative ap-
proach for soliciting individuals' preferences and organizing them to test hypo-
theses. Nonetheless, they are subject to some shortcomings.
The most important problem arises with the specification and interpreta-
tion of the random utility function estimated in these analyses. As a rule,
the model specifications used in Rae's analyses of the respondents' rankings
include income, the suggested price for use of the area (i.e., the fee included
as an attribute of each alternative that is ranked), and one or more measures
of the postulated visibility conditions. It is thus clear from context, though
never explicit in the studies, that the functions are to be interpreted as in-
direct utility functions. As a rule, an indirect utility function would in-
clude the prices of all the goods and services consumed by the individual,
not simply the fee proposed for use of the relevant recreation site. Since
these prices have been omitted from the models, it must be concluded that an
implicit assumption consistent with one of the appropriate forms of aggregation
has been made. There are two possibilities—that all remaining goods can be
treated as a Hicksian composite commodity (see Deaton and Muellbauer [1980]
pp. 120-122 for discussion) or that the utility function exhibits homothetic
separability in two groups of commodities. The first group of commodities con-
sists of the services of the site under evaluation and the second includes all
other goods and services.
6-10
-------
Table 6-1. Summary of Rae/CRA Contingent Ranking Studies
o>
Sample
Study size
Mesa Verde 205
Great 213
Smoky
Description
of
environmental
amenity
Visibility conditions:
Intense plume
Intense haze
moderate haze
clear
Visibility conditions:
intense haze
moderate haze
slight haze
clear
Character
of fee
Entry fee per
vehicle, $2 to
$20 (existing
fee, $2)
Entry fee per
vehicle, $0 to
$30 (existing
fee, $0)
Number
of
Recreational alter-
quality natives
Congestion as 13
measured by
waiting time
at landmark
on site
Availability of 14
full program
of visitor
center
Area
specific
alter-
natives Design choice
8 22 possible combinations
of alternatives; 1 of
10 sets of 8 cards
randomly given to
survey respondents;
combinations of
alternatives always
Include current
conditions; no
clearly dominant
alternative Included
In combinations
8 29 possible combina-
of alternatives;
1 of 10 sets of 8
cards randomly given
to survey respondents;
alternatives always
include current
conditions; no
clearly dominant
alternative included
In combinations
Empirical findings'
Entry fee, negative
and significant;
qualitative variables
for poor visibility.
negative and insig-
nificant; absence of
congestion, positive
and significant
Entry fee, negative
and significant;
qualitative vari-
ables for visibil-
ity provide some
evidence for valu-
ation of better
visibility; Intense
haze, negative and
significant; absence
of program not
important
Benefit .
estimates
(1981 dollars)
Intense haze
to clear,
$0.73 to $0.79
intense plume
to clear,
$1.03 to $1.13
Intense haze
to clear,
$7.39 to $11.22
intense haze
to slight
haze, $11.03
to $14.86
'These results are based on aggregate models and use conventional criteria for significance at the 5 percent level with asymptotic t-statistlcs.
Based on the aggregate model.
-------
Under the first aggregation assumption, the prices of all goods and serv-
ices (other than the site under study) are assumed to change in constant pro-
portion, and this proportion, say k, would be the relevant argument in the
indirect utility function. In this case, because of the nature of the assumed
pattern of price movements, an individual's preference for one good in the set
cannot be distinguished from his preference for any other. Ideally, to define
and estimate an indirect utility function consistent with theory requires a
sample consistent both with the assumption of proportionality in the price
movements of all goods and with some variation in the proportionality con-
stant, k. Since both of these conditions are not often realized in practice,
the Hicksian composite commodity theorem is difficult to use in empirical appli-
cations.* For the Rae analyses, there is no way. either to know whether the
prices of all other goods and services change in a proportional relationship
across all individual respondents or to measure the magnitude of these propor-
tionality constants. These unknowns are important because proceeding under
the assumption that a Hicksian composite can be defined and then arbitrarily
assuming a constant value for it across all individuals in a cross-sectional data
base is equivalent to assuming that there is no change in prices across indi-
viduals. If the respondents all come from a single geographic area (i.e., in
a region immediately around the site), this assumption may be reasonable.
However, based on evidence of substantial regional variation in prices, this
implicit assumption is untenable for sites that draw visitors from around the
nation. Moreover, to the extent the price variation is not simply by a con-
stant multiple for all goods and services, the assumptions of the composite
commodity approach to aggregation would be violated.t
*lt can be used in controlled experiments where the prices confronting
an economic agent (i.e., household or firm) are selected by the analyst. For
the most part it has been an analytical device used in theoretical analysis.
Indeed, Deaton and Muellbauer [1980] raise comparable reservations, noting
that:
The usefulness of this theorem [i.e., the Hicksian composite com-
modity theorem] in constructing commodity groupings for empirical
analysis is likely to be somewhat limited. ... in an open economy
with a floating exchange rate, considerable fluctuation in relative
prices can be expected and even without this, it is not clear that
we could justify the types of aggregates that are usually available.
(pp. 121-122).
They do, however, note that greater justification is available for use of the
theorem with single period aggregation.
tThe Bureau of Labor Statistics (BLS) data used to derive regional cost
of living indexes provide evidence of both variation in the levels of prices by
region and differential patterns of change among these prices for different
goods and services.
6-12
-------
The second approach to structuring an indirect utility function so that it
approximates the models in Rae's analyses v/ould involve assuming that the di-
rect utility function exhibits weak separability. That is, a given general util-
ity function IKXj, X2, . . ., Xn), with X. the ith vector of goods and serv-
ices, can be written as UCU^X^, U2(X2), . . ., Un(Xn))), with each of the
subfunctions (U.(*)) a homothetic function. This specification implies that
the indirect utility function can be expressed in terms of the price (or fee)
for the site's services, income* and a price index for all other goods and
services. This price index can be normalized at unity for a given set of
values for the prices of all other goods and services. However, if it is
assumed to be unity for all respondents, it is implicitly assumed that all
respondents face the same prices (or different price sets that always lead to
a unitary value for the index). As in the case of the composite commodity
aggregate, the plausibility of this assumption--!. e. , holding the price index
at a constant value for all individuals—depends upon whether or not respond-
ents come from a small region surrounding the site. Otherwise, some varia-
tion can be expected, both in price and in the value of the price index.
Aside from this issue, the use of the homothetic separability assumption
also restricts the nature of the income effects_ for goods within each group-
ing--!. e., subfunction as given earlier as U.(X.)--and the nature of the sub-
stitution effects for commodities involved in different groupings. To illustrate
the nature of these constraints, consider the case of Rae's applications where
the utility function is assumed to be composed of two groups of commodities —
the services of the site under study and the set of all other goods and serv-
ices. It is convenient to use the framework of conditional demand functions
to illustrate the demand effects of the separability assumptions.* For example,
the income elasticity of demand for any commodity in the set of goods and
services (other than the site) can be defined as a product of the income elas-
ticity of demand in the conditional demand functiont and the elasticity of the
expenditures on this set of goods with respect to income. More formally, let
q. designate the quantity demanded for the ith commodity in this set; e, the
expenditures on all commodities in the set; and y, the individual income.
Thus, if q is the use of the relevant site's services and p£ is the price per
unit of use,5
e = y - Ps • qs - (6.8)
*For a discussion of conditional demand functions, see Pollak [1969,
1971]. Summaries of his work are available in Deaton and Muellbauer [1980].
tThis elasticity is the percentage change in the quantity demanded of the
good with respect to a percentage change in the expenditures on all goods in
the set. These expenditures play the same role in conditional demand func-
tions as income would in a conventional demand function. In general, the
determination of these expenditure levels will be a function of the level of
income and the prices of all goods and services. See Blackorby, Primont,
and Russell [1978] and Pollak [1971] for further discussion.
6-13
-------
The conditional demand function for q. will be related to the prices, of all
goods in its group, P., and the expenditures on this group (i.e., q} = fj(Pj, e))
will be responsive to income and the prices of all goods. This association can
be used to derive the following relationship between demand responses:
3q. 9f. fta
_l = _i . ^ (69}
8y 9e 3y ' ^ 3J
In elasticity terms, Equation (6.9) can be written as:
Homotheticity of this subfunction that reflects decisions about all other goods
implies that the income elasticity in the conditional demand functions for these
goods will be unity. Thus, the first term on the right side of Equation (6.10)
will be one. Thus, Rae's model implicitly maintains that all goods consumed by
the individual (aside from site services) have equal income elasticities and are
equal to the expenditure elasticity with respect to income.
This analysis can be extended one step further. Budget exhaustion im-
plies that the share weighted sum of the income elasticities will be unity, as
in Equation (6.11)
Ks ' esy + j, Ki eiy = 1 ' <
where
K = share of income spent on the site's services
5
K. = share of income spent on the ith commodity
e = income elasticity of demand for a site's services
e. = income elasticity of demand for the ith commodity.
Using Equation (6.10) to substitute for e. in Equation (6.11) gives
Ks esy + eey Ki = n ' (
where
e
ey e 8y
6-14
-------
While homothetic separability of the utility function does not in general re-
strict e , it does have implications for the cases in which it would be likely
to be plausible. Rearranging the terms in Equation (6.12) gives
n
' Vy5
Since the grouping implicitly required for Rae's model involves all other goods
in the set designated as q., 1=1,2, . . ., n, it Is reasonable to expect that
eey would be close to un>W- That is, expenditures on the majority of the
items in the individual's budget are likely to change in percentage terms as
income does. This implies that the income elasticity of demand for site services
will also have to be close to unity to satisfy the adding-up condition on income
elasticities (i.e., Equation (6.11)). Equation (6.13) illustrates this conclusion.
Of course, this conclusion is a judgment. Indeed, the appraisal of the
plausibility of using the composite commodity to explain Rae's models was also
based on a judgment. What is at issue is an evaluation of the implicit assump-
tions of a model's specification for the properties of its results (or conclu-
sions). The forgoing appraisal suggests that the assumptions necessary to
interpret Rae's model as an indirect utility function are fairly stringent.
Both sites attract visitors from substantial distances. Thus, omitting the rele-
vant price aggregate for other goods may be an important consideration for
the properties of the estimates of compensating surplus derived from Rae's
indirect utility functions.
Regardless of how one judges the plausibility of the assumptions required
to ignore other goods and services, there is a further issue arising from Rae's
definition of the compensating variation. To illustrate the problem, consider
an example. Assume that Equation (6.14) defines the deterministic component
(V) of the random utility model, which is assumed to be a function of the
individual's income (Y), the entry fee (F), and the specified level of visibil-
ity (v):
V = ctjY + a2F + ct3v . v6.14)
Rae's proposed benefit measure is the increment to fee that must accompany a
change in visibility to hold utility constant. When Rae assumes that dY = 0,
this increment is given for the example by Equation (6.15):*
"Assuming dY = 0, this is derived by totally differentiating Equation
(6.14) as: dv =
Holding utility constant in expected value, dV = 0, or
a2dF + a3dv = 0 .
Solving for dF gives:
dF = - =» dv .
a2
6-15
-------
(6.15)
Equation (6.15) is not compensating variation. This Hicksian measure of con-
sumer surplus is defined (see page 2-4) to be the income change required to
hold utility constant in the presence of a change in the quantity of a good or
service, such as visibility.
Thus, the interpretation of these benefit measures depends upon the type
of fee. If it is a fee per unit of use, Equation 6.15, strictly speaking, does
not measure compensating variation. Of course, the extent of error depends
upon the level of repeated use. If, for example, users are expected to visit
the site only once, Rae's measure should not be appreciably different from one
based on the income changes. However, if there are repeat visitors, it may
be a source of error in the benefit estimates. In pragmatic terms, as shown
below, the use of price versus income for measuring the benefits associated
with a specified change in water quality markedly affected the results. More-
over, in the present study, the fee was described as an annual payment rather
than a price per unit of use.*
There are several additional problems with these studies. The Rae ap-
plications fail to include respondents' characteristics in the estimated utility
functions. Presumably, this approach was adopted because two models were
estimated. The first was specified under the assumption of constant param-
eters across all respondents (the "aggregate" form). The second permitted
these parameters to be different for each individual. Thus, this second for-
mat provides the flexibility of permitting all individuals to be different in their
determinants of utility. However, to estimate a model with this flexibility, a
reasonably large number of ranked alternatives is required. It is not clear
that this general framework is helpful to interpreting the results. Detailed
analysis of the parameter estimates across different groups of individuals
would be necessary to understand the importance of an individual's attributes
in determining his preferences for water quality.
Despite these qualifications, Rae's applications have been valuable. They
have identified a new approach for evaluating individuals' preferences for non-
marketed goods and services, and they have contributed to an understanding
of the issues associated with using the random utility model for consistent
benefit measurement.
6.5 MONONGAHELA CONTINGENT RANKING EXPERIMENT: DESIGN AND
ESTIMATES
Since the Monongahela survey was designed to compare approaches for
measuring the benefits of water quality improvements, one section of the ques-
*Since Rae's approach has been followed, and since the role of the prices
of other goods and services has been ignored, the problems raised earlier as
judgmental issues may also have contributed to these findings.
6-16
-------
tionnaire included questions designed to elicit contingent rankings. There
are several important distinctions between the Monongahela survey's contin-
gent ranking component and the procedures used by Rae.
For the Monongahela survey/ individuals were given a smaller number of
alternatives to rank: four combinations of water quality and annual payments
in the form of higher taxes and prices. This number is approximately one-
third that in Rae's experiments and affects the Monongahela survey's ability
to estimate what Rae describes as "individual" models.9" Equally important/
while all the Monongahela survey respondents received the same four sets of
alternatives, individuals in the Mesa Verde and Great Smoky experiments were
randomly assigned one of ten different sets of alternatives to be ranked.
Sufficient experience has not yet been acquired with the estimators of these
models to judge the implications of this difference in experimental design.
A further distinction arises in the composition of each set of alternatives
to be ranked. The procedure used in the Monongahela survey includes four
sets of conditions for the Monongahela River. Table 6-2 details the combina-
tions used, and Figure 6-1 provides an example of the cards presented to
each respondent for ranking. In contrast, the Rae surveys included other
sites in the set of alternatives to be ranked. Specifically/ the Mesa Verde
study included 5 of the 13 alternatives as other sites, and 6 of 14 alterna-
tives in the Great Smoky study were other sites. The rationale for this prac-
tice was described as an attempt to:
reflect the fact that alternative sites are available and to cause re-
spondents to focus broadly on all the characteristics of a site that
contribute to overall enjoyment of National Parks and outdoor recrea-
tion areas. (Rae [1981b], p. 3-1)
Of course, to the extent that one accepts the assumption of independence of
irrelevant alternatives that underlies the random utility models used in these
applications, these other sites should not be important to the rankings pro-
vided by survey respondents.f
The ordered logit estimator permits the estimation of different alternative-
specific effects for each individual in the sample if there are sufficient alter-
natives ranked. See Beggs, Cardell, and Hausman [1981] for a discussion of
the identification problem in such cases.
Rae refers to a constant parameter model for all individuals as the
"aggregate" model and to the model that allows variation in the parameters
describing the effects of the characteristics of alternatives across individuals
as the "individual" model.
TThe procedure used in the Mesa Verde study involved asking respond-
ents first to rank the Mesa Verde alternatives and then to place the non-Mesa
Verde alternatives within the ranking. Presumably, the same procedure was
used in the Great Smoky study.
6-17
-------
Table 6-2. Combinations of Water Quality and Payment for
Monongahela Contingent Ranking Survey
Alternative
Water quality level
Annual payment
RFF water quality index = 0.8
No recreation possible
RFF water quality index = 2.5
Boating possible
RFF water quality index = 5.0
Boating and fishing possible
B
RFF water quality index = 7.0
Boating, fishing, and swimming possible
$5
$50
$100
$175
WATER QUALITY LADDER
$100
IMTPOSSIILE
MATER QUALITY
WORST ra*s»LE
WATCR QUALITY
A
SAFE FOR SWIWKISO
GAME FISH LIKE IASS
CAM LIVE IM IT
§ J
OKAV f OR UOfiTJWG
i*^»i
Figure 6-1. Rank order card.
6-18
-------
Finally, the payment vehicle included in the rankings conducted by Rae
was a price per unit of use--an entry fee to the park. By contrast, the pay-
ment vehicle in the Monongahela survey was independent of the use made of
the river. It is therefore an adjustment to income. This distinction affects,
as we noted earlier, the interpretation of the specifications for the random
utility model.
The rank order cards used to describe each alternative included the RFF
water quality ladder as described earlier in Chapter 4 and repeated in Fig-
ure 6-1. All survey respondents were asked to rank the four alternatives
summarized in Table 6-2. In making these judgments, interviewers were in-
structed to refer to the value card (see Figure 4-6 in Chapter 4) and to ask
individuals to consider actual and anticipated use of the river. The specific
question used was:
First, I would like you to rank the combinations of water quality
levels and amounts you might be willing to pay to obtain those levels
in order from the card, or combination, that you most prefer to the
one you least prefer. I would like you to do this based only on
your use and possible use in the future of the Monongahela River.
That is, keeping in mind only Parts I and II of the value card.
Two hundred thirteen of the 301 survey respondents provided usable
rankings and family income information. Thus, they provide the basis for the
empirical analysis. We have followed Rae's implicit assumptions and inter-
preted our model as an approximation to an underlying indirect utility func-
tion. However, given the incomplete information on an individual's other con-
sumption choices, we have not attempted to include the prices of other goods
or to impose restrictions on the nature of the function estimated. A variety
of specifications for the model were considered under this general format and
the "best" selected based on the ability of the model to "fit" the data and
agreement of the signs of the estimated parameters to a priori expectations.
The final section of this chapter discusses the implications of extending the
model to consider the role of other prices in the indirect utility function.
As noted earlier in this chapter, two estimators have been developed for
random utility functions. One of them, an ordered logit estimator, was used
in Rae's analysis of the Mesa Verde and Great Smoky contingent ranking re-
sults. Because it exhibited rapid convergence and performed reasonably well
in unpublished Monte Carlo experiments performed by V. K. Smith and D.
Waldman to evaluate the estimators, the logit has been used to screen alter-
native specifications for the random utility model.* The second estimator
*To evaluate the relative performance of the ordered logit and ordered
normal models, Smith and Waldman [1982] conducted a limited number of
sampling studies. In general, each estimator performed best with the experi-
ments using the estimator's assumed error (i.e., Weibull for ordered logit,
normal for ordered normal). However, the ordered normal was close to com-
parable to the ordered logit with the Weibull distribution.
6-19
-------
based on a normal specification for the errors in the utility function has much
greater computational costs and was therefore applied to only the "final" model
specifications for comparative purposes.*
Table 6-3 reports a selected set of results for the random utility model
with the ordered logit model. Variables describing the alternatives ranked
and the features of the individual respondent were included in the model.
The models are distinguished according to the variable used to interact with
features of respondents (payments or water quality); the specified form of
the relationship between family income and payment in the model; and the at-
tributes of respondents included in the model. Water quality was measured
using the RFF index scale as it appeared on the rank order cards presented
to survey respondents. The income measure is family income in thousands of
dollars. Age (in years), education (in years), race (1 = white), and sex
(1 = male) qualitative variables were also considered. Three additional quali-
tative variables were also included in some of these models: boat ownership
(Boat own = 1 for owners); participation in any outdoor recreation in the past
year (participate = 1 if yes); and the individual's attitude toward paying for
the costs of controlling water pollution (attitude = 1 if individual considers
himself very or somewhat willing).
This study's results provided stronger support for the methodology than
Rae's findings. Both the payment and water quality measure are statistically
significant and correctly signed in most of the model specifications. The ex-
perimental design induced a high correlation between payment and water qual-
ity (simple correlation = 0.99), and this may explain the results for specifica-
tion (2) in the table. Each equation in the table has three columns to identi-
fy whether it is an individual-specific variable entered individually in the
model (the first column) or a respondent-specific variable entered in interac-
tion form with either the payment (the second column) or water quality (the
third column). Respondent-specific variables must be entered in interaction
form because the rankings are modeled as a function of the differences be-
tween the values of the deterministic portion of the random utility function
for each of the alternatives being ranked. Consider a simple example. Let
V.. designate the utility individual i derives from alternative j. Individual i
will rank alternative j superior to alternative k if V.. > V... Thus, the prob-
IJ IK
ability that alternative j is ranked ahead of k will be equal to the probability
that V.. > V.k. If it is assumed that the deterministic component of V Is a
linear function of one individual characteristic (Zj.) and one variable describ-
ing the alternative (Z2.), V.. can be rewritten as: '
V.. = a0 + B! Zlf + a2 Z2j + £- (6.16)
Using the same relationship to describe Vjk gives the following expression for
V • • ™ V • i •
ij ik
Comparability between the results of logit and probit models for bivari-
ate dichotomous problems, as found in Hausman and Wise [1978], do not neces-
sarily apply. The two error assumptions will yield approaches that are equally
comparable with ranked data.
6-20
-------
Table 6-3. Selected Results for the Random Utility Model with Ranked Logit Estimator'
o>
Independent variables
Alternative specific
Payment (P)
Water quality (WQ)
P x WQ
Individual specific0
Income (x)
Income (t)
Participate (x)
Boat own (x)
Age (x)
Sex (x)
Education (x)
Race (x)
Attitude (x)
Log (L)
(1)
Interaction
with Individual-
specific .
Mturnntivn variables
specific P WQ
-0.151
(-1.437)
-0.10 x 10~3
(-1.760)
0.150
(3.384)
-0.002
(-1.280)
0.077
(1.911)
0.016
(2.762)
-0.015
(-0.247)
-656.25
Model and alternative- specific Interaction
(2) (3)
Interaction Interaction
with Individual- with Individual-
specific fa specific b
*Mt.rr..tilf. variables Aiti.rn.fiu-. variables
specific P WQ specific P WQ
-0.044 -0.046
(-0.236) (-8.922)
0.030 1.364
(3.025) (8.931)
-0.006
(-9.342)
0.42 x 10~5 0.15 x 10~4
(-0.002) (0.900)
-0.055 0.005
(-0.967) (0.949)
-0.004
(-3.342)
0.075
(1.732)
0.017
(2.530)
0.380
(7.455)
-550.69 -628.03
(4)
Interaction
with individual-
specific..
Alternative variables
specific P WQ
-0.067
(-3.957)
1.919
(4.121)
0.25 x 10~3 -0.43 x 10~2
(0.581) (-0.370)
-0.137 0.402
(-0.942) (1.022)
0.0004 -0.015
(1.435) (-1.846)
-628.28
"The numbers In parentheses below the estimated parameters are the asymptotic t-ratios for the null hypothesis of no association; n = 213. (continued)
bThe columns (I.e., P or WQ) Indicate which Interaction Is used In each model specification.
°The multiplication signs (x) Indicate that the Individual-specific variable Is entered In multiplicative Interaction with either the payment or water quality.
The division sign (+) Indicates that Income Is entered In as a division.
-------
Table 6-3. (continued)
O>
r\>
PO
Model and alternative-specific Interaction
Independent variables
Alternative specific
Payment (P)
Water quality (WQ)
P x WQ
Alternative
specific
-0.062
(-9.769)
1.300
(9.113)
(S)
Interaction
with Individual-
specific K
variables
P WQ
Alternative •
specific
-0.053
(-8.215)
0.999
(6.572)
(6)
Interaction
with Individual -
specific b
variables • ... ..
P WQ specific
-0.048
(-8.101)
0.959
(6.520)
(7) (8)
Interaction Interaction
with Individual- with Individual-
specific . specific.
variables Alternative variables
P WQ specific P WQ
-0.043
(-7.764)
0.706
(5.230)
Individual specific
Income (x)
Income (+)
Participate (x)
Boat own (x)
Age (x)
"5
-0.20 x 10
(0.035)
-0.002
(-0.877)
-0.260
(-7.000)
-0.003
(-1.700)
-0.273
(-6.926)
-0.280
(-7.000)
-0.094
(-1.709)
Sex (x)
Education (x)
Race (x)
Attitude (x)
Log (L)
0.0006
(2.476)
0.012
(7.316)
-600.19
0.0004
(2.000)
0.013
(6.944)
-571.69
-0.0001
(-0.066)
0.0006
(3.374)
-598.44
0.010
(1.667)
0.351
(7.468)
-567.99
213.
*The numbers In perentheses below the estimated peremeters ere the asymptotic t-ratlos for the null hypothesis of no association; n
bThe columns (I.e., P or WQ) Indicate which Interaction Is used In each model specification.
cThe multiplication signs (x) Indicate that the Individual-specific variable Is entered In multiplicative interaction with either the payment or water quality
The division sign C+) Indicates that Income Is entered In as a division.
-------
Vij " Vik = (a° + ai Zij + a2 Z2j + ejj> " + ai Zij + 32 Z2R + e.k) (6.17)
Simplified, this expression is:
V,, - V.k = a2(Z2j - Z2k) + s.. - e|k . (6.18)
Thus, the variables describing each individual are not involved in describing
how that individual ranks alternatives since they will remain constant for all
alternatives.*
One of the most puzzling aspects of the results is the effect of the income
variable. Because the payment vehicle was constant regardless of the level of
use, the multiplicative interactions between income and the payment or between
income and water quality would have been expected to provide better results
than income divided by payment. However, the results indicate that the
income divided by payment form is a significant determinant of the utility
function implied by the rankings, while the other forms are not. In all cases,
the signs for the estimated parameters are difficult to interpret. A priori
expectations would have suggested that income relative to payment be a posi-
tive determinant of utility and not negative.
Of the remaining determinants considered, only education and the atti-
tude toward paying for the costs of controlling water pollution were consist-
ently significant determinants of utility. Both variables' parameters are con-
sistent with a priori expectations. Based on the value of the log-likelihood
function at the maximum (LOG[L]> and the significance and consistency of the
estimated parameters, Specification (8) was selected as the final model. It
was reestimated with the Keener-Waldman [1981] ordered normal maximum
likelihood estimator. Table 6-4 reports these results along with estimates for
Model (7) for comparison purposes and repeats the ordered logit estimates for
convenience in comparing the two estimators with each of these specifications.
The two estimators yield quite similar results. The signs and significance
of estimated parameters are comparable for the final model and for Specifica-
tion (7). In general, the Keener-Waldman [1981] estimated parameters are
smaller in absolute magnitude than the ordered logit. There are no specific
implications of this difference, because both estimators involve scaled coeffi-
cients and the estimated parameters do not correspond to the marginal effects
of individual variables on the level of utility. These difficulties in evaluating
the effects of the estimator on the conclusions drawn from these methods sug-
gest that the Rae measure of the benefits associated with a water quality
improvement should be calculated with each of the estimator results for the
final model (i.e., Specification [8]). These results will be considered in the
next section of this chapter.
*See Beggs, Cardell, and Hausman [1981] for further discussion of the
limitations in specifying models based on ordered data.
6-23
-------
Table 6-4. Comparison of Ordered Logit and Keener-Waldman
Ordered Normal ML Estimator
mb (8)c
Independent
variable
Alternative
specific
Payment (P)
Water quality
Individual
specific
Income/p
Boat own
Education
Sex
Attitude
Log (L)
Ordered logit
-0.048
(-8.101)
0.959
(6.520)
-0.273
(-6.926)
—
0.0006
(3.374)
-0.0001
(-0.066)
__
-598.44
Keener-Waldman
-0.039
(-7.073)
0.760
(5.630)
-0.070
(-5.667)
—
0.0006
(3.000)
0.0009
(0.643)
--
-619.46
Ordered logit
-0.043
(-7.764)
0.706
(5.230)
-0.280
(-7.000)
-0.094
(-1.709)
0.010
(1.667)
—
0.351
(7.468)
-567.99
Keener-Waldman
-0.033
(-7.196)
0.510
(3.400)
-0.170
(-4.250)
-0.039
(-0.796)
0.010
(2.000)
0.330
(8.462)
-582.34
The numbers in parentheses below the estimated coefficients are asymptotic
t-ratios for the null hypothesis of no association.
This specification involves payment interaction with the individual-specific
variables.
cThis specification involves water quality interaction with the individual-
specific variables.
6-24
-------
6.6 BENEFIT ESTIMATES WITH CONTINGENT RANKING MODELS
Using ranked data, both estimators for the random utility model provide
scaled values of the parameters. As a consequence, the estimates do not per-
mit direct evaluation of the utility change associated with a change in water
quality. It is nonetheless possible, given that the function is interpreted as
an indirect utility function, to define the compensating surplus associated with
changes in water quality. Compensating surplus would correspond to the
change in income that just offsets the increment to utility associated with the
water quality change. Thus, it could be derived by taking the total differen-
tial of the estimated random utility function with respect to income and water
quality and by solving for the income change that would be equivalent (in its
effects on utility) to any water quality change. This approach is directly
analogous to the definition of compensating surplus in terms of the expendi-
ture function.* Thus, in principle, the model can be used to derive a theore-
tically consistent benefit measure for changes in environmental amenities.
However, as noted earlier, this procedure implicitly assumes that the indirect
utility function is theoretically well behaved.t
Rae's procedure defines benefits as the change in entry fee that would
offset a change in the environmental amenity (see Equation [6.15). The bene-
fit measure for the Monongahela survey was also defined in terms of a total
differential, measuring the change in payment that will offset a water quality
change. As we noted earlier, since the payment vehicle is not a fee per unit
of use but an adjustment to income, regardless of the individual's use of the
river, the measure of compensating surplus should be invariant to the use of
income or of the payment in the total differential equation. If the indirect
utility function is theoretically consistent, the two measures should be equal
and opposite in sign.
Of course, it should be acknowledged that the Monongahela application
has maintained Rae's basic model and therefore implicitly assumes that all
other goods and services are either part of a Hicksian composite commodity or
included in a separable homothetic subfunction. To the extent neither of
these assumptions provides a plausible basis for treating other goods' and
services' prices, estimates of compensating surplus will likely be affected.
One area seems to be an especially clear example of the limitations of this as-
sumption. The Monongahela respondents may well have used other water-based
sites in the region. These sites provide services that substitute for what is
*See Hause [1975]; Freeman [1979a]; and Just, Hueth, and Schmitz
[1982] for further details.
tThe properties of an indirect utility function (IDF) include:
IDF is continuous in prices and income,
i
IDF is nonincreasing in prices and nondecreasing in income,
IDF is quasi-convex in prices, and
IDF is homogeneous of degree zero in prices and income.
See Varian [1978], pp. 89-92.
6-25
-------
proposed for the Monongahela sites under the various hypothetical water qual-
ity scenarios. It must be expected that they will have a different substitution
influence than the remaining goods and services consumed by these individ-
uals. This would suggest that, at a minimum, measures of the "prices" (i.e.,
travel and time costs) of trips to these sites should be included in the spec-
ification of the indirect utility functions for recreation ists in the sample.* In
addition, it implies the need for careful consideration of the relationship be-
tween whether the individual was a user of the sites along the Monongahela
and the corresponding specification of the indirect utility function. Since the
models used in this study do not reflect these considerations, they should be
treated as fairly crude approximations of the indirect utility functions required
for benefit estimation.!
The exact nature of the estimating equation for benefits will depend upon
whether the individual -specific variables enter the model as interactions with
water quality or with the proposed payment. To illustrate the difference,
consider two simple specifications for the random utility function. In Equa-
tion (6.19), the model includes payment (P), water quality (WQ), and an in-
dividual-specific variable (Z) using a payment interaction, whereas Equation
(6.20) uses the water quality interaction. Equations (6.21) and (6.22) report
the corresponding equations for measuring the payment increase equivalent to
water quality improvements for each:
Va = Oi? + a2WQ + a3P-Z . (6.19)
P3WQ-Z . (6.20)
dP = - >g (payment interaction format). (6.21)
dp _ . (Pz + paZ)dWQ (water qua|ity interaction format). (6.22)
It is clear from the specifications that, in either Equations (6.21) or
(6.22), the benefit estimates will vary with the individual—depending on the
individual -specific variables entering the final model used to summarize the
respondents' rankings. Table 6-5 reports the average and range of benefit
estimates for the final specification (i.e., with the water quality interactions)
of the random utility model for using both the ordered logit and ordered
normal models. Because the final specification included a term with income
measured relative to the payment, the estimated' benefits for specified water
*These issues are currently being considered in followup research.
tit should also be acknowledged that the benefit measures calculated with
the income change were several orders of magnitude greater than the price
change and had the wrong sign. These results would be expected because
the estimated parameter for the income variable had an incorrect sign in all
models.
6-26
-------
Table 6-5. Benefit Estimates from Contingent Ranking Models'
Estimator
Average
Range
1
Ordered logit
Ordered normal
11
Ordered logit
Ordered normal
III
Ordered logit
Ordered normal
IV
Ordered logit
Ordered normal
V
Ordered logit
Ordered normal
VI
Ordered logit
Ordered normal
VII
Ordered logit
Ordered normal
VIII
Ordered logit
Ordered normal
Payment = 5
-1.45
-17.72
Payment = 50
62.76
64.30
Payment = 100
60.04
62.12
Payment =175
59.47
61.65
Payment = 5
-2.62
-30.91
Payment = 50
112.97
115.75
Payment =100
108.06
111.81
Payment = 175
107.04
110.97
Water quality change = Beatable to fishable
-72.46 to 208.67
-136.87 to 156.83
Water quality change = Beatable to
39.74 to 83.31
38.54 to 85.51
Water quality change = Beatable to
36.74 to 74.40
36.27 to 78.40
Water quality change = Boatable to
36.12 to 72.66
35.80 to 76.96
Water quality change = Boatable to
-130.42 to 375.61
-246.37 to 282.30
Water quality change = Boatable to
71.53 to 149.96
69.38 to 153.91
Water quality change = Boatable to
66.12 to 133.92
65.29 to 141.12
Water quality change = Boatable to
65.02 to 130.78
64.44 to 138.53
fishable
fishable
fishable
swimmable
swimmable
swimmable
swimmable
These estimates are based on the 213 observations used to estimate the random
utility functions.
For final model, Specification (8).
6-27
-------
dP
quality improvements will change with the payment level at which d^g- is eval-
uated. The results in Table 6-5 are presented for each of the four payment
levels indicated on the rank order cards, as well as for each of two water
quality changes—beatable to fishable water quality and boatable to swimmable
(using the RFF index on the rank order cards). The results are clearly
implausible for the lowest payment level (i.e., P = 5). Because the water
quality change represents an improvement, negative values imply that improved
water quality decreases individual well-being. However, for payment levels
ranging from $50 to $175, the benefit estimates are stable for each water
quality change (i.e., boatable to fishable and boatable to swimmable) and are
approximately the same order of magnitude as the values derived from direct
questioning of survey respondents. (More details on these types of compari-
sons are provided in the next chapter.) These estimates should be interpreted
as being comparable to an option price for each water quality change, because
the question identified both use and anticipated use as the basis for the rank-
ing solicited from survey respondents.
The benefit estimates derived from the order normal model seem slightly
higher than the ordered logit and exhibit a consistently wider range. Finally,
the estimates remain quite stable as the payment level increases from 50 to
175. In Appendix C, comparable benefit estimates are reported for a model
using payment interactions for the individual specific variables (see Equation
[7] in Table 6-3). For this case, the results are also implausible at the low-
est payment level. There is a somewhat larger difference between the ordered
logit and normal estimates, with the averages for logit ranging from $49.17 to
$51.40 for a change in water quality from boatable to fishable (and payments
from $50 to $175) versus $68.75 to $72.45 for the ordered normal. Nonethe-
less, these changes are rather modest overall. The estimated benefits seem
quite stable across the alternative specifications of the random utility model.
6.7 IMPLICATIONS AND FURTHER RESEARCH
This chapter has described and applied the contingent ranking method-
ology for evaluating the benefits from changes in environmental amenities such
as water quality. In the process of developing the background for this ap-
proach, the first applications of the approach by Rae were evaluated. This
appraisal indicated that the empirical results yielded a relatively weak associ-
ation between visibility and the individual's ranking of the alternatives de-
scribing conditions at either the Mesa Verde or Great Smoky Parks. The em-
pirical results for the Monongahela study provide much stronger support for
the method. However, analysis of the theoretical foundations of the method
Rae used for benefit estimation indicated it required quite stringent assump-
tions to be treated as an approximation of a theoretically appropriate benefit
measure. It should be acknowledged that the evaluation of Rae's approach
was based on an attempt to infer the implicit assumptions for his models. The
underlying behavioral model and assumptions were not explicitly described in
either report. Thus, this interpretation should not be attributed to his
reports.
6-28
-------
The analysis performed here has begun the development of the behavioral
underpinnings for the random utility models applied to contingent rankings of
alternatives involving environmental amenities, but the process is not complete.
Models estimated with samples composed of users and nonusers of the Monon-
gahela River sites have been used. A priori expectations would suggest that
nonusers may require specifications for their indirect utility functions that
are different from those of users. The latter should include the prices (i.e.,
travel costs) for all the relevant substitute sites and the payment as an
adjustment to income. By contrast, nonusers' indirect utility functions would
not include these travel cost arguments.
Extensive analysis of this alternative framework for modeling respondents'
rankings of the water-quality/payment alternatives was beyond the scope of
the current project. The primary intention of this analysis has been to apply
and evaluate the Rae/Charles River Associates methodology for benefit estima-
tion. The analysis considered the appropriate interpretation of their proposed
benefit estimator, defined an approach to benefit estimation that more closely
approximated a theoretically consistent measure, and evaluated several models
with two estimators of the random utility framework.
In an attempt to gauge whether these model revisions would be important,
the models used were reestimated for a subset of the respondents—those indi-
viduals who used only one of the sites on the Monongahela River (i.e., elimi-
nating nonusers and those who used more than one site). For this sample (a
total of 49 observations), the implications of treating all sites as perfect sub-
stitutes were considered, and, therefore, only the travel cost of the particular
site used was entered. The results with the ordered logit estimator for models
estimated with this sample under these assumptions were rather poor and sug-
gest that the full sample of users and a more complete specification of the
model will be required to judge the potential importance of the theoretical
arguments calling for different random utility models for users and nonusers.
6-29
-------
CHAPTER 7
A GENERALIZED TRAVEL COST MODEL FOR MEASURING THE
RECREATION BENEFITS OF WATER QUALITY IMPROVEMENTS
7.1 INTRODUCTION
While previous chapters have considered "direct" methods of eliciting
individuals' valuations of water quality changes, all of which require that
individuals be directly asked about their willingness to pay for water quality,
this chapter describes an "indirect" method for benefit estimation. This
method uses individuals' actions and a behavioral model that describes indi-
viduals' decisions in order to infer water quality values. Specifically, using
a generalization of the travel cost model to describe recreation site demand,
this approach involves describing the influence of recreation site character-
istics, such as water quality, on the demand for a site's services. To accom-
modate variations in demand for each site's services, the generalized travel
cost model uses variations in site attributes across a large number of water-
based recreation facilities.
In the process of developing the model, the analysis has attempted to
consider a number of the problems associated with the travel cost framework,
including the following:
The estimation of the opportunity cost of the time spent travel-
ing to a site.
The treatment of time spent at the site during each trip in
relationship to additional trips to the site.
The specification of the model, including the prospects for
biased results from conventional statistical approaches.
The implications of multiple-purpose trips for the validity of
the model.
The estimation of the specific effects of site attributes on the
nature of each site's demand function.
This chapter discusses each of these issues in detail. Specifically, Section 7.2
reviews the economic basis for the travel cost model using Becker's [1965]
household production framework, and Section 7.3 generalizes the conventional
treatment of the travel cost model as a derived demand, assuming site services
are inputs to the production of recreation activities. In particular, Section 7.3
considers the problem of modeling site attributes in developing an appropriate
7-1
-------
quantity index for site services, and it proposes a variant of Saxonhouse's
[1977] generalized least-squares estimator to implement the model. Section 7.4
describes the recreation choice and site attribute data used to estimate the
travel cost model, and Sections 7.5 and 7.6 present and evaluate results from
individual site demand models. In these two sections, as throughout the
chapter generally, a major objective is to gauge the implications of modeling
decisions for each site demand model used to develop the generalized travel
cost model. Section 7.7 presents the generalized travel cost model, and
Section 7.8 describes its use to estimate benefits with survey data from users
of the recreation sites along the Monongahela River in Pennsylvania. Finally,
Section 7.9 presents a brief summary.
7.2 TRAVEL COST MODEL
The travel cost model is widely used to describe demand for recreation
facility services (see Dwyer, Kelly, and Bowes [1977] for a review). Indeed,
the most recent Water Resources Council [1979] guidelines for benefit-cost
analysis call for travel cost methods to estimate the economic value of recrea-
tion sites. Although the travel cost model is usually credited to a suggestion
made by Harold Hotelling to the Director of the National Park Service (that
distance traveled can indicate the implicit "price" recreationists pay for using
a particular facility), Clawson [1959] and Clawson and Knetsch [1966] were
the first to develop empirical models based on it. The travel cost model has
been refined since this early literature, and it is now recognized as an impor-
tant indirect methodology for valuing environmental amenities, especially water
quality (see Freeman [1979a], Chapter 8, and Feenberg and Mills [1980]).
Of course, recognition of the travel cost model has not come without the
parallel development of a behavioral model for the demand patterns it
describes. For example, Becker's [1965] household production model can
analyze individuals' recreation choices.* While the household production model
does not imply new testable hypotheses (see Pollak and Wachter [1975]), it
does offer a useful conceptual framework to describe household behavior,
especially with respect to outdoor recreation.t
The absence of uniform types of household recreation data and the lack
of organized markets for most recreation site services have compounded the
problems of describing consumer demand. Therefore, a framework that can
be constructed using the available recreation data has distinct advantages
over frameworks that do not. Because these advantages have elsewhere been
discussed in detail (see Smith [I975a]; Deyak and Smith [1978]; Cicchetti,
*ln what follows Individual and household are used synonymously.
Based on Becker's [1974] work, such conventions do not require models
specifying a dictatorial decision process for the household. Rather, house-
holds can be seen to act as if guided by a single utility maximizer when
altruistic behavior is recognized as an integral component of the social inter-
actions of family members (see Becker [1981] for more details).
tit can also provide a basis for consistent welfare measurement. See
Bockstael and McConnell [forthcoming].
7-2
-------
Fisher, and Smith [1976]; and Bockstael and McConnell [1981]), they will not
be developed here.
The basic distinction between the household production framework and
other approaches stems from its portrayal of the household as both producer
and consumer. That is, the household is assumed to consume only services
that it produces. For convenience, these services will be designated as fina|
service flows. As with any other production process, these services require
inputs. In this case, however, the inputs involve the household's time, as
well as market-purchased goods and services. Thus, the framework considers
the purchased goods as an indirect means to maximize utility.
The household production framework has two steps or stages, which,
though purely logical abstractions, can explain how households make decisions.
The first step involves selecting market goods and services and allocating
available household time to minimize the costs of each possible set of final
service flows. In the second step, based on the outcomes of the first step,
the household defines for itself the "shadow prices," or marginal costs, of
each of the final service flows. Thus, along with the relevant "full" income
budget, marginal costs are implied by the selection process for final service
flows.
For this study, constrained utility maximization in the household produc-
tion framework highlights several important aspects of the travel cost model,
the first of which is the distinction between the recreation activities under-
taken by a household—such as boating, fishing, or swimming—and the usage
level of a particular recreation site. To readily identify the implicit price of
services of a recreation facility, the former are best treated as measures of
household recreation final service flows, and the latter are best treated as
an input to the production of such service flows.
Furthermore, the household production framework can readily identify
the various ways site services are used. That is, the framework can dis-
tinguish whether an individual uses more of a site's services by visiting it a
greater number of times during a recreation season or by spending more
time at the site during fewer visits. This choice implies a simultaneity problem
in modeling household decisions on visits and onsite time per trip. Past
efforts have implicitly avoided this problem by assuming that all visits (across
all users) are of fixed length (see Cicchetti, Fisher, and Smith [1976]) or by
estimating separate models for each trip length '(Brown and Mendelsohn
[1980]).
Finally, the household production framework permits a general discussion
of a household's use of multiple recreation sites that produce identical recrea-
tion activities, thus allowing the incorporation of site attributes as determi-
nants of the differences in the demands for the services of multiple sites.
In its simplest form, the household production model can describe recrea-
tion decisions by simply distinguishing two types of final service flows
produced and consumed by households. The first is the recreation service
flow, 2 , and the second is a nonrecreation service flow, Z . Because
1 nr
7-3
-------
sets of service flows can be expanded without fundamental changes in the
implications of the model, the present analysis has been confined to this
simple case. Following earlier developments of the model (see Cicchetti,
Fisher, and Smith [1976] as an example), the production function for recrea-
tion services can be specified in terms of five inputs: the purchased goods
associated with recreation (e.g., equipment for fishing, boating, camping,
etc.), Xr; the number of visits to each of two distinct recreation sites, Vx
and V2; and the time per visit to each site, t and t . It is important to
M. ^2
note that this specification greatly simplifies the analysis by maintaining that
onsite time per visit is the same for all visits to a given facility.
Equation (7.1) provides a general functional representation of the recrea-
tion services production function:
lf 2,
(7.1)
The time horizon for production activities is often unspecified. However, the
household must be assumed to make decisions over some predefined time
horizon that involves a full recreation season (or some fraction) during
which multiple visits to different sites are possible.
The production function in Equation (7.1) implicitly maintains that each
(V., t ) pair ideally measures the services provided by each site. Thus,
this function effectively skirts a significant index number problem* because
differences in the productivity of one site's services for the recreation
service flow are embedded in the function itself. The next section adds
further assumptions to this function to investigate the rationale for skirting
the index number problem.
Because the focus here is on decisions related to recreation activities, the
nonrecreation service flow can be expressed in rather simple terms as related
to nonrecreation-related purchased goods, X , and household time spent on the
nonrecreation service flow, t , as in Equation (7.2):
Znr = fnr
-------
In terms of its relationship to practical applications of the travel cost
model, one of the most important aspects of the household production frame-
work arises with the definition of the household's budget constraint. Following
Becker's [1965] original suggestions, the household is assumed to face a "full
income" constraint, Y, including wages, wt , nonwage income, R, and fore-
gone income, L. However, it is not assumed that the household necessarily
treats the market wage as the opportunity cost of its time in all household
production activities. This formulation can be seen as a generalization to that
proposed in Cicchetti, Fisher, and Smith [1976]. Equation (7.3) defines this
budget constraint:
Y = wtw + R + L = PrXp + PnXn
(7.3)
(T«d2 + rt2 + w2tv )V2
where
P , P = the prices of market-purchased recreation -and nonrecreation-
r n related goods
T = the travel cost per mile
d. = the roundtrip mileage to the ith site
r = the individual's opportunity cost of traveling time
t. = time for each roundtrip to the ith site
w. = the individual's opportunity cost for onsite time at the ith
site.
Equation (7.3) identifies three important components of the unit cost of each
visit: the travel costs associated with the vehicle used to reach the site, the
time costs of the trip, and the opportunity costs of time spent on the site.
Only the last of these costs is a choice variable, because the distance and
time to reach a recreation facility are defined by the location of that facility
in relation to the individual's origin point. Because the model assumes that
these locational choices are already determined, their costs are outside the
individual's control.*
*Of course, this statement assumes that the individual's opportunity cost
of traveling, r, is treated as a fixed parameter to the recreation decision
process.
7-5
-------
The past literature has devoted considerable attention to the appropriate
treatment of the travel and time costs of a trip in the formulation of travel
cost demand models. Cesario and Knetsch [1970, 1976] have suggested that
the opportunity cost of travel time, r, is less than the wage rate, w, and, in
some cases, that travel and time costs may not be additive. The latter compo-
nent of the Cesario-Knetsch argument has been difficult to substantiate without
dropping the assumption that the opportunity cost of travel time is a parameter
in the individual's decision process.
For practical purposes, the travel cost literature has tended to focus on
the relationship between the cost of travel time, r, and the wage rate, w.
Cesario, for example, has suggested that since the cost of travel time involved
in urban transportation decisions likely falls between one-fourth and one-half
the wage rate (see Cesario [1976], p. 37), one-third might be used as a
reasonable approximation for travel cost models. In contrast, McConnell and
Strand [1981] have estimated the fraction to be six-tenths for sports fishermen
in the Chesapeake Bay region. Their model assumes that the opportunity
cost of travel time is a parameter estimated from the data and that travel
costs and time costs of travel have equivalent effects on the demand for a
site's services. McConnell and Strand caution that this parameter may vary
among regions and sites.
The only notable exception to the treatment of r as a multiple of the
wage rate arises in Wilman's [1980] recent attempt to compare the Cesario and
McConnell approaches for estimating the costs of recreation trips. Wilman's
analysis sought to distinguish "scarcity" and "commodity" values for time in
modeling the relationship between trips taken and onsite time per trip to
produce recreation service flows.* The Wilman model specifies utility as a
function of goods and services requiring time, goods and services not requir-
ing time, and two measures of a recreation site's use—the number of visits of
a given length to a site and the number of roundtrips to that site. Round-
trips are intended to reflect any satisfaction derived from traveling to the
recreation site. By assuming that the time and budget requirements are fixed
multiples of the number of visits and roundtrips, Wilman links these choice
variables to the household's time and income constraints.
The basis for Wilman's derivation of a different implicit valuation of
travel and onsite times is an assumption that the number of trips and visits
to a site are equal. The resulting first order conditions require equality
between the sum of the marginal utilities of trips and visits and the corres-
ponding goods and time costs of each (weighted by the appropriate Lagrangian
multipliers).
Wilman's definition of commodity and scarcity values of time is simply a
rearrangement in this allocation condition for visits and trips in an attempt to
*lt should be noted that Wilman did not explicitly adopt a household
production framework. However, with relatively minor amendments, her
analysis could be cast in these terms.
7-6
-------
account for the potential utility derived from travel time. It is important to
recognize that the framework maintains that trips and visits are delivered
jointly on a one-to-one basis in this version of her model. They must be
treated as a single commodity, and any cost allocation between them is arbi-
trary. Indeed, once the equality assumption between trips and visits is
dropped, Wilman's model implies that both types of time should be valued at
their scarcity value (see Wilman [1980], Equation [24]). Thus, the existing
recreation literature does not provide an unambiguous theoretical justification
for distinguishing the valuation assigned to the travel and onsite time com-
ponents of a recreation experience.
The household production framework and the procedures used to compile
the data for an empirical estimation of travel time costs permit direct investiga-
tion of the relationship between the travel time costs and the onsite time costs
of the trip. Therefore, the generalized statement of distinct opportunity costs
for each time of travel can be accommodated within the empirical model.
To complete the model it is necessary to maintain that the household's
utility is a function of the levels of the two final service flows produced as
U(Z , Z ). Maximizing this utility function subject to the budget and pro-
duction constraints yields a set of conditions that can be manipulated to
suggest that the marginal utility product of each input (i.e., the product of
the marginal utility of a service flow times the marginal product of the input
in the production of that service flow) relative to its market price, or implicit
unit cost, would be equalized over all inputs. More formally, this result is
given in Equation (7.4):
azr
MU7
zr
r-tj
az
MU7 -%£-
Zr 9tVl
azr
MU7 -fl\7
A a V *>
r £
wiVi (T-d2 + r-t2 + W2tv.
azr azo
MU? jrz- MU7 jrrf MU7 ^
r v* - Zr8Xr _ Zn 8Xn
w2V2 Pp - Pp
W
There are two important aspects of these marginal conditions. First, the
assumption that r and w. are parameters allows all aspects of the costs of an
additional visit to each site to be added (i.e., the full cost of a visit to the
ith site is T-d. + r-t. + w.ty ) and treated as the "price" of that visit.
Second, the joint determination1 of trips and onsite time implied by this formu-
7-7
-------
lation is clearly apparent in the dependency of the unit costs of each of these
inputs on the selected levels of the other.*
Solving the necessary conditions of a utility maximum for the optimal
number of visits to each site as a function of the parameters in the optimiza-
tion problem provides the analytical counterpart to the travel cost demand
model. These derived demand equations can be written in general form as
Equations (7.5) and 7.6):
, R, L, Pp, Pn, T-d! + rtlf T-d2 + rt2, wlf w2, w), (7.5)
V2 = L2(wtw/ R, L, Pr/ Pn, T«d2 + rt2, T-dj + rdt, wlf w2, w). (7.6)
The relationships in Equations (7.5) and (7.6) are clearly more general than
the conventional travel cost demand model. Empirical estimation of these
relationships, however, requires several simplifying assumptions. Specifically,
full income (wt + R + L) is assumed to be approximated by family income,
and choices of market- purchased recreation and nonrecreation goods, as well
as time used in nonrecreation final service flows, are treated as separable
decisions in the consumer's budget allocation process. These assumptions
reduce the input demand equations to a format more closely resembling the
travel cost specifications. In the case of the first site, for example, Equation
(7.7) would result:
T-dt + rtlr T-d2 + rt2, w4, w2) , (7.7)
where
Y = family income as a proxy measure for full income (Y) defined in
Equation (7.3).
Before turning to further refinements in this model to accommodate the
introduction of specific features of recreation sites as determinants of the
variation in the site demand functions, it may be useful to relate the amended
travel cost model to some of the existing travel cost studies. (A comprehen-
sive review is available in Owyer, Kelly, and Bowes [1977].) It is acknowl-
edged at the outset that the features of the existing work can often be
explained by inadequacies in the data available on the usage of recreation
sites. Indeed, many travel cost. studies have been based on aggregate visit
patterns rather than on information on the behavior of individual households.
These data are typically the result of automobile surveys or the aggregation
of user permit information at specific recreational sites. However, information
is now available on the number of visitors to a specified site from a set of
*This framework can also be extended to consider an alternative basis
for deriving a relationship between the opportunity cost of travel time and
the wage rate by assuming that individuals face different types of time con-
straints. See Smith, Desvousges, and McGivney [1983] for details.
7-8
-------
origin zones (often counties) around the site. With such information, the
measure of site usage is generally expressed as a visitor rate (i.e., number
of visits relative to county population) and is interpreted as an expected
"rate of usage" for the "representative" individual in the county. County
summary statistics are used as indicators of the economic and demographic
characteristics of this "representative" individual. As a consequence, there
is often no information with which to estimate the individual's wage rate.
In the presence of these limitations, researchers have taken either of
two approaches. The first assumes a constant wage rate for all individuals in
all origin zones. The second, somewhat more desirable, uses an estimate
based on the wage implied by average family income in the origin zone (i.e.,
family income divided by an estimate of hours worked per year). Clearly,
neither of these options provides a discriminating index of an individual's
wage rate. However, the crude nature of the approximations required by the
data explain in part Cesario's [1976] willingness to propose a "rule of thumb"
for estimating the opportunity cost of travel time.
There are several other problems that arise with travel cost models
based on limited data sets. The first of these stems from controlling for
trips of different lengths with an aggregate data set. In some cases,
researchers have separated data into weekend and weekday visits to ameli-
orate the problem (see Cicchetti, Fisher, and Smith [1976]). An assumption
of constant onsite time is otherwise invoked without empirical justification.
Equally important, the nature of the trips may be quite different as the
distance from the site increases. That is, the trips may have multiple objec-
tives that would imply the full cost of the trip is not an implicit price for the
use of the recreation site but, rather, provides other services as well.*
Recent empirical analyses of the stability of the travel cost model using
data aggregated as distance from a site increases suggest it may be possible
to detect when violations of these assumptions are severe (see Smith and
Kopp [1980]). Of course, this analysis requires the assumptions of constant
onsite time across aggregated visits and single-purpose trips, which are more
untenable as the distance from the site increases.
The second type of data available for travel cost models involves site-
specific user surveys. While these data are in principle superior to the
aggregate visit data, incomplete design of the surveys has limited their ulti-
mate usefulness. One especially important omission involves the treatment of
usage patterns for recreation facilities that might be considered substitutes
for the one whose users are questioned.
*Haspel and Johnson [1982] have considered this issue for a survey of
users of the Bryce Canyon National Park and found that for this site the
assumption of single-purpose trips for. visitors was inappropriate. Their
findings suggest that it would lead to substantial differences in the estimated
travel cost demand functions.
7-9
-------
The micro-level data from the surveys, however, have permitted the
investigation of a number of issues in modeling recreation demand, including
the treatment of: travel costs, the time costs of travel, and the costs of
onsite time. Unfortunately, these efforts have not been entirely successful.
For example, McConnell and Strand [1981] assume that increases in travel
cost and in the time costs of travel should have the same effect on site demand
to infer the relationships between the opportunity cost of travel time, r in
this study's notation, and the wage rate, w. Their data do not include wage
rates that require estimation from family income. The resulting demand equa-
tions can exhibit difficulties in estimating precise (i.e., statistically significant)
separate estimates of the "price" and income effects on site demand.
The most recent attempt to include both travel cost and the time costs of
travel with micro-level data by Allen, Stevens, and Barrett [1981] concludes
that it is difficult to distinguish separate effects for these two variables when
time is entered without attempting to estimate its opportunity cost (see
especially pp. 178-179). The authors suggest collinearity would seem to
prevent precise estimation of separate effects of the two variables. Their
conclusions contrast with earlier suggestions by Brown and Nawas [1973] and
Gum and Martin [1975] that disaggregation would help to resolve these estima-
tion problems.
Theory does imply that travel time should be valued by an opportunity
cost. Thus, the Allen, Stevens, and Barrett findings may simply be a reflec-
tion of a failure to use all available information from theory. Moreover, the
McConnell-Strand empirical results support this optimism.
One important aspect of any attempt to include both travel time and
onsite time costs of a trip will be estimation of micro-level wage rates in a way
that accurately reflects individual rates of compensation and does not preclude
the use of family income as a proxy variable. Such a method is developed in
Section 7.4 of this chapter.
The last remaining facet of the idealized travel cost model given in Equa-
tion (7.7) involves the treatment of the influence of substitute sites on the
demand for any one site's services. This model explicitly identifies sites that
can contribute to the production of the recreation service flow, and it thus
requires an approach that treats the effects of other sites. A variety of meth-
ods have evolved to incorporate the influence of substitute sites on demand.
Because these approaches provide a natural introduction to the extended travel
cost model, which allows a site's characteristics to be determinants of intersite
demand variation, they are considered as a part of the introduction to the pro-
posed model in Section 7.3.
7.3 THE TRAVEL COST MODEL FOR HETEROGENEOUS
RECREATION SITES
As noted in the previous section, the travel cost methodology seeks to
model the demand for a recreation site's services. In general, the operational
forms of travel cost models focus on estimating site-specific demand functions,
7-10
-------
and additional sites are considered only to the extent they might provide
substitute services for a particular site under study. Conventional practice
has incorporated the role of these substitute services using one of three
methods:
Incorporation of an index of the relative attractiveness and
availability of other recreation sites into the relevant site's
demand function (see Ravenscraft and Dwyer [1978] and
Talhelm [1978]).
Specification of the recreation demand models to include the
prices (i.e., travel costs and time costs of travel) of other
substitute recreation sites (see Burt and Brewer [1971] and
Cicchetti, Fisher, and Smith [1976]).
Respecification of the utility function in terms of the attributes
of recreation sites so that use patterns are assumed to be in
response to utility maximizing selections of these attributes
(see Morey [1981]).
Of the three methods, the first is probably the least desirable. It impli-
citly assumes that an arbitrary index can account for substitute sites in the
demand for any given recreation site. Of course, the definition of such an
attractiveness index not only requires knowledge of the exact nature of the
substitute relationships but also assumes that the index form would be a
simple function of the other site's attributes. Thus, this approach requires
the very information it is attempting to derive.
The remaining approaches are consistent with economic models of recrea-
tion demand. The second approach can be interpreted as an empirical state-
ment of the model given in Equation (7.7), which assumes that the effects
of substitute sites on any one site's demand can be captured through the
specification that these other sites' "prices" affect the demand for the site of
interest. Because the demand for each site is measured individually, the
second approach avoids the quantity and price aggregation issues that would
impede the consistent definition of the attractiveness index proposed for the
first approach.
The last approach addresses the quantity and price aggregation issues
directly by assuming a specific format for them in the site attribute specifica-
tion of the recreationist's utility function. All recreation]sts are assumed to
have the same preferences. This method can be limited by the plausibility of
the specification of the utility function.
However, none of these methods offers the ability to consistently relate
conventional travel cost site demands to the site features that produce recrea-
tion services. That is, while the specification of the household production
function for Z in terms of several sites implicitly reflects the prospects for
substituting one site's services for another's, there is no direct means for
explaining the reasons for the degree of substitution observed between any
pair of sites. This inability to explain the source of, or reasons for, these
7-11
-------
substitution possibilities is not a limitation for many applications. As noted
earlier, when sample information identifies the set of sites considered by
individuals as well as their respective patterns of use, cross-price elasticities
of demand can be used to estimate measures of the substitution possibilities.
Unfortunately, this information is not uniformly available in all recreation
surveys. Indeed, this study's data set, described in the next section, is a
survey of users at specific recreation sites, without information on the other
recreation facilities respondents may have used or considered using. In such
cases, the reasons why substitution prospects exist between recreation sites
must be analyzed and some attempt made to reflect them in the modeling of
the overall demand for these sites. In simple terms, what is required is the
addition of further structure to the household production functions—assump-
tions that serve to explain why individual site services contribute differentially
to the production of recreation service flows and, in turn, why they substitute
at different rates.
»
Before the analysis is formally developed, its implications must be
described. This study's approach maintains that each site has a set of charac-
teristics (e.g., size, water quality, camping facilities, scenic terrain, etc.)
and that these attributes contribute to site productivity as inputs to recreation
service flow production functions. If the nature of these contributions is
restricted to a specific form, originally defined as the simple repackaging
hypothesis in problems associated with constructing quality adjusted price and
quantity indexes for consumer demand (see Fisher and Shell [1968] and
Muellbauer [1974]), the measurement of the role of site characteristics as
determinants of the features of site demand will provide an explanation of the
substitution. As Lau [1982] has demonstrated in another context, the simple
repackaging hypothesis implies that site services can be converted into equiva-
lent units based only on their respective characteristics. Thus, after adjust-
ment for their attributes (with Lau's conversion functions), all site services
are perfect substitutes for each other.* If this description is plausible, a
model of site demand that omits consideration of the role of potential substitute
sites will not be biased. Of course, it should be acknowledged that this
assumption is a stringent one and that the models developed from it may be
limited should the assumption prove to be a poor approximation of processes
giving rise to substitution.
To begin the formal development of this model, the original specification
of the household production function for recreation service flows (i.e.,
Equation [7.1]) is replaced with one that includes the characteristics of the
recreation site, Equation (7.8):
Zp = fr (Xp, V., t , a.) , (7.8)
*Berndt [1983] has also recently used this framework to describe the
effects of input quality in neoclassical production models.
7-12
-------
where
X = recreation-related market goods
r
V. = number of trips to the ith recreation site
t = time per trip to the ith recreation site (assumed to be constant
vi across all trips)
a. = vector of attributes for the ith recreation site.
In this form, the relationship between V., ty , and a. in the household pro-
duction function for recreation service flows determines the appropriate
index for transforming one site's services into their equivalents for another
site. More specifically, given strict monotonicity of the household production
function, Equation (7.8) can be solved for V..* This resulting function might
be designated a site-service requirements function and would be given (in
general form) by Equation (7.9):
V. = h(Zp, Xp, tv , a.) . (7.9)
Thus, to convert one site's services into equivalent units of another site, the
ratio of the equivalent h(.) functions for each site is needed. t For example,
if there are two sites (designated with subscripts 1 and 2), and if the differ-
ences in the production technologies for Z. using each site can be captured
with a., the equivalence between trips to each is given by Equation (7.10):
h(Z , X , t ,
-V.. (7-10)
This relationship can be further simplified if the ratio Vi/V2 is assumed
to be independent of Zp, Xp, and ty (i = 1,2).f Under this assumption, the
*A monotonic function implies that there is a one-to-one association be-
tween the set of independent variables and the dependent variable. In the
context of a production function this assumption implies that, if an output Q
can be produced with a certain input bundle x, the same output can be pro-
duced with more of every input (provided it is possible to costlessly dispose of
what is not needed).
tThis analysis of the role of site characteristics adapts work recently
developed by Lau [1982] for the definition and measurement of a raw materials
aggregate within neoclassical models of production.
fThe assumption of independence of t can be easily modified by incor-
porating it as one of the set of attributes assumed to be available with each
visit to the site. Indeed, this format is equivalent to the assumption made
earlier that onsite time is the same for all visits.
7-13
-------
site-service requirements function would be given as Equation (7.11), and the
household production function corresponding to it by Equation (7.12):
V. = h (Zr, Xr, tv ) - R(a.) , (7.11)
Zp = fp (Xp, ty/ R(a.) • V.) , (7.12)
where R(a.) = the augmentation function.
R(a.), the augmentation function, provides a specific index that permits
each site's services to be transformed into equivalent units. It maintains that
this transformation will be constant regardless of the level of the site's serv-
ices used and will only vary with changes in the attributes (the a.'s) for a
site. Consequently, the augmentation function describes how sites would sub-
stitute for each other in the production of the recreation service flow, Zp.
This form of the household production function—used in the following dis-
cussion—implies that the effects of a site's characteristics on household
demands for that site's services can be derived if households can be viewed
as engaged in a two-step optimization process to allocate their time and
resources.* One of these steps involves minimizing the costs of producing a
given output, suggesting that the patterns of trips to recreation sites will
be adjusted so the relative unit costs of a trip to any pair of sites would be
proportionate to their respective marginal products in contributing to the
recreation final service flow. In other words, the effective price of a site's
services will be equalized across all recreation facilities considered for use
in the production of the recreation service flow.
If the prices of site services are equalized across sites, the augmentation
function, R(a.), provides the means of relating each site's marginal product.
Thus, for example, using the augmentation function to compare two sites with
different levels of water quality (one with levels permitting recreation fishing
and the other permitting only boating), this distinction is captured analytically
by a higher augmentation coefficient for the site with cleaner water. Desig-
nating the sum of the travel costs and time costs of travel by P. (i.e.,
P. = T-d. + r-t.) then yields:t
p^ PN
(7.13)
or the equivalent of a hedonic price function for sites' services:
Pr = g(aj) . (7.14)
*For further discussion of the application of the household production
model to modeling outdoor recreation behavior, see Deyak and Smith [1978]
and Bockstael and McConnell [1981].
tThis relationship assumes that onsite time is constant and equal for both
sites and that the opportunity costs of onsite time are equal for the two sites.
7-14
-------
This approach is simply an alternative derivation of the first-stage estimating
equation for the Brown-Mendelsohn (1980) hedonic travel cost model. It does
not, however, necessarily imply that the marginal prices of attributes will be
constant.*
A second implication of the above approach is that the household's cost
function for producing Z will be a function of the site's attributes. More-
over, the attribute augmentation function, R(a.), will adjust the effective
price of the site's services in the household's cost function, as in Equation
(7.15):
C = C(Zp, Pp, w., P/R(a.)) , (7.15)
where
P = price of recreation related commodities
w. = price of onsite time.
This cost function provides the basis for a generalized travel cost model.
It is assumed that a given recreation site's attributes do not change dur-
ing a recreation season. Thus, estimates of a travel cost recreation demand
for a single site cannot isolate the role of these attributes. Nonetheless, these
characteristics should, in principle, affect the form of these demand functions
across sites, as seen when Equation (7.15) is differentiated with respect to the
site's price, P.. Following Shepherd's [1953] lemma, the partial derivative is
the Individual's demand for the site's services. Equation (7.16) illustrates
that this demand must be a function of the site's characteristics^
To make the framework in Equation (7.16) operational, a number of
complications must be considered. The first of these issues involves the
recreation service flow, Z , for which there is no measure. As a rule, the
*The Brown-Mendelsohn [1980] hedonic travel cost model proposes a two-
stage framework. In the first stage, the hedonic price function is estimated
for each origin zone by considering the set of recreation sites available to
users in that zone, their respective travel and time costs for trips, and their
attributes. With these data a separate hedonic price function can, in principle,
be estimated for each zone. The partial derivatives of these price functions
(which are assumed to be linear in their application) define the implicit prices
of the sites' attributes for users in each zone. Using the recreation site
choices, their implied levels of attributes, and these implicit prices for attri-
butes, Brown and Mendelsohn then estimate demand functions for each attribute
across all origin zones.
tC4(-) is a short-hand expression for the partial derivative of the cost
function, C(-), with respect to its fourth argument, P./R(a.).
7-15
-------
flows are not part of travel cost demand models—an exclusion that is justified
if the production technology is homothetic and if the levels of production are
uncorrelated with other determinants of the demand for a site's services.*
That is, the first assumption implies Equation (7.16) can be rewritten as:
V* = H(Zr) • g(Pr/ w., P./R(a.), R(a.)) . (7.17)
Rewriting Equation (7.17) in logarithmic form yields:
1n V* = In H(Zr) + 1n(g(Pp, w., P./R(a.), R(a.))) . (7.18)
When Z is uncorrelated with the arguments of g(.), and when 1n(g(.)) is
linear in parameters, the ordinary least-squares estimates of these parameters
will be unbiased. t Of course, this framework assumes that all individuals
produce the same types of activities. t
The second complication arises from the treatment of onsite time. The
model developed in Section 7.2 described the cost of a site's service by
considering the travel and time costs of traveling to the site and the time
spent at the site per visit. For simplicity, the time spent onsite was assumed
constant for all visits. Thus, the full cost, C., of all trips to the ith facility
is given as:
C. = (T-d. + rtj + Wjtv ) V. , (7.19)
where
T = travel cost per mile (operating costs for an automobile)
d. = roundtrip distance in miles
r = opportunity cost of traveling time
*Any production function that can be written as a monotonic, increasing
function of a homogeneous function is a homothetic function. This specification
implies that the marginal technical rate of substitution between all pairs of
inputs will be constant along rays from the origin. In terms of the cost
function corresponding to this production function, the returns to scale (as
measured by the elasticity of cost with respect to output) will be a function
of the output level.
tTo make this judgment, it has been implicitly assumed that the site
demand equations include an additive, classically well-behaved error.
fThe framework implicitly assumes that approximately the same mix of
recreation activities is undertaken by users. The rationale follows from
the assumption that users have comparable household production functions (or
that the factors leading to differences in household production technologies
can be specified). The assumption on the mix of recreation activities is
equivalent to treating Z as an aggregate index of all of the recreation under-
taken at the site. r
7-16
-------
t. = travel time to and from the facility
w. = opportunity cost of onsite time.
A change of one trip involves a full cost of T*d. + rt. + w.tv . The first two
components of these costs are given to each individual once the recreation
site is selected. However, this same conclusion does not follow for w.tv .
The time spent at the site, t , is a choice variable. Thus, if onsite cosb
V»
are included in a travel cost demand model, amending Equation (7.17) to
reflect the restriction implicit in the previously described definition of the
price of a trip, the estimation of the model must reflect simultaneity in the
choice of V. and t . In past studies, this issue has been avoided by assum-
ing that onsite tim'e was constant for all trips.* Section 7.6 evaluates the
importance of this simultaneity for the recreation sites in this study.
The measurement of the opportunity cost of travel time, r, and of onsite
time, w., is also a difficult issue. As noted in the previous section, there
has been considerable controversy over the appropriate treatment of the first
of these implicit prices. Cesario and Knetsch 11970, 1976] and Cesario [1976]
have argued that the wage rate is not an appropriate index of the first of
these costs. Rather, based on individual travel choice studies, they have
proposed that the opportunity cost of traveling time is a fraction of the wage
rate. In this study's sample, the wage rate is estimated based on a wage
model derived from the 1978 Current Population Survey that permits specific
wage predictions to be made for each individual. These predictions take
account of the individual's background, including education, age, occupation,
sex, race, and other socioeconomic characteristics. As a result, it is possible
to separate the estimation of the wage rate from the respondent's reported
family income. The next section provides more complete details on the wage
model and its predictions for the sample of recreationists.
Finally, the theoretical model does not offer explicit guidelines as to how
a site's attributes affect the derived demand functions for that site's services.
The analysis assumes that all of the demand parameters can be affected by a
site's features. With the natural log of visits specificied as a function of the
travel and time costs of visiting the site, income, and a variety of other
determinants ,t using a semi log specification gives the generalized travel cost
specification in its simplest form as:
This assumption was one of the reasons offered by Smith and Kopp
[1980] for a spatial limit to travel cost models estimated from aggregate visit
rate information by origin zone.
tEarlier attempts to discriminate between the popular specifications for
the travel cost model have not met with great success. Using tests for
nonnested models, Smith [1975b] found a slight preference for the semi-log
with aggregate visit rate data. Ziemer, Musser, and Hill [1980] have also
found support for the semilog specification. However, neither set of results
could be regarded as definitive.
7-17
-------
., a2j, . - . , akj)
j, a2j, . . . , ak|) P.. (7.20)
agj/ • • •• , akj) Y. .
The double subscript for V., and P.. permits the identification of the site (i)
and the individual recreationist (j). Thus, V.. is the number of trips to the
ith site by the jth individual, P.. is the travel and time costs per trip for the
jth individual, and Y. is that individual's income. Significantly, each param-
eter of the demand equation is specified as a function of the site attributes.
Thus, for individuals using the services of a single site, the demand func-
tion's parameters are assumed to be constant. Nonetheless, this model has
the ability to describe how the demand for a site's services changes with the
attributes of that facility. Thus, separate estimates of the demands for
individual recreation sites together with measures of their characteristics
provide, in principle, the information needed to determine the demand for new
sites or for existing sites that experience changes in their available character-
istics. These changes might include improvements in water quality, capital
additions increasing access points, or improvements to the camping facilities.
Thus, this analysis demonstrates that the observed variation in the estimated
parameters of travel cost site demand models across sites may be the result of
differences in these sites' characteristics. It therefore provides the basis for
evaluating the implications of water quality for recreation behavior. Indeed,
as suggested in Section 7.7, the estimates of travel cost demand models
together with the attributes explaining the variation in the estimated parame-
ters of these models can be used to construct the demand relationships
required for a benefits analysis of water quality changes.
It is also important to recognize that the structure of the model provides
sufficient information to permit efficient estimation of the role of site attributes
for the parameters of site demand. To illustrate this point, consider a general
statement of the site demand model :
Y. = Xjp. + e. , (7.21)
where
Y. = N x 1 vector of the measures of the quantity demanded for the
1 ith site's services by each of N individuals
X. = N x K matrix of demand determinants for the N sampled users
1 of the ith site
p. = K x 1 parameter vector for the ith site
e. = N x 1 vector of stochastic errors for the ith site.
7-18
-------
Operationalizing the theoretical specification given in Equation (7.20) is equiv-
alent to assuming that variations in the vector of parameter estimates, p., can
be explained by the attributes of each recreation site, as in Equation (7.22):
p. = 6A. , (7.22)
where
6 = K x M matrix of parameters describing the effects of site
attributes on the parameters of the site demand equations
A. = M x 1 vector of the M characteristics of the ith site.
The specification for the determinants of site demand parameters will
affect the form of the efficient estimator of this two-component model. Under
the present specification, a two-step estimation scheme can be considered.
The first would involve the estimation of each site demand equation. Assuming
there are S sites, the process yields S vectors of estimates for each of the
parameters in the p. vector. Consider the ith such estimate. If e. is classi-
cally well behaved, the ordinary least-squares estimate, p., of p. will be un-
biased. It can be written as:
p. = (X.T X.)"1 XjT Y. . (7.23)
Or, substituting for Y. from Equation (7.21) yields:
p. = p. + (X.T X.)"1 X.T e. . (7.24)
Because p. is not observed, it is necessary to consider the use of estimates
I A
in its place. The ordinary least-squares estimate, p., is one such possibility.
If the model given in Equation (7.21.) has classically well-behaved errors and
A
nonstochastic independent variables as determinants, p. is the best linear
unbiased estimate of p}. Substituting for p. in Equation (7.22) using Equation
(7.24) provides the basis for a second-step estimator:
p. = p. - (X.T Xj)"1 X.T e. = 6A. . (7.25)
Rearranging terms yields:
Pj = 6A. + (X.T X.)"1 X.T e. . (7.26)
Equation (7.26) clearly suggests that, even if E(e.2) = o2 for all sites (i.e.,
i = 1 to S), efficient second-stage estimates require a generalized least-
squares estimator. That is, the model given in Equation (7.26) must be
estimated taking into account the relative precision of estimation of the Ps
7-19
-------
vector across sites. This will be given in each case by the corresponding
diagonal elements of Equation (7.27):
E(p. -
. - p.)T =
X.)
'1
(7.27)
The (X. X.) will not be identical across sites, even if the error variances
are constant and equal.
Unlike many instances, the nonspherical errors in this framework provide
a consistent estimate of the covariance matrix needed for generalized least-
A
squares estimation of the models in terms of pjt These estimates are contained
in the ordinary least-squares estimates of the respective parameter estimates'
covariance matrices (i.e., Equation [7.27]).
The generalized travel cost model can be efficiently estimated with a
.two-step procedure. Each site demand model is estimated with ordinary
least-squares (ignoring for the moment any potential simultaneity introduced
by the onsite time costs variable). The estimated parameters in these models,
together with their estimated variances, provide the basis for the second-step,
generalized least-squares estimates of the role of site attributes as determi-
nants of the individual demand parameters. If the jth member of p. for
i = 1, 2, ..., S, if the vector of these estimates is b. (an S x 1 vector of
the ordinary least-squares estimates for the jth parameter in the original p.
vector), and if a..2 is the corresponding diagonal element for a.2 (X. X.)~ ,
the generalized least-squares estimator of 6. (the sth row of 6) is given as
follows: '
where
" T T~
e.T = (A'Z
0222
-
AZ . ,
(7.28)
ss
A = SXM matrix of A. for each of S sites .
This estimator is somewhat different from that described by Saxonhouse
[1977]. However, the overall logic is completely parallel. The two generalized
7-20
-------
least-squares estimators differ in two respects. Saxonhouse [1977] assumes
that the first stage models will be jointly estimated with a Zellner [1962]
seemingly unrelated regressions estimator. This approach is more efficient
than ordinary least-squares estimates of the individual equations when there
is contemporaneous correlation between the stochastic errors across the equa-
tions and when the independent variables in all models together are not
highly correlated.* As originally formulated, the Zellner estimator maintains
that there is an equal number of observations for all models. While Schmidt
[1977] has developed variations on the estimator that relaxes this assumption,
there is no reason in this application to expect contemporaneous correlation
between the errors of the site demand equations. Each will be based on inde-
pendent surveys of users with little prospect that the same individuals would
use more than one site. In the absence of this contemporaneous correlation,
there is no advantage to the Zellner estimator. It is identical to the ordinary
least-squares estimates for each equation.
The second distinction arises in the specification of the covariance struc-
ture for the second step estimates. Saxonhouse's model assumes that Equation
(7.22) includes a stochastic error. By maintaining that these errors are
independent of the site demand errors, it is possible to develop consistent
estimates of the required covariance matrix using the residuals from ordinary
least-squares estimates of the second-step models. Saxonhouse's approach
can be viewed as a generalized random coefficient model because the parame-
ters of the site demand models are treated as random variables. However,
the observed variation in these parameters (across sites) arises from both
systematic (i.e., the differences in each site's characteristics) and random
influences. This interpretation has been avoided here in preference for a
framework that treats the demand parameters as constants that change with
site attributes. Because the true parameters are unobservable, estimates of
them must be used to determine the role of these attributes. Thus, random
influences enter the framework through the estimates of these parameters and
not as an inherent component of the demand model.
In summary, it has been argued that it is possible to develop a theoreti-
cally consistent method for determining the effects of a recreation site's char-
acteristics on the features of the demand for that site's services. Moreover,
the framework developed here does not require information on all recreation
sites considered by each potential user. This is an important distinction be-
tween the approach developed here and the Brown-Mendelsohn [1980] hedonic
travel cost model. Equally important, it is possible, using a straightforward,
two-step estimation procedure, to provide efficient estimates of the model.
It should be acknowledged that the approach presented here is not new.
Freeman [1979a] suggested such a scheme (without explicit consideration of
*Of course, it is important to recognize that the models discussed here
may be biased as a result of the assumption that all sites' services can be
transformed into common units using conversion functions in terms of their
respective attributes. This framework maintains that, after adjustment for
these characteristics, all sites are perfect substitutes in the production of
recreation service flows.
7-21
-------
the estimation problems) as one of a number of ad hoc approaches to treat-
ing water quality effects in modeling the demand for recreation sites. This
framework has extented Freeman's suggestion by demonstrating that it is not
ad hoc. Rather, it is completely consistent with a household production frame-
work of recreation participation patterns and with the theory of adjusting
quantity and price indexes for quality changes in goods and services.
7.4 SOURCES OF DATA
The 1977 Nationwide Outdoor Recreation Survey was conducted by the
Heritage Conservation and Recreation Service as part of the Department of
Interior's mandate to periodically develop National Recreation Plans. In con-
trast to past recreation surveys, which only included a general population
component, the 1977 survey included general population and site-specific user
surveys.
The Federal Estate Survey component of the survey, the primary basis
of this study, consists of interviews with recreationists at each of a set of
recreation facilities. All federally owned areas with public outdoor recreation
were considered to comprise the Federal Estate, and sites were chosen on a
basis of specific agency control. The majority of interviews were conducted
in areas managed by the National Park Service, the National Forest Service,
the U.S. Army Corps of Engineers, and the Fish and Wildlife Service. Each
agency was then stratified by Federal Planning Regions, and areas were
randomly chosen with weight given to annual visitation in 1975.
Interviewing time at each site was based on visitation, which also deter-
mined the number of interviews. The final Federal Estate Survey contains
13,729 interviews over 155 recreation areas. Information collected included
socioeconomic characteristics, current outdoor recreation activities, and atti-
tudes toward recreation for each respondent. Data requirements for develop-
ing travel cost models that describe demand for individual recreation sites are
met by the Federal Estate Survey.
Given that the scope of this study is water-based recreation and that
the analysis requires detailed descriptions of the activities at each site, only
U.S. Army Corps of Engineer sites were chosen for modeling. These 46 sites
also ensured consistent management of recreation activities. Three were elim-
inated from the analysis because of data inconsistency or ambiguous interview
site locations.
A number of the sites selected for analysis from the Federal Estate Survey
had observations with incomplete information. Rather than being eliminated
from the sample, these observations were classified according to whether or
not the missing information affected either the measurements of the use of the
relevant recreation sites or the travel and time costs of that use versus the
socioeconomic characteristics of the individuals involved. Observations that
did not permit evaluation of recreation choices (i.e., those missing the use
and travel information) were eliminated. The remaining incomplete observations
were replaced by the mean values of the relevant variables at that site because
the demand models were estimated at the site level. This procedure corre-
sponds to the zero-order method for treating missing observations.
7-22
-------
Section 7.6 discusses the results of using regression diagnostics to
evaluate the sensitivity of each site's estimates to sample composition. In
addition to gauging the sensitivity of the estimates to the assumptions of our
models, this index also provided a means to evaluate the implications of the
procedures used for missing observations.
Several variables in the Federal Estate Survey were reported by discrete
intervals. Answers to questions concerning time spent at site, number of
visits to site, travel hours to site, and annual income were treated as contin-
uous variables. In all cases the interval's midpoint was used. Open-ended
intervals were converted using the previous interval, with the difference
between the previous interval's midpoint and minimum value added to the
open-ended minimum value.
One component of the model described in Section 7.3 is the travel cost of
a trip, which is defined as the number of miles traveled multiplied by a per
mile cost. An independent estimate of travel cost was developed by measur-
ing each respondent's actual road distance traveled to a site based on his
reported zip code. All distances were calculated with the Standard Highway
Mileage Guide [Rand McNally, 1978], which lists road miles between 1,100
cities. National interstate highways and primary roads were used in all calcu-
lations. Other routes were used only for the distance to the nearest primary
road. In cases where cities have multiple zip codes, the center of the city
was used as the origin.
The second part of the travel cost calculation requires a per mile cost of
a trip. The marginal cost of operating an automobile in 1976 is estimated to be
approximatley $0.08 per mile. This estimate is based on costs of repairs and
maintenance, tires, gasoline, and oil as reported by the U.S. Census Bureau
in the U.S. Statistical Abstract [1978]. Mileage costs for operating an average
automobile were then calculated by using the round trip miles to the site multi-
plied by $0.08. This assumes that the respondent drove directly to the site
using the routes in the Standard Highway Mileage Guide. Unfortunately,
information was not available on the primary purpose of the respondents' trip
or further driving plans.
The Federal Estate Survey includes annual household income of respond-
ents but does not indicate any hourly wage rate. Because the use of reported
income in calculating opportunity cost of time precludes determination of
income's role in the site demand models, an independent estimate of each
individual's wage rate is important to a complete specification of the model.
A hedonic wage model estimated from the 1978 Current Population Survey
(CPS) was used to derive these estimates. This model specifies the market
clearing wage rates to be a function of individual-, job-, and location-specific
characteristics. The specific model was developed by Smith [forthcoming,
1983]. By substituting each individual's characteristics (including location-
specific and occupation-specific variables), predicted wage rates were derived.
Equation (7.29) provides a general statement of the procedure, with X.. the
determinants of the wage rate: 'J
7-23
-------
N ~
W = exp ( Z B.X..) , (7.29)
1 = J IJ
where
W. = the predicted wage rate for the ith individual
ys.
B. = the estimated coefficient of the jth variable
X.. = the ith individual's value of the jth variable
N = the number of explanatory variables.
Explanatory variables usually include age, sex, education, occupation, and
various other job- and location-specific characteristics.
The estimates made in this study should be regarded as proxy measures
for actual wage rates. Since the wage model is a semilog, the predictions
can be expected to understate the estimated conditional expectation for the
wage rate. While Goldberger's [1968] proposed unbiased estimator for this
conditional expectation would be superior for large degrees of freedom (the
CPS sample contained 9,077 for males and 7,067 for females) and for a small
error variance of the estimated model, the bias in this study's estimates will
be small. A 10-percent discrepancy would be a generous outer bound on the
magnitude of the percentage difference between the direct predictions of these
wage rates and the estimates based on Goldberger's method. Indeed, in most
large sample applications (see the examples in Goldberger [1968] and Giles
[1982]), the actual differences are under 5 percent. Thus, despite this
limitation, these estimates provide a better set of proxy measures for wage
rates than the available alternatives since they take explicit account of indi-
vidual and job characteristics. In specifying and estimating the wage model,
consideration was also given to measures of job risks, air pollution, climate,
crime, access to cultural and sporting activities, and local labor market condi-
tions.
The nominal wage model includes a cost-of-living variable as one of the
determinants of wages. Smith used the Bureau of Labor Statistics budget-
cost-of-living index for this variable. In the Standard Metropolitan Statistical
Areas (SMSAs) where the index was not known, information available for 27
SMSAs was used to model the determinants of variations in the cost of living.
As shown in Equation (7.30), the index, C., was related to population
density, D.; the size of the SMSA population in 1975 in thousands, POP.; and
the percent of the population under 125 percent of the poverty standard,
POOR.. The t-ratios for the hypothesis of no association are shown in paren-
theses :
C. = 111.81 + 0.005 D. - 0.001 POP. - 1.30 POOR
J (37.73) (7.38) J (-2.40) J (-4.36) J
R2 = 0.787
F(3, 23) = 28.34 . (7.30)
7-24
-------
The Federal Estate Survey does not directly identify respondents' SMSA.
Thus, a cost of living variable was generated at the State level to minimize
computation cost. This index was calculated as the average of the SMSAs
within each State, avoiding the need to match each respondent to an SMSA.
The estimated 1977 nominal wages for the recreationists at each site were
developed using the equations in Table 7-1. The characteristics necessary
for the model were generally available in the Federal Estate Survey, and
classifications between the model and the survey were compatible. Problems
do arise, however, for respondents who were not labor force participants at
the time of the survey. For example, students and housewives could not be
considered in the sample used to estimate the hedonic wage model. In these
cases, the wages were treated as an opportunity cost estimated to be the
mean value by sex of the predicted wage rates in the recreation survey.
Table 7-2 provides a summary of predicted hourly wage rates by income and
occupation of the respondents. The predicted Wage rate is used to calculate
the opportunity cost of both onsite time and travel time. For at least two
reasons, there are substantial differences in these estimates for the upper
income members of the sample. The first stems from the coding of the wage
measure in the Current Population Survey. Specifically, the reporting format
limits the reported usual weekly earnings (the basis for the hourly wage
rate—usual weekly earnings divided by usual hours worked) to $999. Thus,
there is censoring in wages for individuals above approximately $52,000 per
year. The second reason is that family income can reflect the effects of
nonwage income and the impact of dual earner households. Unfortunately,
the extent of these influences cannot be sufficiently determined to improve
wage rate estimates for individuals in these higher income households.
The U.S. Army Corps of Engineers maintains the Recreation Resource
Management System for evaluation and planning. Data from this system are
compatible with the sites chosen for the Federal Estate Survey and have been
available since 1978. Information is collected annually on each water resource
project with 5,000 or more recreation days of use. For 1978, this information
included financial statistics, facilities available, natural attributes, recreation
participation, and number of employees.
The Recreation Resource Management System is used to define attributes
of the 43 Federal Estate Survey sites. Attributes of an area considered
include land area, shore miles, pool elevation, the number of multipurpose
recreation areas, and facilities provided. Table 7-3 provides descriptive
statistics for both the characteristics of the sites and of a selected set of
variables for the survey respondents at these sites.
The National Water Data Exchange (NAWDEX) is a membership of water-
oriented organizations and is a major source of water quality information.
The NAWDEX system is under the direction of the U.S. Geological Survey,
and its primary function is to exchange data from various organizations.
Major sources of information are usually State agencies, the U.S. Geological
Survey, the U.S. Army Corps of Engineers, and the U.S. Environmental
7-25
-------
Table 7-1. Hedonic Wage Models
Male
Female
Variable
Coefficient
t-statistics
(of no association)
Coefficient
t-statistics
(of no association)
Intercept
Education
Education squared
Experience
Experience squared
Race
Veteran
Unemployment
Professional
Managerial
Sales
Clerical
Craftsman
Operative
Transport equipment
Nonfarm labor
Service worker
Injury rate
Cancer
TSP
Household head
Union member
OJT x Experience
Crime rate
Percent sunshine
Dual job holder
Know x Cancer
Log (cost of living index)
0.631
0.030
0.001
0.031
-0.001
0.113
0.035
-0.012
0.086
0.142
-0.0003
-0.099
0.015
-0.149
-0.118
-0.131
-0.251
0.011
0.299
0.0007
0.229
0.178
-0.002
0.000005
-0.002
-0.042
3.77
0.559
8.71
4.01
3.45
25.83
-22.35
8.75
3.50
-3.51
2.76
4.48
-0.01
-3.01
0.48
-4.47
-3.35
-3.87
-7.70
10.40
2.93
2.31
16.75
17.52
-1.64
1.89
-2.31
-1.75
4.58
7.22
0.179
0.028
0.001
0.018
-0.0002
-0.024
—
0.002
0.563
0.521
0.199
0.390
0.445
0.235
0.366
0.199
0.166
0.012
0.105
0.0003
0.069
0.191
-0.001
-0.000008
0.0001
-0.025
5.727
0.606
2.03
2.61
2.15
15.91
-11.83
-1.73
—
0.57
19.17
16.15
6.30
15.38
8.68
8.09
5.34
3.97
6.26
7.67
0.86
0.97
6.01
13.81
-0.54
2.39
0.12
-0.81
4.24
6.56
R2 = 0.47
degrees of freedom = 9,077
F ratio = 292.92
R2 = 0.33
degrees of freedom = 7,067
F ratio = 135.52
SOURCE: Smith [1983].
aThe variable definitions are as follows:
(1) Education—measured as the years corresponding to the highest grade of school attended (this variable
is entered in linear and quadratic terms).
(2) Experience—measured using the conventional proxy of age minus years of education minus six (this
variable is entered in linear and quadratic terms).
(3) Socioeconomic qualitative variables—dummy variables for race (white = 1), sex (male = 1), veteran
status (veteran = 1 and relevant only for males), member of a union (yes = 1), head of the household
(yes = 1), and dual job holder (yes = 1).
(4) Occupational qualitative variables—dummy variables to define the respondents occupation as: profes-
sional, managerial, sales, clerical, craftsman, operative, transport equipment operator, nonfarm labor,
or service worker.
(5) Cancer—index of exposure to carcinogens.
(6) TSP—average suspended particulates in 1978.
(7) OJT--on-the-job training program available.
(8) Know—relative number of workers within an industry covered by collective bargaining with health and
safety provisions.
The omitted occupational category was defined to correspond to a composite of those occupations that might
lead the estimated hourly wage to understate actual earnings. The omitted occupatfons were farm laborers
and private household workers.
A measure of price uncertainty was constructed to provide some basis for adjusting the experience measure
to reflect the different levels of provision of on-the-job training (OJT) across firms. To evaluate the impor-
tance of these effects, price uncertainty was measured as the unexplained variation (i.e., 1-R8) for linear
trend models fit to monthly wholesale price indexes for each of 14 product categories for each year over
the period 1976 through 1978. After evaluating each year's index, 1977 was selected for th.s analysis.
The indexes were assigned to individuals according to their industry of employment in an attempt to match
products as closely as possible. The variable was entered as an interaction term with experience.
7-26
-------
Table 7-2. Summary of Predicted Hourly Wage Rates (1977 $)
Total
sample Male Female
Overall mean 5.44 6.27 4.34
Number of observations 3,460 1,971 1,489
Mean by annual household income
Under 5,999 5.08 5.79 4.06
6,000 to 9,999 4.92 5.49 4.10
10,000 to 14,999 5.32 6.01 4.38
15,000 to 24,999 5.72 6.70 4.39
25,000 to 49,999 5.98 7.17 4.65
50,000 or more 5.73 6.53 4.65
Mean by occupation of respondent
Professional, technical, and
kindred workers
Farmers
Managers, officials, and proprie-
tors
Clerical and kindred workers
Sales workers
Craftsmen, foremen, and kindred
workers
Operatives and kindred workers
Service workers
Laborers, except farm and mine
Retired widows
Students
Unemployed
Housewives
Other
No occupation given
7.05
5.15
7.17
4.34
5.18
5.89
4.97
4.11
4.44
5.92
5.30
5.46
4.37
5.71
5.49
7.89
5.71
7.74
5.94
6.24
6.05
5.15
4.71
4.74
6.27
6.27
6.27
6.27
6.27
6.27
5.65
2.75
4.94
4.10
3.29
4.31
3.56
3.18
3.11
4.34
4.34
4.34
4.34
4.34
4.34
aTotal number of observations is 3,282.
Total number of observations is 3,460.
Protection Agency (EPA). All water quality data used in the analysis were
retrieved from NAWDEX in a series of steps. Collection of useful water
quality data was completed by identifying potential monitoring stations and by
then obtaining actual data. Potential monitoring stations were identified by
defining the recreation area in terms of latitude and longitude. A general
retrieval was then obtained that listed station name, location, parameter
collected, years of data collection, and agency responsible for the data collec-
tion.
7-27
-------
I able /-3. The characteristics of the Sites and the Survey Respondents
Selected from the Federal Estate Survey
Characteristics of survey respondents
Site characteristics
Property Recreation
Project name code days
Allegheny River
System, PA
Arkabutla Lake, MS
Lock & Dam No. 2
(Arkansas River
Navigation
System), AR
Beaver Lake, AR
Belton Lake, TX
Benbrook Lake, TX
Berlin Reservoir, OH
Blakely Mt. Dam,
Lake Ouachita, AR
Canton Lake, OK
Clearwater Lake, MO
Corded Hull Dam &
Reservoir, TX
DeGray Lake, AR
Dewey Lake, KY
Fort Randall, Lake
Francis Case, SD
Grapevine Lake, TX
Greens Ferry Lake, AR
Grenada Lake, MS
Hords Creek Lake, TX
Isabella Lake, CA
Lake Okeechobee and
Waterway, FL
Lake Washington Ship
Canal, WA
Leech Lake, MN
Melvern Lake, KS
Millwood Lake, AR
Mississippi River Pool
No. 3, MN
Mississippi River Pool
No. 6, MN
Navarro Mills Lake, TX
New Hogan Lake, CA
New Savannah Bluff
Lock & Dam, GA
Norfork Lake, AR
Ozark Lake, AR
Perry Lake, KS
Phllpott Lake, VA
Pine River, MN
Pokegama Lake, MN
Pomona Lake, KS
Proctor Lake, TX
Rathbun Reservoir, TX
Sam Rayburn Dam &
Reservoir, TX
Sardls Lake, MS
Waco Lake, TX
Whitney Lake, TX
Youghlogheny River
Lake, PA
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
327
328
329
330
331
332
333
334
335
336
337
338
339
340
343
344
345
.
2,011,700
343,700
4,882,600
2,507,000
1,978,000
1,179,000
2,104,300
3,416,500
888,000
2,167,900
1,659,700
1,116,800
4,756,000
5,139,100
4,407,000
2,553,900
359,500
1,489,200
2,894,584
712,900
950,600
2,034,600
2,042,300
1,323,700
645,500
1,111,500
335,200
207,600
3,066,500
1,102,000
3,388,000
1,454,900
1,615,100
948,300
1,460,400
975,200
2,332,200
2,728,700
2,488,900
3,371,600
1,976,400
1,122,600
Shore
miles
.
134
96
449
136
37
70
690
45
27
381
207
52
540
60
276
148
11
38
402
80
316
101
65
37
55
38
44
32
380
173
160
100
119
53
52
27
156
560
110
60
170
38
Area
acres
.
52,549
32,415
40,463
30,789
11,295
7,990
82,373
19,797
18,715
32,822
31,800
13,602
133,047
17,828
45,548
86,826
3,027
15,977
451,000
169
162,100
24,543
142,100
20,350
11,292
14,286
6,162
2,030
54,193
39,251
41,769
9,600
22,177
66,542
12,301
15,956
36,072
176,869
98,590
21,342
53,230
4,035
Predicted
wage rate
X
S.45
5.23
5.24
5.59
5.52
5.00
5.44
5.24
5.09
5.43
5.43
5.17
5.83
5.43
5.20
5.15
5.13
5.26
5.64
5.38
6.26
5.90-
5.69
5.49
6.36
5.79
5.16
5.57
5.28
5.65
5.02
5.52
5.33
5.95
5.70
5.42
5.49
5.74
5.32
5.41
5.46
5.25
5.56
a
1.65
1.45
1.03
1.70
1.51
1.21
1.24
1.53
1.54
1.38
1.58
1.58
2.10
1.69
1.58
1.45
1.56
1.42
1.48
1.20
2.07
1.40
1.65
1.87
2.23
.42
.41
.28
.13
.61
.22
.48
.55
1.80
1.46
1.36
1.63
1.56
1.35
1.31
1.25
1.29
1.59
Household Income
X
15,667
13,184
10,409
18,150
17,279
19,135
16,459
17,144
17,392
17,943
15,491
19,235
18,021
20,696
19,309
15,890
9,199
16,263
15,938
13,849
16,686
18,886
18,087
18,630
29,571
19,589
13,739
18,954
12,609
17,667
12,654
16,565
14,268
20,097
16,816
17,265
17,510
20,543
19,515
13,141
16,396
18,688
16,682
a
8,625
8,974
3,991
9,946
11,913
10,065
10,161
9,524
10,553
8,456
9,215
10,612
9,559
11,705
10,992
8,562
4,833
9,699
11,445
9,541
5,815
10,986
9,015
1,319
10,895
10,693
4,652
11,270
9,414
8,889
7,568
6,925
6,668
9,370
9,476
7,330
11,167
7,473
11,331
7,223
12,454
11,651
11,051
Visits
X
2.6
5.4
6.8
3.5
6.0
2.3
5.2
4.3
4.6
4.0
5.7
4.8
2.4
3.3
6.3
4.7
6.4
4.4
3.3
4.1
3.3
2.5
4.3
5.6
3.0
4.8
4.6
4.0
5.8
3.2
4.9
4.7
5.8
2.1
3.3
5.4
5.4
4.3
4.1
6.5
6.9
5.0
5.4
a
2.5
2.7
2.0
3.0
2.8
1.2
2.9
2.8
3.2
2.7
2.9
2.7
2.0
3.1
2.6
3.0
2.6
3.0
2.5
3.0
3.0
1.8
3.0
3.0
2.4
3.0
2.8
3.1
2.7
2.5
3.0
2.7
2.6
1.4
2.7
2.8
2.9
2.9
2.7
2.3
2.2
2.8
2.9
(T+M) Cost
X
45.19
20.04
3.04
94.55
33.18
30.23
21.15
45.39
32.30
50.51
29.65
42.04
90.75
100.29
38.45
54.16
24.57
39.46
55.59
24.91
98.63
104.08
31.48
37.62
99.20
52.23
27.68
34.10
18.65
94.89
58.71
28.79
26.09
69.80
100.63
25.38
46.08
41.78
40.23
36.08
33.02
35.40
24.67
a
28.30
27.94
13.01
88.64
52.35
58.93
26.63
49.31
22.97
42.24
34.70
43.42
122.44
93.59
64.32
70.00
32.90
48.25
45.54
11.03
130.14
84.35
29.39
55.21
79.14
55.19
30.29
14.55
23.78
59.65
98.54
24.02
46.00
50.54
122.30
23.33
40.96
29.18
31.90
42.17
45.10
38.03
9.48
Miles9
X
106
45
55
266
67
73
40
121
95
140
60
115
243
260
92
154
65
108
127
76
338
268
84
90
196
141
61
72
37
268
199
79
47
178
376
65
109
96
85
123
99
96
47
a
57
90
33
296
142
223
130
139
99
192
87
164
519
295
217
306
165
170
100
258
605
313
137
176
288
240
70
29
77
75
433
109
100
188
590
115
103
41
74
234
263
195
58
Number
of
obser- t
vations
69
61
41
226
53
46
96
91
74
74
104
49
46
50
92
217
75
54
48
30
37
48
45
53
49
70
42
41
39
42
52
28
38
75
68
31
52
31
67
205
61
201
31
aOne-way distance to the site.
bNumber of observations are based on the final models estimated for site.
NOTES: X is the arithmetic mean.
o is the standard deviation.
(T+M) cost is the sum of vehicle and time-related costs of a visit.
-------
One major problem in the data collection process is the identification of
appropriate monitoring sites. Ideally, monitoring stations should be located in
the area where recreation occurs. Monitoring sites could only be identified
by obscure station names. Furthermore, information is not available according
to area names used by survey respondents. Proximity of a water quality
monitor to actual recreation could not be determined.
Monitoring sites that could be identified as relevant were then chosen,
and the actual water quality data were obtained through NAWDEX. Several
problems are inherent in this type of data collection. A brief discussion of
the data collection process and some problems encountered follow. The reader
is referred to Appendix E for a more detailed discussion of water quality.
Water quality parameters were selected on a basis of previous use and
availability among sites. The parameters collected are temperature, pH,
dissolved oxygen, biological oxygen demand, turbidity, nitrates, phosphates,
fecal coliform, dissolved solids, flow, and Secchi-disk transparency. Of the
43 sites, 16 had no data due to a lack of known monitoring sites.
Actual water quality data were collected for 27 sites for the years 1972
to 1981. Most of these sites were missing information for the year the survey
was completed. As a result, calculations were carried out using 1972 to 1981
data. Monthly means for each site were calculated for June through September.
An overall mean was also calculated using the four monthly means. In cases
where sites were completely missing a parameter, the mean for all sites was
used.
*.
Individual parameters and indexes are used in the analysis, including
both monthly values and a summer average. Index methods include the
National Sanitation Foundation and the Resources for the Future measures.
Linear combinations of parameters were also tested, although the degree of
correlation between parameters was regarded with caution.
The treatment of missing values for these variables led to a lack of
variation between sites. This is caused by two factors. First, the averaging
of several years distorts the actual water quality for a particular year.
Consideration is not given to improvements or deterioration of water quality.
Secondly, replacing missing observations with the means smooths out the
variation between sites. Any predictions of water quality benefits with the
travel cost model will become more reliable as missing observations are replaced
with actual data.
The choice of parameters to be measured at a monitoring site varies
according to a water body's local characteristics and the agency collecting the
sample. This inconsistency in data collection may cause problems when the 43
Army Corps of Engineers' areas are compared. For example, if suspended
solids are not considered a problem in an area, they are not likely to be
measured. Consequently, several parameters were not available in all areas
or during the appropriate time.
In summary, three generally compatible data sources were used. Data
obtained from each source are consistently defined across sites.
7-29
-------
7.5 EMPIRICAL RESULTS FOR SITE-SPECIFIC TRAVEL COST MODELS
The theoretical model of the consumer's recreation decisions identified
three aspects of the process that may influence the use of the travel cost
model for an analysis of the benefits (or costs) of a change in the attributes
of a recreation site. Two of these aspects arise in defining the relevant
measure of site usage and the associated cost to the individual for a "unit" of
the site's services (assuming an ideal quantity index could be derived). In
the formal model of household choice, the individual was able to produce
additional units of the recreation service flow with more trips of a given
length or by increasing the time spent onsite during a fixed number of trips.
The household production framework did not specify these choices as perfect
substitutes, but it did admit the possibility of substitution. This type of
input substitution is plausible because the time horizon for production has
been interpreted to be the recreation season. This specification of the prob-
lem implies that the number of visits to a given site and the times spent
onsite per visit will be jointly determined variables. Indeed, the demand
model for visits (i.e., Equation [7.7]) was expressed as a reduced form
equation. Of course, the specific analytical model simplified the issues
involved by assuming the time spent onsite was the same for all the visits in
a given season. Actual behavior is more complex, with the prospects for
different amounts of time spent onsite for every visit. There are several
aspects of this problem described below in greater detail. The discussion
portrays the treatment of each issue in this analysis and how this treatment
compares with earlier literature.
The second aspect of modeling an individual's recreation choices arises
in the definition of the cost of a visit to a given recreation site. The analyti-
cal model indicated that this cost would be composed of the costs of transpor-
tation to the site (i.e., the product of roundtrip mileage and a vehicle operat-
ing cost per mile) and the opportunity costs associated with the time spent
traveling to the facility. As noted earlier, the appropriate definition of these
opportunity costs has been addressed in several papers in the past literature.
The model identifies the cost as r and does not attempt to relate it to the
individual's wage rate. Of course, in practice r is unknown and requires
estimation. Since the treatment of this variable has important implications for
the estimated costs of a trip, the issues involved in this study's modeling
choices are detailed below.
Finally, the third aspect of the representation of recreation decisions
stems from this chapter's overall objective, which is to evaluate the influence
of site characteristics on the demand for the services of a recreation facility.
As developed in Section 7.3, some analytical restrictions on the role of site
attributes for the production of recreation service flows, together with a
diversity of these features across sites, provide sufficient information to
estimate the relationship between each site's demand model and its attributes.
To estimate this relationship, however, requires the adoption of a common
demand specification for all the individual site demand equations. While the
sample sites provide the ability to engage in an approximately comparable set
of recreation activities, this is not a sufficient reason, in itself, for expecting
the site demand models to be comparable. Thus, before turning to the
generalized least-squares models for explaining the variation in an individual
7-30
-------
site's estimated demand parameters, the implications of using a common specifi-
cation must also be considered. To adequately treat these three issues, a
fairly detailed set of statistical analyses of site demand models was undertaken.
The explanation of these results will be developed in this and the next
two sections of this chapter. This exposition begins with a more detailed
discussion of the conceptual dimensions of each of these issues in the first
three subsections of this section. The ordinary least-squares estimates for
the general model applied to all 43 sites follow that discussion. The remainder
of this section discusses the implications of using conventional pretesting
criteria for selecting individual specifications for each site demand, as well as
the influence of different approaches for treating the opportunity cost of
travel time to each site. Section 7.6 discusses the results of the analysis of
onsite time and visits within a simultaneous equation model and the several
specific statistical issues that arise for travel cost models because of the
nature of the available measures of site usage. The final component of the
model is developed in Section 7.7, where the results of the generalized least-
squares model for the determinants of the features of recreational site demand
equations are presented.
7.5.1 The Treatment of Onsite Time
Ideally, the measurement of site demand models would involve both the
number of trips to a particular site and the time spent onsite for each trip.
Unfortunately, in practice this information is rarely available.* The source of
data for this analysis (the Federal Estate Survey) includes information on the
amount of time spent at the site during the trip in which the respondent was
interviewed and not the corresponding information for all trips taken during
the season. Thus, any attempt to deal with the relationship between onsite
time and visits will require further assumptions.
There have generally been two treatments of onsite time in the recreation
demand literature. The first of these corresponds to the most common practice
in the literature—onsite time is assumed to be constant across trips and
across individuals. In this case, the number of visits is a consistent index
of the use of a site's services. With this approach, the onsite time (or cost)
term is dropped from the travel cost model (and thus the wage rate would not
enter Equation [7.7]).t
*Brown and Mendelsohn [1980] is one notable exception.
tThis practice is, strictly speaking, not correct. Even though onsite
time is constant and not considered a choice variable, it does influence the
cost of a trip (see Equation [7.4]). Moreover, it cannot be treated as a
constant displacement to the demand model's intercept because the opportunity
costs of time can be expected to vary across individuals.
We considered a role for onsite time under the assumptions that adjust-
ment for simultaneity was unnecessary and that the results were uniformly
unsatisfactory. Without an explicit recognition of the simultaneity between
visits and onsite time costs, ordinary least-squares estimates of the role of
onsite time costs would lead to the conclusion that these costs were unimpor-
tant influences on the demand for each site's services.
7-31
-------
The second approach specifies the travel cost demand function for each
site to include the costs of onsite time for the trip in which the individual
was interviewed. This case implicitly assumes that the time spent onsite is
constant for all trips but may well be different across individuals. Thus, the
empirical model corresponds to the theoretical structure developed at the
outset of this chapter. The first approach corresponds to the basic model
and is reported in this section. The second approach is used to gauge the
implications of ignoring onsite costs. These results are summarized in Section
7.6.
7.5.2 The Opportunity Cost of Travel Time
As noted earlier, it has often been argued that the opportunity cost of
travel time is less than the wage rate. If this cost is known, theory suggests
that travel costs and the cost of travel time have equivalent effects on the
demand for the site's services (i.e., their parameters, in a linear demand
model would be equal). In the absence of information on these opportunity
costs, and if it is possible to assume they are a constant fraction of every
individual's wage rate, separate effects can be identified for travel cost and
the cost of travel time. The relationship between the estimated parameters
provides one basis for estimating the constant—essentially the McConnell-
Strand [1981] approach. Of course, to apply this approach, independent
estimates of roundtrip distance to the site and travel time must be available.
Since few travel cost studies have had access to this type of information,
many studies accept Cesario's [1976] suggestion that the opportunity cost of
travel time is a multiple of the wage rate ranging from one-fourth to one-half
and use it in calculating the cost of a trip. In these cases, travel costs and
travel time are both based on roundtrip distance. Of course, the latter also
requires an assumed velocity of travel, a wage rate, and the Cesario constant
to estimate the opportunity cost of travel time.
Since the Federal Estate Survey reports travel time and the Zip codes of
each respondent's residential location, it was possible to develop independent
estimates of both components of the cost of a trip. Thus, tests for each
model evaluate the appropriate treatment of travel costs and the costs of
travel time. These tests simply translate the economic issues and ad hoc
practices into restrictions on the parameters of the site demand models.
7.5.3 Results for the Basic Model
Table 7-4 provides the ordinary least-squares estimates for the semi log
specification of our travel cost demand models. The general form for the
model is given in Equation (7.31) below:
ln(V.) = «0 + a^TC.+MCj) + a3 INC. + ef , (7.31)
where
V. = number of visits during the recreation season for the ith
1 respondent
7-32
-------
w
Table 7-4. Regression Results of General Model, by Site
LN VISITS = a0 + cri (T+M) COSTS3 + of3 INCOME6
Site
Allegheny River System, PA
Arkabutla Lake, MS
Lock and Dam No. 2 (Arkansas River
Navigation System), AR
Beaver Lake, AR
Belton Lake, TX
Benbrook Lake, TX
Berlin Reservoir, OH
Blakely Mt. Dam, Lake Ouachita, AR
Canton Lake, OK
Clearwater Lake, MO
Corded Hull Dam and Reservoir, TN
DeGray Lake, AR
Dewey Lake, KY
Ft. Randall, Lake Francis Case, SD
Grapevine Lake, TX
Greers Ferry Lake, AR
Grenada Lake, MS
Hords Creek Lake, TX
Isabella Lake, CA
Lake Okeechobee and Waterway, FL
Lake Washington Ship Canal, WA
Site
number
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
Intercept
0.53
1.58
2.31
1.61
1.69
1.83
1.40
1.70
1.77
1.51
1.86
1.79
0.42
1.32
1.80
1.48
2.04
1.73
1.26
1.68
0.96
(2.04)
(9.99)
(9.76)
(16.07)
(9.38)
(10.70)
(8.47)
(10.08)
(8.61)
(5.97)
(14.13)
(7.71)
(2.27)
(6.00)
(16.12)
(14.08)
(12.61)
(8.22)
(5.55)
(3.68)
(2.69)
T+M cost
-0.0005
-0.0093
-0.0125
-0.0066
-0.0052
-0.0054
0.0014
-0.0079
-0.0206
-0.0032
-0.0139
-0.0070
-0.0024
-0.0066
-0.0073
-0.0065
-0.0095
-0.0050
-0.0073
-0.0268
-0.0037
(-0.13)
(-3.09)
(-2.30)
(-12.77)
(-2.47)
(-4.11)
(0.43)
(-5.14)
(-5.28)
(-1.42)
(-6.00)
(-3.00)
(-2.95)
(-5.93)
(-8.80)
(-9.02)
(-4.36)
(-2.11)
(-3.15)
(-1.72)
(-3.79)
8.2 x
6.2 x
-1.8 x
-3.5 x
2.6 x
6.0 x
-4.1 x
-7.6 x
7.1 x
-1.0 x
-1.2 x
-6.9 x
2.0 x
7.5 x
8.5 x
8.4 x
-1.0 x
-2.1 x
7.9 x
1.9 x
1.7 x
Income
10'6
10'6
10'5
10'6
10~6
io-6
10'7
10'6
10'6
ID"5
10-8
10'5
10'5
10'6
io"6
10'6
10*5
10'5
10~6
ID'7
10'5
(0.74)
(0.67)
(-1.08)
(-0.78)
(0.29)
(0.80)
(-0.05)
(-0.98)
(0.86)
(-1.21)
(-0.01)
(-0.73)
(2.02)
(0.91)
(1.70)
(1.42)
(-0.68)
(-1.76)
(0.81)
(0.01)
(0.84)
R2
0.01
0.15
0.14
0.43
0.12
0.30
0.01
0.24
0.28
0.04
0.34
0.17
0.18
0.43
0.47
0.28
0.22
0.19
0.20
0.10
0.26
DF
66
58
38
224
50
43
93
88
71
71
101
46
43
47
89
214
73
51
45
27
41
F-
ratio
0.29
4.93
3.11
86.07
3.39
9.11
0.09
13.67
13.98
1.61
25.57
4.68
4.72
17.61
39.12
40.79
10.02
5.95
5.47
1.56
7.18
DF = Degrees of freedom. (continued)
aT+M represents the respondents' round trip cost. It Is composed of travel time cost (TCOST) and a constant per mile cost of operating
an automobile (MCOST).
bt-values of no association are shown in parentheses.
-------
Table 7-4. (continued)
Site
Leech Lake, MN
Melvern Lake, KS
Millwood Lake, AR
Mississippi River Pool No. 3, MN
Mississippi River Pool No. 6, MN
Navarro Mills Lake, TX
New Hogan Lake, CA
New Savannah Bluff Lock & Dam, GA
Norfork Lake, AR
Ozark Lake, AR
Perry Lake, KS
^ Philpott Lake, VA
** Pine River, MN
Pokegama Lake, MN
Pomona Lake, KS
Proctor Lake, TX
Rathbun Reservoir, TX
Sam Rayburn Dam & Reservoir, TX
Sard Is Lake, MS
Waco Lake, TX
Whitney Lake, TX
Youghiogheny River Lake, PA
Site
number
321
322
323
324
325
327
328
329
330
331
332
333
334
335
336
337
338
339
340
343
344
345
Intercept
0.87
1.30
1.43
1.33
1.41
1.66
1.04
1.88
1.13
1.66
1.50
1.90
0.81
1.44
1.54
2.06
0.77
1.46
1.81
1.95
1.41
0.29
(3.88)
(4.47)
(7.94)
(4.20)
(7.45)
(6.40)
(2.58)
(8.39)
(4.27)
(8.52)
(4.17)
(9.28)
(4.65)
(7.28)
(5.35)
(13.61)
(1.85)
(7.06)
(20.73)
(15.04)
(13.07)
(0.60)
T + M cost
-0.0022
-0.0079
-0.0081
-0.0057
-0.0074
-0.0057
-0.0040
-0.0067
-0.0047
-0.0046
-0.0042
-0.0087
-0.0017
-0.0033
-0.0058
-0.0134
-0.0015
-0.0094
-0.0030
-0.0006
-0.0025
0.0263
(-1.83)
(-1.66)
(-3.99)
(-4.62)
(-4.39)
(-1.39)
(-0.41)
(-1.44)
(-2.55)
(-4.44)
(-0.74)
(-4.40)
(-1.27)
(-4.46)
(-1.11)
(-7.50)
(-0.27)
(-2.83)
(-3.17)
(-0.32)
(-1.80)
(1.61)
3.5 x
4.1 x
1.8 x
4.7 x
1.3 x
-1.4 x
7.1 x
-9.8 x
9.3 x
-8.8 x
-1.0 x
-1.7 x
-6.4 x
-1.4 x
8.4 x
1.2 x
1.4 x
1.0 x
4.3 x
-7.4 x
3.2 x
1.7 x
Income
10~6
ID'6
ID'5
10'6
10'5
10'5
10'6
ID'6
ID'5
10'6
10'5
io-6
10'6
10'5
10'6
io-6
10'5
10'6
10'6
io-6
10"6
10'5
(0.37)
(0.32)
(2.14)
(0.54)
(1.53)
(-1.14)
(0.60)
(-0.70)
(0.79)
(-0.66)
(-0.68)
(-0.13)
(-0.91)
(-1.57)
(0.62)
(0.19)
(0.82)
(0.13)
(0.78)
(-1.25)
(0.72)
(1.55)
R2
0.07
0.06
0.25
0.34
0.22
0.06
0.01
0.06
0.14
0.31
0.03
0.36
0.04
0.24
0.13
0.54
0.02
0.11
0.05
0.03
0.02
0.14
DF
45
42
50
46
68
39
38
36
39
49
25
35
72
67
28
49
28
64
202
58
201
28
F-
ratio
1.68
1.37
8.26
11.67
9.68
1.33
0.23
1.25
3.30
11.18
0.41
10.03
1.39
10.36
1.35
28.39
0.34
4.10
5.22
0.93
1.80
2.35
DF = Degrees of freedom.
aT+M represents the respondents' round trip cost. It is composed of travel time cost (TCOST) and a constant per mile cost of operating
an automobile (MCOST).
t-values of no association are shown in parentheses.
-------
TC. = time costs of travel for the ith respondent, defined as
1 product of the estimated wage rate for the person (see
Section 7.6) and the roundtrip travel time
MC. = travel costs for the ith respondent
INC. = family income for the ith respondent
e. = stochastic error for ith respondent.
Several alternative functional forms were .considered. However, the
results uniformly favored the semilog form based on the ability to precisely
estimate the site demand parameters. Moreover, this specification is generally
selected in evaluations of functional forms for the travel cost model (see
Smith [I975a], Smith and Kopp [1980], and Ziemer, Musser, and Hill [1980]).
In general the implicit price (TC+MC) of a trip to the site is statistically
significant and correctly signed. There is a fairly large range for values for
the estimated parameters for the implicit price—ranging from -0.0005 to
-0.0139. Only one site exhibited a positive coefficient for the implicit price,
and in this case the coefficient would not be judged to be significantly differ-
ent from zero. In the balance of the models, 27 sites had coefficient estimates
that would lead to the judgment of a demand effect significantly different from
zero at least at the 5-percent level. The balance of the estimated price
coefficients is negative and in many cases is also statistically different from
zero at a higher significance level—i.e., 10 percent.
The effect of income is poorly measured in all of these models. In most
cases the parameter estimates would lead to the conclusion that income is not
a significant determinant of the demands for these sites. Indeed, in a number
of the models the estimated parameters were negative. However, these esti-
mated parameters would lead to the conclusion that income's effect was not
significantly different from zero.
At first, the lack of significance of income may seem surprising. How-
ever, when it is considered in comparison to other recreation applications of
the travel cost framework, it is more plausible. For the most part these sites
provide high-density camping, swimming, boating, etc. These are activities
where the participation decision and level of use decisions were either some-
what insensitive to family income or where income's marginal effect increased
and then decreased with increases in the level of income. Table 7-5 summar-
izes the role of income in the Cicchetti, Seneca, and Davidson [1969] analysis
of recreation participation decisions. Of course, it should be acknowledged
that these participation models are reduced form equations reflecting the
influence of both demand and supply influences (see Smith [I975a] and Deyak
and Smith [1978] for further discussion of these approaches). Nonetheless,
they provide some information based on the likely implications of the mix of
activities a site can support for the nature of the demand for that site's
services.
7-35
-------
Table 7-5. Summary of Cicchetti, Seneca, and Davidson [1969]
Participation Models
Equations8
Activity Probability of participation Level of participation
Water-based
Swimming Marginal effect of income on Effect sensitive to
probability changes with region of residence
level of income
Water skiing Constant marginal effect*3 Income not a signi-
ficant determinant
Other boating Constant marginal effect Marginal effect of
income changes with
level of income
Canoeing Constant marginal effect Income not a signi-
ficant determinant
Other Activities
Camping developed Income not a significant Constant marginal
determinant effect of income
aThese results are based on the estimates reported in Chapter 5 of Cicchetti,
Seneca, and Davidson [1969].
These estimated parameters were substantially smaller in numerical magnitude
than the estimated parameter for income in the probability equation for
fishing.
Finally the overall explanatory power, as measured by R2, is also quite
variable across sites. In some cases, such as sites 303 (Beaver Lake,
Arkansas), 313 (Ft. Randall, Lake Francis Case, South Dakota), 314 (Grape-
vine Lake, Texas), and 337 (Proctor Lake, Texas), the R2 is comparable to
most cross-sectional analyses. For the remainder it is somewhat low, indicat-
ing that there may be other major factors influencing these site demands.
7.5.4 Results for the Tailored Models
It should be acknowledged that while the basic model provides a plausi-
ble specification for a site demand equation, there may well be a number of
other determinants of these demands. Indeed, the low R2 would certainly
support this conclusion. Since the overall objective is to develop a general
model for projecting the effects of changes in any water-based site's charac-
teristics on the site demand, site demand equations must adhere to a common
7-36
-------
specification. Nonetheless, this does not prevent an appraisal of the sensitiv-
ity of the basic model's parameter estimates to the inclusion of additional
variables. As a consequence, the analysis plan considered a wide array of
alternative specifications of each demand function. These models include
additional socioeconomic information—age, sex, education, and race—as well as
an attitudinal variable (coded as zero and 1), with 1 designating individuals
who regarded outdoor recreation as very important in comparison to their
other interests (RECIMP).
Table 7-6. Comparison of Basic Model With Tailored Model: Coefficient
for (TC+MC)
Site name
Site No. Basic model
Range
of estimates
tailored models
Lock and Dam No. 2 (Arkansas
River Navigation System)", AR
Beaver Lake, AR
Blakely Mt. Dam, Lake
Ouachita, AR
Cordell Hull Dam and
Reservoir, TN
Dewey Lake, AR
Grapevine Lake, TX
Greers Ferry Lake, AR
Genada Lake, MS
Lake Washington Ship Canal, WA
Melvern Lake, KS
Millwood Lake, AR
Mississippi River Pool No. 3, MN
Mississippi River Pool No. 6, MN
Ozark Lake, AR
Philpott Lake, VA
Pine River, MN
Proctor Lake, TN
Sardis Lake, MS
Whitney Lake, TX
302
303
307
310
-0.0125 -0.010 to -0.013
-0.0066 -0.0060 to -0.0070
-0.0079 -0.0070 to -0.0080
-0.0139 -0.0013 to -0.0015
312 .
314
315
316
320
322
323
324
325
331
333
334
337
340
344
-0.0024
-0.0073
-0.0065
-0.0095
-0.0037
-0.0079
-0.0081
-0.0057
-0.0074
-0.0046
-0.0087
-0.0017
-0.0134
-0.0030
-0.0025
-0.0020 to -0.0030
-0.0060 to -0.0090
-0.0060 to -0.0070
-0.0080 to -0.0100
-0.0030 to -0.0400
-0.0070 to -0.0090
-0.0070 to -0.0090
-0.0050 to -0.0060
-0.0070
-0.0030 to -0.0050
-0.0070 to -0.0090
-0.0010 to -0.0020
-0.0013 to -0.0014
-0.0030 to -0.0040
-0.0020 to -0.0030
7-37
-------
Table F-5 in Appendix F presents a sample of these models for a selected
set of the 43 sites. These cases represent the site demands where one or
more alternative specifications would have been regarded as equivalent or
better to the basic model. In evaluating these models, the focus was on the
estimated parameters for variables that were common between the basic model
and each variation to it. In general, the most important parameter—the
coefficient for the implicit price—was remarkably stable. Table 7-6 provides
a comparison of these estimates from the tailored specifications with the basic
model estimates reported in Table 7-4.
Since it is widely acknowledged in the econometrics literature that pretest-
ing and sequential estimation practices affect the kinds of inferences that can
be drawn concerning the properties (i.e., unbiasedness, efficiency, etc.) of
the "final" model's estimated parameters, these types of sensitivity analyses
gauge whether the decisions required to select the final models were important
to the parameters of central importance to the overall objectives.* The general
criteria used for selecting the specifications reported in Table 7-4 were based
on three considerations: (1) agreement between the sign of the estimated
parameters with what was expected from economic theory; (2) statistical signif-
icance of the estimates using conventional criteria as appropriate indexes of the
precision of the estimates; and (3) robustness of the measured effects for
important variables (such as TC+MC) to model specifications.
7.5.5 Evaluation of Measures of the Opportunity Cost of Travel Time
Tables 7-7 and 7-8 report the results of two sets of tests for the basic
model and tailored models, respectively. The tests have been structured to
evaluate alternative definitions of the opportunity cost of travel time. The
two models can be readily described. The first maintains that the wage rate
is the most appropriate measure. This would imply that the measure of the
time costs of travel, TC, can be added to the travel costs as in Equation
(7.31). Alternatively, if, as several authors have argued, the opportunity
cost is a different, constant multiple of the wage, the model should be written
as:
j = a0 + aiTCj + ciaMCj + a3INC. + e . (7.32)
Thus, if the wage rate is the appropriate measure of the opportunity cost of
travel time, otj should equal or2. Rejection of this null hypothesis would
therefore provide support for the arguments against the use of the wage rate
as the opportunity cost. The sixth column of Table 7-7 reports the relevant
F-statistic and significance levels for this hypothesis using the basic model.
Overall the hypothesis is rejected for 9 of the 43 sites with the basic model at
the 5-percent significance level. These decisions are generally repeated with
the tailored models for the sites reported in both cases.
*This approach is clearly in the spirit of the suggestion made by Klein
et al. [1978] for dealing with estimation problems.
7-38
-------
Table 7-7. F-Test for Restriction of General Model
10
to
Hypothesis 1, Full-Time Cost:
Hypothesis 2, Cesarlo Hypothesis:8
Unrestricted model:
Site
Allegheny River System, PA
Arkabutla Lake, MS
Lock and Dam No. 2 (Arkansas River
Navigation System), AR
Beaver Lake, AR
Belton Lake, TX
Benbrook Lake, TX
Berlfn Reservoir, OH
Blakely Mt. Dam, Lake Ouachlta, AR
Canton Lake, OK
Clear water Lake, MO
Corded Hull Dam and Reservoir, TN
DeGray Lake, AR
Dewey Lake, AR
Ft. Randall, Lake Francis Case, SD
Grapevine Lake, TX
LN
LN
LN
Stte
number
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
Visits = er0 + a,
Visits = oi0 + of,
Visits = o0 + a,
Sum of
squared
residuals,
Hypothesis
45.27
24.84
7.58
104.64
25.90
11.52
62.13
45.00
41.48
45.84
47.11
22.45
16.03
24.34
22.64
(T + M) Cost * «3 Income
(T 1/3 * M) Cost + o3 Income
T Cost + «2 M Cost +
Sum of
squared
residuals,
1 Hypothesis 2
45.27
24.12
8.02
109.61
25.82
11.29
61.93
44.02
43.49
45.51
46.18
22.62
16.45
26.34
25.40
a 3 Income
Sum of squared
residuals,
unrestricted
model
44.99
23.93
6.91
104.09
23.71
11.20
61.70
43.96
41.25
45.37
46.18
22.45
15.91
24.05
21.34
F* statistic
Ho: ot = cr2
0.53
0.14
0.07
0.27
0.04
0.28
0.43
0.15
0.53
0.40
0.16
0.99
0.58
0.46
0.02
level of significance
Ho: o = 1/3o2
0.53
0.50
0.02
0.01
0.04
0.56
0.56
0.73
0.06
0.64
0.99
0.56
0.24
0.04
0.01
"(T 1/3 + M) cost represents the total cost of a round trip where travel time Is evaluated at one-third of the predicted wage rate.
(continued)
-------
Table 7-7. (continued)
Hypothesis 1, Full-Time Cost:
Hypothesis 2, Cesario Hypothesis:*
Unrestricted model:
Site
Greers Ferry Lake, AR
Genada Lake, MS
Hords Creek Lake, TX
Isabella Lake, CA
Lake Okeechobee and Waterway, FL
Lake Washington Ship Canal, WA
Leech Lake, MN
Melvern Lake, KS
^ Millwood Lake, AR
i Mississippi River Pool No. 3, MN
° Mississippi River Pool No. 6, MN
Navarro Mills Lake, TX
New Hogan Lake, CA
New Savannah Bluff Lock & Dam, GA
Norfork Lake, AR
Ozark Lake, AR
Perry Lake, KS
LN
LN
LN
Site
number
315
316
317
318
319
320
321
322
323
324
325
327
328
329
330
331
332
Visits = 00 + ot
Visits = 50 + a,
Visits = 00 + a,
Sum of
squared
residuals,
Hypothesis
110.96
27.65
30.61
23.59
21.84
26.22
21.18
31.37
28.67
20.68
37.73
23.44
30.71
16.67
18.45
24.31
12.06
(T + M) Cost + a3 Income
(T 1/3 + M) Cost + 53 Income
T Cost + at M Cost +
Sum of
squared
residuals,
1 Hypothesis 2
120.65
27.39
30.32
23.45
22.71
28.31
20.78
31.16
28.36
22.59
39.47
23.59
30.76
16.65
19.58
25.53
12.01
a 3 Income
Sum of squared
residuals,
unrestricted
model
104.06
27.39
30.18
23.45
21.59
24.90
20.64
31.15
28.35
20.63
37.49
23.30
30.60
16.44
17.53
21.93
12.00
F- statistic level
Ho: al = otj
0.01
0.41
0.40
0.61
0.59
0.15
0.29
0.59
0.46
0.74
0.51
0.64
0.72
0.49
0.17
0.03
0.73
of significance
Ho: a = 1/3a2
0.01
0.99
0.63
0.99
0.26
0.02
0.59
0.91
0.90
0.04
0.06
0.50
0.66
0.51
0.04
0.01
0.89
a(T 1/3 + M) cost represents the total cost of a round trip where travel time is evaluated at one-third of the predicted wage rate.
(continued)
-------
Table 7-7. (continued)
Hypothesis 1, Full-Time Cost:
Hypothesis 2, Cesarlo Hypothesis:8
Unrestricted model:
Site
Phllpott Lake, VA
Pine River, MN
Pokegama Lake, MN
Pomona Lake, KS
Proctor Lake, TN
Rathbun Reservoir, IO
Sam Rayburn Dam & Reservoir, TX
Sardl* Lake, MS
Waco Lake, TX
Whitney Lake, TX
Youghlogheny River Lake, PA
LN
LN
LN
Site
number
333
334
335
336
337
338
339
340
343
344
345
Visits = B0 + fflj
Visits = a0 + or,
Visits = o0 + a,
Sum of
squared
residuals,
Hypothesis
10.42
22.96
37.31
14.42
13.25
21.70
34.21
64.10
20.07
113.80
20.17
(T + M) Cost + O3 Income
(T 1/3 + M) Cost + a3
T Cost + c»2 M Cost +
Sum of
squared
residuals,
1 Hypothesis 2
9.97
23.44
38.26
14.18
12.41
20.83
33.16
66.42
20.02
115.40
21.35
I ncome
a 3 Income
Sum of squared
residuals,
unrestricted
model
9.85
21.25
36.81
13.27
12.24
17.29
33.15
52.76
17.23
96.77
18.17
F-statistic level
Ho: al = a2
0.17
0.02
0.35
0.13
0.05
0.01
0.16
0.01
0.01
0.01
0.10
of significance
Ho: o = 1/3o2
0.52
0.01
0.11
0.18
0.42
0.03
0.89
0.01
0.01
0.01
0.04
"(T 1/3 + M) cost represents the total cost of a round trip where travel time Is evaluated at one-third of the predicted wage rate.
-------
Table 7-8. F-Test for Restriction of Tailored Models*
Site
Lock and Dam No. 2 (Arkansas River
Navigation System), AR
Beaver Lake, AR
Blakely Mt. Dam, Lake Ouachita, AR
Cordell Hull Dam and Reservoir, TN
Dewey Lake, KY
Grapevine Lake, TX
Greers Ferry Lake, AR
V Grenada Lake, MS
M Lake Washington Ship Canal, WA
Melvern Lake, KS
Millwood Lake, AR
Mississippi River Pool No. 3, MN
Mississippi River Pool No. 6, MN
Ozark Lake, AR
Philpott Lake, WA
Pine River, MN
Proctor Lake, TX
Sardis Lake, MS
Whitney Lake, TX
Site
number
302
303
307
310
312
314
315
316
320
322
323
324
325
331
333
334
337
340
344
F-statistic level of significance
Model 1
0.07
0.29
0.18
0.20
0.49
0.03
0.01
0.35
0.59
0.41
0.49
0.99
0.54
0.03
0.16
0.03
0.06
0.01
0.01
Model 2
0.07
0.32
0.13
0.20
0.63
0.04
0.01
0.47
0.20
0.61
0.46
0.88
0.28
0.02
0.08
0.03
0.09
0.01
0.01
Model 3
0.05
0.06
0.17
0.46
0.16
0.02
0.01
0.36
0.10
0.46
0.84
0.88
0.56
0.03
0.17
0.02
0.16
0.01
0.01
Model 4
0.03
0.20
0.30
0.16
0.58
0.03
0.01
0.35
0.18
0.61
0.46
0.64
0.76
0.03
0.14
0.02
0.07
0.01
0.01
Model 5
0.05
0.34
0.30
0.22
0.87
0.05
0.02
0.20
0.16
0.99
0.46
0.75
0.55
0.13
0.04
0.02
0.05
0.01
0.01
F-tests are calculated using the five restricted
travel time and mileage cost are separate.
models in T.able 7-6 against unrestricted models where
-------
The second hypothesis considers Cesario's suggestion that the opportunity
cost is a multiple of the wage rate. The explanation for the parametric
treatment of this hypothesis stems from the definitions of the components of
the cost of a trip. TC, the time costs of travel, is defined as the predicted
wage rate, say w, times the travel time, t, or wt. If the opportunity cost of
travel time is some multiple, k (k < 1) of the wage rate and can be assumed
to be constant across individuals, the true measure of TC (designated TC)
should be kwt. Both travel costs and the time costs of travel should, when
the latter is correctly measured, have the same effeclt on the demand for a
site's services. Thus, if the maintained hypothesis (TC = kwt) is correct, ax
can be expected to be equal to a2. However, k cannot be measured._ By
using wt as a proxy and assuming that k is co_nstant, the estimates of o^ in
the model using TC = wt can be expected to be o^ = kat- Since it is expected
that ofj = (*! and that a2 will, under ideal conditions, _equal a2, the Cesario
suggestion can be treated as the hypothesis that o^ = ka2 in terms of Equation
(7.32). Since Cesario's_ specific suggestion was that k = 1/3, the second
hypothesis is oij = 1/3 52. The seventh column of Table 7-7 reports the
results for this test. Nearly twice as many sites (16) reject this null hypoth-
esis with the basic model.
Thus, there is greater support for the use of the wage rate as a measure
of the opportunity cost of travel time than the Cesario one-third adjustment
to the wage. However, there is no unambiguous choice, because some sites
fail to reject both sets of restrictions.
7.6 FURTHER EVALUATION OF THE TRAVEL COST MODELS
Section 7.5 presented estimates of the final models for each of 43 recrea-
tion sites. As noted earlier, the methodology developed in this chapter
requires that the individual site demand equations adopt the same specification.
In some cases this specification would have been adopted as "best," and, for
others, the choice was not as clearcut. As a consequence, it was necessary
to evaluate the sensitivity of important demand parameter estimates to the
model specification. There are several additional aspects of these travel cost
models that require further consideration. Therefore, this section collects
the results of the further evaluations of these models. This analysis was
conducted in an attempt to identify potential shortcomings with the models and
to appraise their importance for the estimated values. Most of these difficul-
ties arise from either econometric problems with the model or limitations that
would be expected based on the economic model of consumer behavior devel-
oped at the outset of the chapter.
The first aspect of these travel cost models requiring further consideration
arises from the data and the model specification themselves. The visit measure
used in this analysis is a positive integer by definition. This raises a number
of potential econometric problems. For the purpose of this study these prob-
lems have been ignored.* However, where possible, appraisals have been made
The implications of these features for the site demand models and benefit
estimates are currently being evaluated using appropriately structured maximum
likelihood estimators and recent method of moments approximations proposed by
Greene [1983].
7-43
-------
of the potential implications of one of the most important aspects of the sample—
that it observes only the behavior of individuals who have visited each site at
least once. To evaluate the potential importance of the bias in ordinary least-
squares estimates as a result of the truncated form of the measure of site
usage, Olsen's [1980] method of moments approximation of the maximum likeli-
hood estimates for models with truncated dependent variables has been used.
The Olsen method relies on approximating the mean of the conditional
distribution for the dependent variable (i.e., E(y | y >0)). His proposed
scaling factors use a first-order approximation to derive a relationship between
the ordinary least-squares estimates of a model's parameters and the maximum
likelihood estimates. They can be estimated from the moments of the incom-
plete (i.e., truncated) distribution. These scaling factors are used to gauge
the magnitude of the differences between an approximate maximum likelihood
estimator and ordinary least squares. Thus, as Olsen suggests, they provide
a crude index of the potential severity of the problems with truncation.
Greene [1981] has also proposed an approach for adjusting ordinary least-
squares estimates in the present Tobit and truncated dependent variable
models. He found that Olsen's approximation tends to overstate the bias.
Olsen's approximation will be closest to Greene's approach for models with
small coefficients of determination (i.e., R2). As R2 increases the Olsen
adjustment will tend to overstate the extent of bias. Thus, this study's
screening of estimated site demand models provides a fairly conservative basis
for gauging the bias due to the truncation of the measure of site usage.
Table 7-9 reports these scaling factors for the 33 sites in which the general
model performed well.
The scaling factors in the fourth column of Table 7-9 can be interpreted
as the multiplicative adjustment coefficients required for the ordinary least-
squares parameter estimates to approximate the maximum likelihood estimates
(based on the assumption of a truncated distribution). Thus, for site No. 301,
the maximum likelihood estimates would be 15 percent greater than the ordinary
least-squares parameter estimates in absolute magnitude. These comparisons
suggest that several sites exhibit pronounced truncation effects. For at least
11 of these sites, the bias associated with the ordinary least-squares estimates
may well be quite substantial. As a consequence, the potential for differential
bias in the estimates of these site demand functions is accounted for in the
final model. That is, the generalized least-squares estimates relating the
features of each site demand function to the site's characteristics have been
derived using two samples—one including all sites with complete data (i.e.,
sites with plausible demand models and complete information on water quality
and other site characteristics) and a second omitting those sites with potentially
important truncation effects.
A second source of qualification to the travel cost demand model arises
from the assumption that all users of each individual site have the same
derived demand for that site's services. In most cases, disparities in onsite
time could not be accounted for. Moreover, it has not been possible to adjust
for the different mixes of activities undertaken by different individuals at the
7-44
-------
Table 7-9. Effects of Truncation on the Travel Cost Models' Estimates
Site
Site name number
Arkabutla Lake, MS
Lock & Dam No. 2 (Arkansas River
Navigation System), AR
Beaver Lake, AR
Belton Lake, TX
Benbrook Lake, TX
Blakely Mt. Dam,
Lake Ouachita, AR
Canton Lake, OK
Cordell Hull Dam & Reservoir, TX
DeGray Lake, AR
Dewey Lake, KY
Ft. Randall, Lake Francis Case, SD
Grapevine Lake, TX
Greers Ferry Lake, AR
Grenada Lake, MS
Hords Creek Lake, TX
Isabella Lake, CA
Lake Okeechobee and Waterway, FL
Lake Washington Ship Canal, WA
Leech Lake, MN
Melvern Lake, MS
Millwood Lake, AR
Mississippi River Pool No. 3, MN
Mississippi River Pool No. 6, MN
New Savannah Bluff Lock & Dam, GA
Norfork Lake, AR
Ozark Lake, AR
Philpott Lake, VA
Pine River, MN
Pokegama Lake, MN
Proctor Lake, TX
Sam Rayburn Dam & Reservoir, TX
Sardis Lake, MS
Whitney Lake, TX
301
302
K
303°
304
305
307
308
310
311
31 2b
31 3b
314
315
316
317
318°
3l9b
320b
321 b
322
323.
324b
325
329.
330b
331
333
334b
335°
337
339
340
344
Incomplete
mean/
standard
deviation
2.115
4.115
0.975
2.080
3.001
1.425
1.299
1.855
1.818
0.866
0.817
2.458
1.493
2.401
1.374
1.169
1.119
0.876
0.994
1.269
1.739
1.020
1.557
2.137
1.139
1.577
2.413
0.949
1.018
1.960
1.474
3.107
1.821
Olsen
ML scaling
factor3
1.15
1.01
13.55
1.18
1.01
2.18
2.92
1.29
1.34
13.55
13.55
1.05
1.85
1.07
2.44
5.33
7.87
13.55
13.55
3.30
1.39
13.55
1.75
1.15
6.69
1.67
1.07
13.55
13.55
1.25
1.95
1.01
1.34
These scaling factors are assigned approximately using Olsen's Table I by
selecting the closest value for the reported mean to standard deviation with
the incomplete distribution.
These sites were omitted for truncation bias in second estimation of the
model.
7-45
-------
same site.* Thus, it might be conjectured that the same demand model is not
equally well suited to all survey respondents. Such a hypothesis would imply
that the parameter estimates would be sensitive to sample composition. That
is, deleting individual observations associated with individuals with especially
long onsite time or rather different sets of activities may well have a pro-
nounced effect on the ordinary least-squares estimates of the model's parame-
ters. Moreover, this impact may be differentially important to subsets of the
sites considered for this analysis because there are substantial differences in
the number of respondents for these sites.
To investigate this issue, DFBETA was calculated (Belsley, Kuh, and
Welsch's [1980] regression diagnostic). This index was designed to act as an
aid in identifying influential or outlying observations. It is not a statistical
test. It has been used to judge the "influence" of specific observations on
this study's estimates of site demand parameters. With this evaluation, it is
then possible to consider the features of these survey respondents to evaluate
whether there are economic reasons for expecting that the demand patterns of
these individuals must be explained in a different framework. The specific
index used is defined as the difference between the ordinary least-squares
estimate for each parameter based on the complete sample and the correspond-
ing estimate based on the sample with the omission of one observation. These
indexes were calculated for each parameter and each observation. A review
of these estimates indicated that no single observation had an important effect
on the estimated parameters. This conclusion was found for all sites, includ-
ing those with a somewhat limited number of sampled recreationists. While
this finding does not guarantee that the effects of onsite time and the mix of
recreation activities are inconsequential influences on site demand, it does
suggest that they are unlikely to have pronounced effects on these estimates.
f
The final aspect of the travel cost models that requires further considera-
tion stems from the relationship between decisions on the number of trips to
each recreation facility and the amount of time spent onsite per trip. As
noted previously, the onsite time measure relates to the trip each respondent
was undertaking at the time that he was interviewed, but there is no informa-
tion as to how representative this trip was. That is, the survey does not
identify for all visits during the season the amount of time spent onsite per
trip. Thus, the analyses of travel and onsite time costs (both the time and
the distance components) implicitly assume that the onsite time for the current
trip is a good indicator of the onsite time for all past trips.
If this assumption is reasonable, then it is also plausible to consider the
prospects for a simultaneous equation model to describe the decisions for
visits to a site and the time on the site per trip. When using simultaneous
equation models with the Federal Estate Survey data, two aspects of consist-
ency in recreation choices must be considered.
The feasibility of including measures of the activities undertaken into
the second stage models for the estimated site demand parameters is currently
being investigated.
7-46
-------
First, individuals may decide the amount of time to be spent onsite--first
based on the activities they wish to undertake and then based on the number
of visits to a site to engage in these activities. Within this decision frame-
work, onsite time can be treated as exogenously determined. Visits may be
conditional upon these onsite time choices. This would not imply that onsite
time was not important to decisions on visits to a recreation facility. Rather,
it would suggest that they are not joint decisions. Indeed, for some cases it
may be necessary to segment the samples of users according to their lengths
of stay on the site.*
Secondly, the onsite time may not be constant for all trips, and thus the
measure available for per-trip time onsite is inappropriate. These prospective
difficulties in evaluating the relationship between visit and onsite time deci-
sions will therefore influence any effort to model their respective roles in
recreation site demand functions. Nonetheless, in an attempt to account for
these simultaneous equation effects, onsite time has been treated as an endo-
genous variable, and a variety of specifications have been considered for it as
well as for the site demand models themselves. In general, this study has
attempted to instrumentalize the measures of the variable costs of onsite time.
More specifically, onsite cost is specified as a nonlinear combination of exoge-
nous and endogenous variables as a result of the respective roles for the
opportunity cost of time and onsite time.
Following conventional practice (see Kelejian [1971]), the combination is
treated as a right-hand-site endogenous variable and the models were estimated
with two-stage least squares.! The first-stage instruments were composed of
the included predetermined variables in each specification for the travel cost
model along with age, sex, and a qualitative variable to reflect whether the
recreation activities included camping. Several variations in these instruments
were considered. However, this set of variables provided acceptable models
for the largest set of site demands. Table 7-10 reports the two-stage esti-
mates for 21 of the sites.t As with earlier results (i.e., using ordinary least
squares and ignoring onsite time), the role of income appears quite limited for
nearly all sites. Only one site demand, Millwood Lake, Arkansas (No. 323)
yields a statistically significant estimate for the coefficient of family income.
The results for onsite time are encouraging but certainly not clearcut. As
suggested by the theoretical model, onsite time (SCOST) affects the "price"
of a trip to the site (since the model assumes all trips have the same onsite
time), and it also contributes to the production of recreation service flows.
*Our analysis with regression diagnostics indicated that these problems
were unlikely to be present in our models because the results were not sensi-
tive to deleting individual observations.
tldeally, the Kelejian method calls for polynomials in the predetermined
variables as first-stage instruments. This was not attempted in our case
because of the limited number of observations for several of the sample recre-
ation sites.
21 sites are the result of two screenings of the 43 sites in the
survey. The first screening eliminates 10 sites with implausible demand func-
tions. The second eliminates 12 sites that experienced truncation bias.
7-47
-------
Based on the first of these impacts, it would be expected to have a negative
impact on the demand for visits to a site. It is a component of the price of a
visit. In addition, however, increases in the time spent onsite provide one
means of substituting for visits. Thus, one might hypothesize a positive
"substitution" effect on the demand for trips to a recreation site. Of course,
the demand model reflects a composite of these two influences.
The empirical results are consistent with the presence of these opposing
influences on site demand. For some sites, the effect of SCOST is positive,
while, for others, it is negative. Five of the 21 site demands exhibit statis-
tically significant estimates for onsite costs, based on the asymptotic t-ratios.
In all of these cases the estimated coefficients are negative.
Table 7-10.
Two-Stage Least-Squares Estimates for Selected
Travel Cost Site Demand Models
Site^Name
Beaver Lake, AR
Benbrook Lake, TX
Blakely Mt. Dam,
Lake Ouachita, AR
Canton Lake, OK
Cordell Hull Dam & Reservoir, TX
DeGray Lake, KY
Ft. Randall, Lake Francis Case, SD
Grapevine Lake, TX
Greers Ferry Lake, AR
Grenada Lake, MS
Hords Creek Lake, TX
Lake Washington Ship Canal, WA
r
Leech Lake, MN
Millwood Lake, AR
Mississippi River Pool, No. 6, MN
Norfork Lake, AR
Philpott Lake, MN
Pokegama Lake, MN
Proctor Lake, TX
Sam Rayburn Dam & Reservoir, TX
Whitney Lake, TX
The numbers in parentheses below
association.
Site
No
303
305
307
308
310
311
313
314
315
316
317
320
321
323
325
330
333
335
337
339
344
the
Intercept
1.705
(11.45)
1.999
(6.05)
1.721
(3.077)
1.787
(5.83)
1.603
(7.58)
1.587
(4.51)
1.778
(4.74)
2. 154
(12.73)
1.607
(10.60)
1.5S1
(5.06)
1.938
(5.81)
0.505
(0.66)
0.293
(0.79)
0.829
(2.33)
1.198
(3.81)
0.666
(1-65)
2.209
(7.21)
1.344
(3.62)
1.783
(6.47)
1.157
(3.38)
1.527
(8.30)
Estimated
TC+MC
-0.0056
(-8.12)
-0.0052
(-3.84)
-0.0081
(-4.57)
-0.0172
(-2.83)
-0.0137
(-4.53)
-0.0083
(-3.24)
-0.0042
(-2.43)
-0.0053
(-5.06)
-0.0066
(-7.70)
-0.0073
(-2.86)
-0.0050
(-2.06)
-0.0038
(-2.40)
-0.0032
(-2.33)
-0.0091
(-3.88)
-0.0062
(-3.24)
-0.0055
(-2.85)
-0.0074
(-3.75)
-0.0030
(-3.54)
-0.0149
(-5.99)
-0.0098
(-2.92)
-0.0027
(-1-44)
travel cost models8
SCOST
-0.0003
(-2.21)
-0.0001
(-0.55)
-0.0002
(-0.40)
-0.0008
(-0.80)
-0.0002
(-0.48)
-0.0004
(-1.38)
-0.0021
(-2.28)
-0.0001
(-2.31)
0.00002
(0.08)
-0.0016
(-2.64)
-0.0001
(-0.19)
0.0465
(1.07)
0.0004
(1.26)
0.000009
(0.02)
-0.0006
(-1.66)
-0.0008
(-0.28)
-0.0007
(-1.49)
-0.0004
(-1.32)
0.0003
(0.84)
-0.0001
(-0.29)
-0.0009
(-4-15)
INC
-0.000003
(-0.61)
0.000003
(0.34)
-0.000009
(-0.81)
0.000005
(0.47)
0.000003
(0.35)
-0.000002
(-0.21)
0.000009
(0.79)
0.000009
(1.72)
0.000009
(1.48)
0.00001
(0.71)
-0.00002
(-1-77)
0.00002
(0.78)
0.000011
(0.95)
0.00002
(2.54)
0.00002
(1.95)
0.00001
(0.91)
0.000002
(0.18)
-0.000009
(-0.86)
0.000002
(0.33)
0.000002
(0.19)
0.000006
(1.11)
estimated coefficients are asymptotic t-ratios for the null
AGE
-0.0009
(-0.27)
-0.0020
(-0.29)
0.0048
(0.78)
0.0043
(0.58)
0.0072
(1.71)
0.0104
(1-44)
-0.0087
(-0.94)
-0.0129
(-2.91)
-0.0045
(-1.09)
0.0100
(1.99)
-0.0030
(-0.36)
-0.0018
(-0.15)
0.0069
(0.93)
0.0134
(1.88)
0.0070
(0.97)
0.0149
(1.55)
-0.0082
(-1.07)
0.0020
(0.34)
0.0049
(1.01)
0.0102
(2.00)
0.0078
(1.88)
hypothesis
R*
0.42
0.31
0.21
0.26
0.35
0.21
0.38
0.50
0.28
0.26
0.20
0.21
0.17
0.30
0.24
0.20
0.47
0.24
0.56
0.17
0.10
of no
7-48
-------
The remaining sites also were modeled within a simultaneous framework.
However, in these cases the parameters estimates were inferior to those
derived using ordinary least squares under the assumption of constant onsite
time. As a rule, the estimated effect of travel cost (TC+MC) was not statis-
tically significant and, in some cases, suggested a positive effect on site
demand. Moreover, the estimated effects of onsite costs were generally
statistically insignificant. Thus, the models reported in Table 7-10 are the
cases in which the simultaneous estimates were judged to be equivalent or
better than the ordinary least-squares results reported in Section 7.5.
These results are important for two reasons. They attempt to deal with
onsite time costs and travel costs within a single demand framework. Most
authors (see Brown and Mendelsohn [1980] as a notable example) have either
attempted to partition their samples according to the time spent onsite and
estimate separate demand models for each grouping or have assumed that
onsite time was not important to the decisions for trips to a recreation facility.
This latter assumption might be the result of features of the recreation activi-
ties undertaken and site selected or simply because the time onsite was
approximately constant across trips.
Table 7-11. Comparison of Ordinary Least-Squares and Two-Stage
Least-Squares Estimates of Travel Cost (TC. + MC.) Parameters
Site name Site
Beaver Lake, AR
Benbrook Lake, TX
Blakely Mt. Dam, Lake Ouachita, AR
Canton Lake, OK
Cordell Hull Dam & Reservoir, TX
De Gray Lake, AR
Ft. Randall, Lake Francis Case, SD
Grapevine Lake, TX
Greers Ferry Lake, AR
Grenada Lake, MS
Hords Creek Lake, TX
Lake Washington Ship Canal, WA
Leech Lake, MN
Millwood Lake, AR
Mississippi River Pool No. 6, MN
Norfork Lake, AR
Philpott Lake, VA
Pokegama Lake, MN
Proctor Lake, TX
Sam Rayburn Dam & Reservoir, TX
Whitney Lake, TX
No.
303
305
307
308
310
311
313
314
315
316
317
320
321
323
325
330 ,
333
335
337
339
344
Ordinary
least-
squares
estimate
-0.0066
-0.0054
-0.0079
-0.0206
-0.0139
-0.0070
-0.0066
-0.0073
-0.0065
-0.0095
-0.0050
-0.0037
-0.0022
-0.0081
. -0.0074
-0.0047
-0.0087
-0.0033
-0.0134
-0.0094
-0.0025
Two-
stage
least-
squares
estimate
-0.0056
-0.0052
-0.0081
-0.0172
-0.0137
-0.0083
-0.0042
-0.0053
-0.0066
-0.0073
-0.0050
-0.0038
-0.0032
-0.0091
-0.0062
-0.0055
-0.0074
-0.0030
-0.0149
-0.0098
-0.0027
7-49
-------
Table 7-12. Hausman Test for Differences Between Two-Stage Least-Squares and
Ordinary Least-Squares Estimates
-J
o
Site
303
305
307
308
310
311
313
314
315
316
317
320
321
323
325
330
333
335
337
339
344
Notes :
NA
~ 2SLS * OLS
ON -Oti
X A
-0.0010
0.0002
-0.0002
0.0034
0.0002
-0.0013
0.0024
0.0020
-0.0001
0.0022
-0.0000248
-0.0001
-0.0010
-0.0010
0.0012
-0.0008
0.0013
0.0003
-0.0015
-0.0004
-0.0002
= The t-statistic
fi2SLS
0.0000005
0.0000018
0.0000031
0.0000368
0.0000092
0.0000066
0.0000030
0.0000011
0.0000007
0.0000065
0.0000058
0.0000027
0.0000018
0.0000055
0.0000037
0.0000037
0.000003859
0.0000007
0.0000062
0.0000112
0.0000034
could not be
.OLS
0.0000003
0.0000017
0.0000024
0.0000152
0.0000054
0.0000054
0.0000012
0.0000007
0.0000005
0.0000047
0.0000056
0.0000010
0.0000014
0.0000041
0.0000028
0.0000034
0.000003895
0.0000005
0.0000032
0.000011
0.0000019
calculated as
VAR,,CI e-VAR^, c
2SLS OLS
0.000447
0.000316
0.000837
0.004648
0.001949
0.001095
0.001342
0.000633
0.000447
0.001342
0.000447
0.001304
0.000633
0.001183
0.000949
0.000548
NA
0.000447
0.001732
0.000447
0.001225
the variance since the c
t-statistic
2.237
0.633
-0.239
0.731
0.103
-1.187
1.788
3.160
-0.224
1.639
-0.05
-0.077
-1.580
-0.845
1.264
-1.460
NA
0.671
-0.866
-0.894
-0.163
jrdinary least'
squares estimate was greater than the two-stage least-squares estimate.
3L = the estimated coefficient of the travel plus mileage cost variable.
VAR = the variance of Si.
2SLS = the two-stage least-squares model.
OLS = the ordinary least-squares model.
-------
Of course, this perspective is implicitly adopted for the results in the
previous section. Thus, the second potential use of these findings is to
gauge how important an error the failure to take account of simultaneity might
be for the use of the general models to derive a benefit estimation framework.
Table 7-11 reports a comparison of the ordinary least-squares estimates of the
travel cost parameter versus the two-stage results for each of the sites where
the two-stage least squares were judged to be at least as good as the ordinary
least-squares models. Overall the results are quite similar. There are two
types of comparisons that can be made between these estimates. As a practi-
cal matter, for benefit estimation, the numerical differences between the
ordinary least squares and two-stage least-squares estimates are of concern.
For the most part, the two sets of estimates for the (TC} + MC.) parameter
are quite comparable. A second comparison involves considering whether the
null hypothesis that the parameters for the travel and time cost variable were
equal in the two models would be rejected based on these estimates. It is
possible to develop an asymptotic test for this hypothesis using Hausman's
[1978] approach to specification tests. Hausman derives an expression for
the variance of the difference between two estimators of the same parameter.
These estimators are defined for two hypotheses. It must be assumed that
one is a consistent estimator under both the null and alternative hypotheses
and that the second estimator is asymptotically efficient under the null but
inconsistent under the alternative hypothesis. Given asymptotic normality and
these assumptions, the variance of the difference between the estimators is the
difference in their respective variances. This application considers the differ-
ence between the two-stage least-squares and ordinary least-squares estimates
of the coefficient for the travel cost variable. Constructing the corresponding
t-ratio gives the following:
A 2SLS a OLS
t = "1 - "1 - . (7.33)
Table 7-12 reports the details of the calculation of these test statistics.
The t-ratio will follow an asymptotically normal distribution. Considering
these statistics as an approximate basis for testing the difference between
these coefficient estimates gives only two cases (Sites 303 and 314) in which
the null hypothesis of equality would be rejected at the 5-percent significance
level. Thus, these findings largely confirm the informal judgmental inspection
and indicate that the ordinary least-squares models, which assume onsite
costs to be constant, are unlikely to have serious errors because of this
assumption.
7.7 ANALYZING THE ROLE OF WATER QUALITY FOR RECREATION
DEMAND
The last step in the empirical modeling involves estimating the role of
water quality and other site attributes in the demands for a site's services.
The structure of the model has been detailed in Section 7.3. Thus, what
remains to be presented is a specific description of the results of the applica-
tion. The overall objective is to attempt to explain the observed variation in
7-51
-------
each of the estimated demand parameters across sites by the characteristics of
those sites. With such a model, it is possible, in principle, to characterize
the change in a site's demand in response to a change in any of the factors
influencing those demand parameters. Thus, it would be feasible to evaluate
the implications of a change in water quality for the demand for the site's
services, even though the change has not been experienced. This ability
arises from the fact that this model provides a general description of the
factors that influence the features of site demands within a single framework.
The model has been derived from two subsets of the 43 site demand
models described in Section 7.5 above. The first of these included 33 sites
with plausible site demand functions.* The second restricts the sample further
by eliminating 11 of these sites, based on estimates of the Olsen scaling
factors reported in Table 7-9. As noted earlier, these scaling factors provide
some indication of the prospects for bias due to the truncation in the measures
of site usage. These 11 sites exhibited the largest values of the estimated
scaling factor, ranging from 5.33 to 13.55. The specific sites eliminated from
the sample are footnoted in Table 7-9 on page 7-45.
To develop estimates of the influence of site characteristics on the
parameters describing a site's demand function, the attributes involved must
be identified. As indicated in Section 7.4, the information on the site charac-
teristics was obtained from U.S. Army Corps of Engineers. These data were
augmented with information on water quality from the U.S. Geological Survey.
As a rule, the Corps of Engineers data were measures of the size of the area
and types of equipment available. The water quality information consisted of
monthly readings from June through September of the year of the survey for
seven measures of water quality, including dissolved oxygen, fecal coliform
density, pH, biochemical oxygen demand, phosphates, turbidity, and total
suspended solids. Two water quality indexes were also developed from these
data for each month—the RFF water quality index (see Vaughan in Mitchell
and Carson [1981]) and the NSF index. Since the specific features of these
indexes were described in Section 7.4, their definitions will not be repeated
here. Table 7-13 summarizes the primary site characteristics considered from
the Corps of Engineers data.
Unfortunately, there are few a priori insights one can derive from
economic theory regarding which subset of these variables is most likely to
influence the estimated parameters of site demand models. While the primary
focus was on the water quality measures, the analysis considered a number of
alternative specifications, including subsets of the site characteristics reported
in Table 7-13. The variables with the most consistent association with the
demand parameters over the specifications considered included a measure of
the size of the site (i.e., SHORMILE), its access points (i.e., MULTI +
ACCESS), and the size of the water body relative to the overall site size
(i.e., AREAP/AREAT). This selection does not seem particularly surprising.
Each variable can be interpreted as a crude measure of the capacity of the
*Appendix F presents the benefit estimates if all 33 sites are used in the
model.
7-52
-------
Table 7-13. Description of U.S. Army Corps of Engineers
Data on Site Characteristics
Variable name
Description
SHORMILE
AREAT
AREAP
MULTI
ACCESS
CORPICK
OTH PICK
CORCMPD
OTH CMPD
CORLN
OTH LN
DOCK PR
DOCKCO
FLOAT
Total shoreline miles at the site during peak
visitation period
Total site area, land and water in acres
Pool surface acreage on fee and easement
lands during peak visitation period
Number of developed, multipurpose recrea-
tion areas onsite
Number of developed onsite access areas
Number of Corps-managed onsite picnic
locations
Number of other agency-managed onsite
picnic locations
Number of Corps-managed developed camp
sites
Number of other agency-managed developed
camp sites
Number of Corps-managed onsite boat launch-
ing lanes
Number of other agency-managed onsite boat
launching lanes
Number of onsite private boat docks
Number of onsite community docks
Number of onsite floating facilities (e.g.,
water ski jump, swimming floats, fishing
floats, etc.)
7-53
-------
site to provide services that would support different types of recreation
service flows.
It was more difficult to isolate measures of water quality that appeared
to influence the estimated site demand parameters. While the final generalized
least-squares estimates for the determinants of site demand parameters seem
exceptionally good, there are a number of reasons for caution in interpreting
these findings, as shown by a review of the approaches used to develop
them.
The modeling of the role of water quality considered a wide array of
potential specifications of its effects, including each of the following:
The monthly and average (across the 4 months of the summer
season) readings for the two water quality indexes and meas-
ures of the variation in the index over the 4 months were
considered.
The monthly and average readings for specific components of
the index (i.e., dissolved oxygen, total suspended solids,
etc.) were considered individually and in sets using existing
information, where possible, to avoid the joint presence vari-
ables that might be measuring common phenomena.
Temporal effects of individual pollutants were considered in an
attempt to isolate "best" or most relevant indexes of water
quality.
With a few notable exceptions these results led to either insignificant or un-
stable estimates of the effects of water quality on the site demand parameters.
Only in the case of dissolved oxygen did this pretesting of model specifi-
cations lead to a stable and statistically significant association between the
variation in the estimated site demand parameters and the mean and variance
in the level of dissolved oxygen over the summer period. This association is
more clearcut with the smallest samples. Clearly, these findings are consistent
with the earlier Vaughan-Russell [1981] and Nielsen [1980] analyses supporting
the use of dissolved oxygen as an ideal measure of water quality for evaluating
recreation fishing. Nonetheless, it should be acknowledged that the missing
data problem is especially important for this study's water quality variables
(see Section 7.4 above). The procedure has been to use the sample mean for
those sites with missing water quality information. Thus, a smaller number of
actual readings on water quality are what should be regarded as the basis of
the measured association between water quality and the estimated site demand
parameters. This does not imply that the use of means was inappropriate.
Rather, it indicates that there was little observed variation in any of the
water quality variables to associate with the estimated demand parameters.*
The indexes of water quality (i.e., the RFF and NSF) tend to reduce
the variation present in their components. Thus, there was very little varia-
tion in these indexes across sites.
7-54
-------
Approximately half of the 22 sites in the restricted sample had complete water
quality information. Thus, the preference for the dissolved oxygen measure
might well be altered with more complete water quality data.
Table 7-14 reports the generalized least-squares estimates for the final
model with both samples.* The parameters, a0, alf and a3, correspond to
the general model specifications as given in Equation (7.31). These results
clearly favor the model based on the restricted sample. Increases in the
average level of dissolved oxygen would be improvements in water quality.
The results using this restricted sample indicate that such increases would
increase the demand at all implicit prices (i.e., travel costs) and would also
increase the degree of inelasticity in the demand curve. This second effect
simply reflects the site's ability to support a wider range of recreation activi-
ties with the improved water quality.
Given the poor performance of income as a determinant of the demand for
any one of the site's services, it is not surprising that the second step model
for the income parameter is incapable of explaining the variation in the site
demand parameters.
The most striking difference between the results estimated with the two
samples arises with the estimated coefficients for the travel cost variable. The
estimated effects of the site attributes, including the water quality measures,
are all significantly different from zero and generally consistent in sign with
a priori expectations. The differences between the two samples would seem to
provide indirect evidence of the importance of truncation effects on the travel
cost site demand models.
These generalized least-squares results do not include R2 measures of
goodness of fit because the coventional R2 statistic is no longer confined to
the 0 to 1 interval when calculated based on the generalized least-squares
residuals. Thus, it does not have the same interpretation as the R2 statistics
reported with the ordinary least-squares results (see Cicchetti and Smith
[1976] Appendix B for more details).
*See Section 7.3 for a detailed discussion of the construction of the
generalized least-squares estimator. It should be noted that Vaughan and
Russell [1981] have used a similar methodology in their valuation of recreation
fishing days. However, their approach combined the two equations by substi-
tuting the second step model for the determinants of site demand parameters
(Equation 7.22) into Equation (7.21) to derive:
Y. = X.6A. + e. .
This model includes interaction terms in the determinants of site demands and
site attributes. It provides an equivalent description of the two-step approach
used in this study. However, there is one advantage to the two-step approach
in specification analysis of the models. It allows the specification of the
determinants of site demand to be treated separately from the determinants of
variations in site demand parameters. Each specification for the combined
model includes assumptions about both!
7-55
-------
Table 7-14. Generalized Least-Squares Estimates of Determinants of Site Demand Parameters
an a, a.
•sj
1
in
0)
Independent
variable
Intercept
SHORMILE
(MULTI + ACCESS)
AREAP/AREAT
Mean dissolved
oxygen
Variance in
dissolved oxygen
33 site
1.2959
(3.768)
-0.0003
(-1.304)
0.0017
(0.464)
-0.1686
(-1.116)
0.0049
(1.220)
0.0003
(1.131)
22 site
1.5106
(4.081)
0.0003
(1.250)
-0.0059
(-1.502)
-0.3950
(-1.752)
0.0045
(1.065)
0.0005
(1.862)
0
(0
0
(0
-0
(-1
-0
(-2
-4
(-1
-0
(-0
33 site
.0005
.203)
-6
.47 x 10 °
.256)
.41 x 10"4
.586)
.0025
.190)
.2 x 1Q~4
.514)
.17 x 10~5
.751)
-0
(-9
-0
(-6
0
(2
0
(2
0
(5
0
(4
22 site
.0246
.480)
-4
.13 x 10
.763)
.77 x 10~4
.810)
.0033
.273)
.0002
.992)
.98 x 10~5
.077)
0
(0
-0
(-1
0
(1
0
(1
-0
(-0
-0
(-0
33 site
.53 x 10
.330)
-7
.14 x 10 '
.408)
.22 x 10~6
.299)
.10 x 10"
.423)
.12 x 10
.642)
.73 x 10~8
.617)
22
0.54 x
(0.308)
0.97 x
(0.089)
0.47 x
(2.562)
-0.19 x
(-0.181)
-0.12 x
(-0.604)
0.94 x
(0.007)
site
10~5
-9
10
10'6
10"5
10~6
io-10
aThe numbers in parentheses below the estimated coefficients are the asymptotic t-ratios for the null
hypothesis of no association.
-------
7.8 A MEASURE OF THE BENEFITS OF A WATER QUALITY CHANGE
The objective of the analysis of recreation behavior has been to develop
a model capable of measuring the benefits associated with improving the water
quality for any site that provides water-based recreation activities. Given
information on the site characteristics found to be important determinants of
site demand, it is possible to use the model for each demand parameter to
estimate a "representative individual's" demand function for the desired water-
based facility. Consequently, this section reports the results of such an
application using information on the 13 sites along the Monongahela River that
were used by the contingent valuation survey respondents.
Table 7-15 provides a description of these sites and their attributes.
The model estimates the representative individual's demand for each site's
services. Because the survey asked each respondent about his use of the
river, including an identification of the site (or sites) used, it is possible to
develop an estimate of these demand functions for each site. Moreover,
because the model includes water quality information, the change in these
demands can be estimated to accompany each of the water quality changes
used in the survey instrument. In Chapter 8, this information provides the
basis for. a comparison of direct and indirect methods for measuring the
Table 7-15. Recreation Sites on the Monongahela River
Identification MULTI AREAP/
Site name number SHORMILE + ACCESS AREAT
Pittsburgh area
The confluence of the
Youghiogheny and Monongahela
Rivers
Elrama
Town of Monongahela
Donora and Webster
Near Charleroi
California and Brownsville
Maxwell Lock and Dam
Point Marion
Morgantown
Fairmont
9th Street Bridge
Cooper's Rock
15
16
17
18
19
20
21
23
25
26
29
37
44
2
2
2
3
2
3
12
2
2
4
3
1
2
1
2
2
4
1
4
6
7
1
2
1
1
1
0.99
0.99
0.99
0.99
0.99
0.96
0.96
0.93
0.99
0.77
0.67
0.99
0.99
SOURCE: U.S. Army Corps of Engineers Resource Management System.
7-57
-------
benefits from a water quality improvement. The direct methods correspond to
the results from the survey, while the indirect methods use the information
on the survey respondents' recreation behavior together with the generalized
travel cost model developed in this chapter.
Seventy-five of the survey respondents were users of 1 or more of the
13 sites along this section of the Monongahela River. Because several individ-
uals used more than one site, there are a total of 94 observations identified
as an individual/site combination. These data provide the basis for construct-
ing 94 separate demand models to evaluate the implications of water quality
changes as measured by dissolved oxygen. For example, the model implied
that the estimated price elasticities of demand (at the average travel costs for
users in the survey used for the contingent valuation experiment) for the 13
recreation sites along the Monongahela River—the area for the contingent
valuation survey—ranged from -0.069 to -0.075 at current water quality
levels. Improving the water quality to permit game fishing would imply a
change in DO from 45 to 64 (percent saturation). These changes reduce the
absolute magnitude of the estimated demand elasticities to -0.052 to -0.059.
The benefits from a water quality improvement are measured by the
increment to the ordinary consumer surplus experienced by each individual.*
This increment can be defined for each individual user as follows:
/•Pfk /*Pfk
Bjk = / Fjk(p, WQ*)dp - / Fjk(p, WQ) dp (7.34)
P P
Jk Pjk
where:
P.. = travel cost (mileage plus travel time) experienced by the
J jth user to kth site
P* = maximum price the jth user would be willing to pay for
^ the kth site's services (i.e. where the quantity demanded
is zero)
WQ* = improved water quality level
WQ = initial water quality level (i.e., WQ* > WQ)
F. (.) = demand function for the kth site's services by the jth
Jk user.
*The measurement of the benefits from water quality improvement has
ignored the potential for congestion effects. It has been assumed that conges-
tion is negligible both before and after the change in water quality. Without
this assumption, the implications of management practices would need to be
considered in the definition of the benefit measure (see McConnell and Sutinen
[1983]).
7-58
-------
Implementation of this benefit estimator required several amendments. The
specification of the site demand functions in semilog terms implies that they
will not have a price intercept. Rather, they asymptotically approach the
horizontal (price) axis. To estimate a finite consumer surplus, a maximum for
the price, P* , was selected to correspond to the maximum travel cost paid by
J *^
any of the survey users of any Monongahela site. The specific value was
$22.65 for a roundtrip, including both the mileage and time costs of travel.*
This modification implies the benefit estimates for the water quality improve-
ment will be ABCD as given in Figure 7-1, with P. corresponding to the jth
user's travel costs and P* the maximum value for the travel cost.
VisiU/yr
Price
(S/visit)
Figure 7-1. Measurement of consumer surplus increment due to
water quality improvement (WQ to WQ*}.
Table 7-16 details the dissolved oxygen levels associated with each of
three use designations (see Vaughan in Mitchell and Carson [1981]) employed
in the calculations rather than the actual water quality levels for the sites
along the river. The reason for this approach follows from the key project
objective—to compare benefit estimates based on the travel cost models with
those based on the survey responses. All survey respondents were told that
the water quality was consistent with boatable conditions. Thus, the corre-
sponding value for dissolved oxygen was used as the base value for the esti-
mates. Because the model requires a mean level of dissolved oxygen for the
*This maximum travel cost is generally smaller than the maximum travel
costs experienced by the Federal Estate Survey respondents used to estimate
the generalized travel cost model. Indeed, it is less than the majority of the
sample means of the FES travel costs (see Table 7-3).
7-59
-------
Table 7-16. Dissolved Oxygen Levels for
Recreation Activities
Assumed level
of dissolved
Use designation
Boatable water conditions
Fishable water conditions
Swimmable water conditions
These estimates for dissolved oxy
oxygen required
45
64
83
gen are based
on
Vaughan in Mitchell and Carson [1981].
summer recreation season, the means were assumed to correpond to each of the
levels given in Table 7-16. The variance in monthly levels of dissolved oxygen
was set at the sample mean for the sites used to estimate the model—8.187--and
was assumed to be unaffected by water quality changes.
Table 7-17 presents the mean values for the incremental benefits associ-
ated with three types of changes in water quality conditions:
An assumed deterioration in water quality making it unavailable
for boating or other recreation activities.
An improvement in water quality from boatable conditions to
fishable conditions.
An improvement in water quality from boatable conditions to
swimmable conditions.
All three of these changes were assumed to take place at all 13 of the Monon-
gahela sites. The first was treated as the equivalent of losing the use of
the recreation site completely. The benefit loss was measured as the con-
sumer surplus associated with the site under boatable conditions—P.ADP* in
Figure 7-1.
The remaining two scenarios correspond to different levels of the new
demand functions for the water quality associated with fishable and swimmable
conditions. Table 7-17 presents the mean consumer surplus increment for
each of the three changes for our 94 user-site combinations. It also reports
the range of values for the increment to consumer surplus. The mean benefits
correspond to the increase in an "average" individual's willingness to pay
over the recreation season. The average user in the survey used one or
more Monongahela sites 7.22 times. Thus, the loss of the site completely
translates to a loss of $7.39 per unit in 1977 dollars, or $11.46 in 1981 dollars,
the date of the contingent valuation survey.*
This adjustment used the consumer price inaex (CPI) for all commodities.
Using a 1967 base, the 1977 CPI for all items was 181.5. In 1981 it closed
the year at 281.5. See Economic Report of the President 1982, Council of
Economic Advisors [1982],
7-60
-------
Table 7-17. Mean and Range of Benegt Estimates for
Water Quality Scenarios
Minimum Maximum
Water quality change Mean value value
Scenario (1) $53.35 $0.00 $70.80
Loss of use of site (7.39)
under boatable
conditions
Scenario (2) $4.52 $0.00 $8.60
Improvement of (0.63)
water quality from
boatable to fishable
conditions
Scenario (3) $9.49 $0.00 $18.30
Improvement of (1.31)
water quality
from boatable to
swimmable conditions
aThese calculations are in 1977 dollars, the year of the
Federal Estate Survey.
The numbers in parentheses below the overall increment
report the corresponding consumer surplus increment on
a per visit basis.
Because these benefits estimates are available for each of the 94 user/site
combinations') the estimates in several classifications were alsa tabulated—by
size of family income reported by the respondents and by the magnitude of
their travel costs. The results for the consumer surplus loss due to loss of
the use of the river for boating are given in Tables 7-18 and 7-19. The
results for each of the two increments to water quality compared with income
are given in Tables 7-20 and 7-21. It should be noted that the income levels
are in 1981 dollars while the consumer surplus increment is in 1977 dollars.
Scaling the latter by 1.55 will convert them to equivalent dollars. Since it
was a simple multiple of the estimates and would not change the distributions,
they were not converted for these tables.
These results indicate that it is possible to use a generalized form of the
travel cost model to estimate the benefits from a water quality change. By
using the recreation use patterns for a number of sites, it was possible to
develop a general model that, in principle, is capable of being used to estimate
the recreation benefits associated with water quality changes at any site
providing similar water-based recreation activities.
7-61
-------
Table 7-18. Consumer Surplus Loss Due to Loss of Use of the
Monongahela River by Survey Users' Income
Consumers surplus loss (1977
1 ncome
(1981 dollars) 0-10
0-5,000
5,000-10,000
10,000-15,000
15,000-20,000
20,000-25,000 1
25,000-30,000 1
30,000-35,000
35,000-40,000
40,000-45,000
45,000-50,000
50,000 and above --
Total 2
10-20 20-30
1
_.
—
1 1
1 1
2
—
—
—
—
—
2 5
30-40
2
--
--
2
--
2
3
—
1
--
--
10
40-50
--
--
--
--
1
3
--
--
1
--
--
5
50-60
4
2
2
3
1
8
2
2
--
3
2
29
dollars)3
60-70
3
7
6
11
1
6
2
1
1
1
--
39
70-80 Total
10
2 11
8
18
6
22
7
3
3
4
2
2 94
To convert to 1981 dollars multiply the endpoints of the benefit scale by
1.55.
Table 7-19. Consumer Surplus Loss Due to Loss of Use of the
Monongahela River by Survey Users' Travel Cost
Travel
cost
dollars) 0-10
0-5
5-10
10-15
15-20
20-25 2
Total 2
Consumer surplus loss (1977 dollars)
10-20
-
-
-
2
-
2
20-30
-
-
4
1
-
5
30-40
-
2
8
-
-
10
40-50 50-60
19
5 10
-
-
-
5 29
60-70 70-80 Total
39 2 60
17
12
3
2
39 2 94
7-62
-------
Table 7-20. Consumer Surplus Increments Due to Water Quality Improvement--
Boatable to Fishable by Survey Users' Income
Consumer surplus increment
Income
(1981 dollars)
0-5,000
5,000-10,000
10,000-15,000
15,000-20,000
20,000-25,000
25,000-30,000
30,000-35,000
35,000-40,000
40,000-45,000
45,000-50,000
50,000 and
above
Total
0-10
_
-
-
-
1
1
-
-
-
-
2
4
10-20
-
-
-
2
2
3
4
3
3
4
-
21
20-30
3
-
-
16
3
18
3
-
-
-
-
43
(1977 dollars)3
30-40
7
11
8
-
-
-
-
-
-
-
-
26
Total
10
11
8
18
6
22
7
3
3
4
2
94
To convert to 1981 dollars multiply the endpoints of the benefit scale by
1.55.
Table 7-21. Consumer Surplus Increment Due to Water Quality
Improvement—Beatable to Swimmable by Survey Users' Income
Consumer surplus increment (1977 dollars)3
1 ncome
(1981 dollars)
0-5,000
5,000-10,000
10,000-15,000
15,000-20,000
20,000-25,000
25,000-30,000
30,000-35,000
35,000-40,000
40,000-45,000
45,000-50,000
50,000 and
above
Total
0-10
_
-
• -
-
1
-
-
-
-
-
2
3
10-20 20-30
_ „
-
-
1
2
1 3
3
3
3
4
-
5 15
30-40
1
-
-
3
1
18
4
-
-
-
-
27
40-50
1
-
-
6
2
-
-
.
-
-
-
9
50-60
2
2
8
8
-
-
-
_
-
_
-
20
60-70
6
9
-
-
-
-
-
_
-
_
_
15
Total
10
11
8
18
6
22
7
3
3
4
2
94
a
convert to 1981 dollars, multiply the endpoints of the benefit scale by
1.55.
7-63
-------
7.9 SUMMARY
The findings from the application of the travel cost approach are of
equal, if not greater, importance. The research in this project developed a
generalized travel cost model that predicts the recreation benefits of water
quality improvements at a recreation site. Estimating the benefits for users
of the Monongahela River, the travel cost model predicted benefits of $83 per
year for a user if a decrease in water quality is avoided. Water quality
improvements to swimmable water in the Monongahela were estimated at $15
per year (in 1981 dollars).
Two features of the generalized travel cost model are of particular impor-
tance. The model can be applied to predict the value of water quality im-
provements for a substantial range of sites, and it is especially relevant for a
large number of water quality standards applications. Including the effect of
key site features in addition to water quality-like access and facilities—and
relying on data frequently available in the public domain makes the model a
viable tool for future benefits applications.
7-64
-------
CHAPTER 8
A COMPARISON OF THE ALTERNATIVE APPROACHES FOR ESTIMATING
RECREATION AND RELATED BENEFITS
8.1 INTRODUCTION
One of the primary objectives of this research has been to compare avail-
able methods for measuring benefits of water quality improvement. Of course,
the "true" value of benefits associated with a specific increment of water qual-
ity can never be known, and a comparison of measurement methods cannot be
interpreted as a validation of any one of them. Nonetheless, it is important
to recognize that contingent valuation methods for estimating the benefits of
environmental quality improvements are viewed with considerable skepticism by
many (if not most) economists. Presumably, these economists assume that in-
dividuals will experience difficulty in responding to valuation questions for
nonpriced goods and that their responses will exhibit significant strategic bias.
By contrast, indirect methods have been more favorably regarded by most
economists, and this study's use of benefit estimates derived from one in-
direct method—the travel cost recreation demand model—as a benchmark for
the contingent valuation estimates reflects this perspective. Of course, it
should be recognized that indirect and direct benefits measurement approaches
can be distinguished according to the assumptions each makes and that a com-
parison of them reflects in part the plausibility of their assumptions as
descriptions of real-world behavior and constraints.
To aid in the interpretation of the comparisons of benefit estimation ap-
proaches, this chapter highlights the specific features of the approaches and
how they are applied in this study. The Monongahela River case study pro-
vides the basis for the evaluation of the approaches. The types of possible
evaluations are bounded by its scope. More specifically, Section 8.2 of this
chapter introduces the conceptual basis for a comparative evaluation of benefit
estimation approaches. Following this discussion, Section 8.2 also relates the
evaluation scheme used in this chapter to that used in earlier comparisons,
including those of Knetsch and Davis [1966], Bishop and Heberle'rn [1979],
and Brookshire et al. [1982]. Section 8.3 discusses the results of the com-
parison of approaches, including the findings of a numerical comparison of the
mean estimates of the user and intrinsic components of benefits for specific
water quality changes by methodology. This discussion is followed by pair-
wise comparisons of the contingent valuation and travel cost methods and of
the contingent valuation and contingent ranking methods. Finally, Section
8.4 summarizes the findings and discusses their implications for the practical
use of benefit measurement approaches.
8-1
-------
8.2 THE CONCEPTUAL FRAMEWORK FOR A COMPARISON OF RECREATION
BENEFIT ESTIMATION APPROACHES
8.2.1 Background
Improvements in water quality associated with water bodies that support
recreation activities can lead to both user and intrinsic nonuse benefits. User
benefits arise because water quality can be expected to affect the types of rec-
reation activities at the site experiencing the changes. Individuals who wish to
participate in activities made possible by the improvement will be able to, thus
enhancing their levels of economic well-being. User benefit estimates of water
quality improvements attempt to measure the magnitude of these changes in
well-being. Intrinsic benefits, on the other hand, arise either because indi-
viduals are uncertain of their potential use of a site or because they experience
enhanced utility merely from knowing of improved site conditions. The first
recognition of the importance of intrinsic benefits has most often been associ-
ated with Krutilla's [1967] discussion of the rationale for public involvement in
the management of natural environments. Intrinsic benefits have been identified
under a variety of classification schemes to include option and existence values.
Because preceding chapters have presented detailed discussions of both
user and intrinsic benefits, the definitions of each are not repeated here.
Rather, this chapter considers the relationship between benefit estimation ap-
proaches and the two benefit classes. This relationship is important because it
affects the types of comparisons that can be undertaken across approaches.
The measurement of the economic benefits of water quality improvement
requires a mechanism for linking the water quality change to a consistent meas-
ure of benefits. As noted in Chapter 2, this linkage provides one basis for
classifying methods used to measure benefits of a change in any environmental
amenity not exchanged in an organized market. While Chapter 2 identifies sev-
eral types of assumptions that provide these links, two classes of assumptions
are especially relevant to the approaches considered in this project for benefit
measurement.
The first relevant class of assumptions involves the use of the technical
association between water quality and recreation site services. Use of a water
body's recreation services involves a corresponding (and, indeed, simultaneous)
use of the water quality at the site. Thus, the types of activities that can be
undertaken at a particular site are affected by the site's water quality (a point
explicitly made throughout the analysis in Chapters 4 through 7). Given both
a behavioral model to describe how individuals allocate their resources and ex-
ogenous measures of their use of recreation sites with differing levels of water
quality, this approach maintains that it may be possible to estimate individuals'
willingness to pay for water quality indirectly. This recognition is, of course,
the basis for the approach used in the generalized travel cost model developed
in Chapter 7.* However, more important for comparing measurement approaches
This model assumes that each set of users for each of the sites included
in our sample from the Federal Estate Survey acts as the "representative" in-
dividual would under the circumstances defined by the site's availability and
the survey respondent's economic characteristics.
8-2
-------
is that this approach—using "indirect" technical linkages between water quality
and recreation site services—only measures user values.
The second relevant class of assumptions, identified in Chapter 2 as insti-
tutional assumptions, explicitly recognizes that ideal markets would provide the
benefit measures required for any good or service, providing the good could
be exchanged in them. However, attempts to estimate the valuation of such
environmental amenities as water quality face difficulties because ideal markets
are not available. Thus, the contingent valuation approach—using "direct"
institutional linkages—assumes that, if individuals are confronted with a hypo-
thetical market (in the form a survey questionnaire) for these amenities, their
responses will measure their true valuation of the resources (or amenities) in-
volved. Thus, the contingent valuation approach assumes it is possible to
mimic the outcomes of ideal markets by completely describing the conditions of
exchange in a hypothetical market for the service to be valued. As a result,
these methods assume that an individual's responses to the conditions presented
in this hypothetical market will be equivalent to the actual responses that would
be made if the exchanges took place in actual markets. Since the market is
simply an institution, a hypothetical market can be defined to suit any partic-
ular nonmarketed service and does not require that it actually be feasible to
exchange the services described. Thus, contingent valuation methods can
measure both user and intrinsic benefits.
In comparing the two classes of assumptions and the approaches for bene-
fit measurement arising from them, it is important not to confuse the flexibil-
ity of the approaches using institutional restrictions with judgments that these
approaches require less stringent assumptions.* Alternative approaches re-
quire different assumptions. Therefore, appraisals of the severity of one ap-
proach's assumptions relative to another's should be regarded as individual
judgments, not necessarily as objective comparisons.
8.2.2 Research Design and Comparative Analysis
The research design of this project permits several types of comparisons.
Chapters reporting each approach's estimates have discussed the first type—
those within a benefit estimation framework. For example, the contingent valu-
ation survey was designed to consider five different approaches for eliciting an
individual's valuation of water quality changes. In four of these approaches,
only the valuation question differed:
A direction question
A question using a payment card
*The classification scheme for benefit estimation methods given by Schulze,
d'Arge, and Brookshire [1981], pp. 154-155, is somewhat misleading in that it
implies the contingent valuation approach has the least a priori assumptions.
While this is true as a description of the assumptions concerning constraints to
actual behavior, it ignores the implicit assumption that responses to hypothet-
ical institutions will provide a good, guide to the responses made to the actual
institutional arrangements implied by their "constructed" markets.
8-3
-------
The conventional iterative bidding framework with a $25 start-
ing point
The conventional iterative bidding framework with a $125 start-
ing point.
Each questioning format was applied to an approximately equal proportion of the
sample and provides independent estimates of an individual's valuation of the
specified water quality changes. Because the design of the questions elicited
the individual's option price and user values, comparisons of these questioning
formats were undertaken for the estimates of option price, user value, and
option value with the results described in detail in Chapters 4 and 5.
This chapter focuses on comparisons between benefit estimates across
methodologies—e.g., travel cost vs. contingent valuation. These comparisons
will also involve the effect of question format, but the effect of format may
differ from the within methodologies comparison because the standards for the
comparisons are different. Equally important, the comparisons across methods
cannot consider each method's performance in measuring combined user plus
intrinsic benefit (i.e., option price) as well as their separate estimates (e.g.,
in form of option value). The travel cost method measures only user value,
and the contingent ranking only a composite of the two.
The specific details of the within method comparison involved two types
of evaluations:
Statistical tests for the differences in means between all pairs
of question formats for the full sample and for users and non-
users of the Monongahela River.
Multivariate regression analysis, including dummy variables for
the question formats along with other prospective determinants
of the relevant dependent variables.
The option price results exhibit the most differences among question for-
mats, with some evidence of a starting point bias. The regression models also
exhibit the most cases of significant effects for the question format variables in
this case. This finding contrasts with several (but not all) of the past con-
tingent valuation studies.* With the option value estimates there is also some
evidence of starting point bias, but these findings are not as pronounced as in
the analysis of the option price estimates. These differences are not necessar-
ily surprising since only the first stage of the individual's response (i.e., the
option price) had distinct questioning formats. Thereafter, the questions
calling for separation of the option price into components (i.e., user values)
were (by practical necessity) direct questions.
The Schulze, d'Arge, and Brookshire [1981] summary concludes, based
on an analysis of several contingent valuation experiments, that starting point
bias is not a serious problem. Our results do not conform to this conclusion
and indicate that the prescreening of data used to eliminate inconsistent obser-
vations may affect their conclusions. Of course, It should also be emphasized
that our results relate only to a single experiment.
8-4
-------
Finally, the results are quite sensitive to the screening of observations
judged to be refusal to participate in, or inconsistent with, the contingent
valuation framework. As noted in Chapter 4, while procedures used to identify
these observations are based on a statistical index of the influence of each
individual observation (and are therefore capable of replication), the effects
of specific socioeconomic characteristics of the survey respondents cannot be
distinguished from the question format (see Table 4-8 in Chapter 4). Thus,
the results for starting point bias and for other pairs of question formats
(iterative bidding with $125 starting point) would have been more pronounced
with the inclusion of the observations judged to be inconsistent with the con-
tingent valuation framework.
Comparisons across approaches are limited because the methods do not
uniformly measure the same components of the benefits associated with a water
quality improvement. As we noted earlier, the contingent valuation method
design measures both user and intrinsic benefits and permits these estimates
to be separated. By contrast, the travel cost and contingent ranking meth-
ods are more limited. The travel cost approach measures only user values
(i.e. ordinary consumer surplus). The contingent ranking design measures
option price but does not divide the estimates into the user value and option
value. Therefore, comparisons here are limited to examining the relationship
between the user value estimates of the contingent valuation and travel cost
approaches and the option price estimates for contingent valuation and con-
tingent ranking.*
The comparison of the estimated user values derived using the contin-
gent valuation approach (with all four question formats) and the consumer sur-
plus estimates derived from the generalized travel cost model is the most
interesting comparison. It provides an extension to the recent work of Brook-
shire et al. [1982] for the valuation of air quality using hedonic property
value and contingent valuation methods.
Using a subset of the survey respondents who visited specific Mononga-
hela River sites to derive consumer surplus estimates from the generalized
travel cost model (presented in Chapter 7) allowed a matching of each re-
spondent's expressed user value for a comparable water quality change with
the values predicted from the travel cost model. This comparison of the
travel cost and contingent valuation methods can be made for each user in
this survey, in contrast to the Brookshire et al. [1982] analysis.t Thus,
both the mean estimates derived from the two approaches and the association
in the estimates can be compared across individual users.
*For the sake of simplicity in the use of terms in this chapter contingent
valuation is used to refer to the four question formats in the contingent valu-
ation survey. While contingent ranking is a subset of contingent valuation
(and this distinction was made in Chapter 1), the easier terminology of con-
tingent ranking vs. contingent valuation is used in this comparison chapter.
tThis is one of the aspects of our extension over this work. A second
involves replacing the broad bounds for contingent valuation estimates with a
potentially more restrictive upper threshold.
8-5
-------
Several features limit the ability to compare the estimates derived from
the travel cost and contingent valuation methods. The simplest of these fea-
tures is different dollar values in each method because the travel cost model
was developed with 1977 dollars and the contingent valuation estimates was
developed with 1981 dollars. Using the consumer price index, an adjustment
can approximately account for this difference. A more important reason for
differences stems from what is being measured. The user values derived
using contingent valuation methods estimate an individual's expected willing-
ness to pay or compensating surplus (for improvements in water quality),
while the generalized travel cost model estimates ordinary consumer surplus.
A long literature on the theoretical foundations of consumer surplus estimates
has suggested that there are good reasons why these two measures should di-
verge.* However, for price (Willig [1976]) and quantity (Randall and Stoll
[1980]) changes, the difference between the two measures can be bounded
under specific conditions (see Chapter 2 for a brief review).
At first, the comparison of welfare measures in this project might seem
to involve a case that falls outside the scope of the bounds, because it in-
volves a change in water quality rather than a price or quantity change.
Fortunately, this conclusion is premature. One of the assumptions used to
develop the generalized travel cost model—that water quality augments the
effect of a recreation site's services in the production of recreation activities
(see Equations (7.11) and (7.12) in Chapter 7)--implies that a water quality
change can be translated into an equivalent change in either the quantity of
a site's services or in the "effective" price of using the site (see Equations
(7.12) and (7.13), respectively).t Therefore, for changes in water quality
*See Just, Hueth, and Schmitz [1982] for further discussion.
tin general terms the consumer surplus increment due to a water qual-
ity change, w, with a demand function Q - F(P,w) (P = price, Q = quantity)
is given as
/P* /»p*
F(p.,w2)dp - J F(p.,w1)dp
P: Pi
i
where
CS. = consumer surplus for individual facing price P.
P* = price at which the quantity demanded would be zero
w2 = improved level of water quality
wj = existing level of water quality.
The form of the household production technology assumed in the development
of our travel cost model implies that a change in water quality can be con-
sidered equivalent to a change in the quantity of or price of a site's serv-
ices. This implies that the change from wt to w2 can be treated as equiva-
lent to some change in the price of a site's services from P(wx) to P(w2).
8-6
-------
that translate into relatively small price (quantity) changes, the Willig (Randall
and Stoll) bounds can be applied to judge the relationship between Marshallian
consumer surplus and the willingness to pay for a water quality change.
From a practical perspective, one might assume that the discrepancies be-
tween the Marshallian consumer surplus and the willingness to pay for a water
quality change associated with recreation water sites would be small. Most
households' expenditures on water-based recreation activities would be a very
small fraction of their income. This judgment is also supported by the esti-
mated travel cost demands developed for this study in that they imply income
is not a significant determinant of the demand for the services of water-based
sites comparable to sites on the Monongahela River. Thus, the difference be-
tween the willingness to pay and the consumer surplus for a comparable change
in water quality can be expected to be less than 5 percent.* The evidence
necessary for judging the implications of income for survey respondents who
were users of the Monongahela River can be derived using the same type of
information required by the travel cost model.t That is, because individual
estimates of the ordinary consumer surplus require travel cost and income in-
formation, these variables were combined with the respondents' reported use
patterns for the Monongahela sites, thus treating all 13 sites as if they shared
a common demand function, even though the generalized travel cost model does
not require this assumption. These data permit the estimation of a travel cost
model for the Monongahela in its current state. The results are given in
Equation (8.1) below:
In V = 0.7983 - 0.0195 (T+M) cost + 0.000015 income (8.1)
(3.153) (-0.785) (1.636)
R2 = 0.032
The numbers in parentheses are the t-ratios for the null hypothesis of no
association. These results indicate that income is not a significant determi-
nant of user trips to the Monongahela sites. Therefore, these findings would
be consistent with judgments based on the generalized travel cost model, and
willingness to pay would be expected to be less than the Marshallian consumer
surplus for water quality improvements (the equivalent, in the generalized
*See Freeman [I979a] or Just, Hueth, and Schmitz [1982] for a complete
discussion of the implications of the Willig [1976] bounds for applied benefit
analysis.
tThe generalized travel cost model assumes that a water quality change
can be translated into either an equivalent price or quantity change. Thus,
the site demand equation is the relevant basis for judging income responsive-
ness. Survey responses for compensating surplus (referred to as user value
in Chapter 5) are expected to provide equivalent results if these two sets of
information provide consistent descriptions of the individuals' demand charac-
teristics. An examination of the role of income in the user value equations
confirms this a priori expectation. The coefficients estimated for income are
never judged to be statistically significant determinants of user values.
8-7
-------
travel cost model, of price decreases for the site's services or quantity in-
creases). However, these estimates of income effects imply that the differ-
ence between willingness to pay and consumer surplus should be small.
These results can also be compared with the predicted demands for each
of the 13 Monongahela sites based on the generalized travel cost model and
the characteristics of each of these sites. Of course, this comparison cannot
be treated as an evaluation. The estimates given in Equation (8.1) are pooled
across sites and assume the demand parameters are invariant with respect to
site attributes. Nonetheless, the comparison may serve to identify whether
the implied demand features are completely incompatible with these crude esti-
mates available as a byproduct of the survey data. The focus is on the pa-
rameters of greatest influence for estimates of consumer surplus change in re-
sponse to a water quality change. Table 8-1 reports these predicted param-
eters for the intercept and coefficient of (T+M) cost for each of the 13 sites
under the assumption of boatable water quality. The absolute magnitude of
the price coefficient is in all cases smaller than any estimates based on the
survey, but they are reasonably close to the survey estimates. The intercept
predictions are substantially larger than the survey estimates.
The second comparison across benefit methodologies involves the contin-
gent valuation and contingent ranking approaches. Because all survey re-
spondents were asked one of the four types of contingent valuation questions
and the contingent ranking, the estimates from these approaches are not inde-
pendent estimates of the option prices for water quality changes. Indeed, it
is possible that an individual's responses to the contingent valuation questions,
which preceded the ranking questions on the survey instrument, influenced the
rankings. Therefore, this comparison reflects both the effects of the methods
used to estimate benefits and an individual's consistency in responding to com-
parable water quality increments in different formats.
Table 8-1. Predicted Demand Parameters for Monongahela Sites
Coefficient for
Site Intercept T+M cost
Pittsburgh area 1.323 -0.0133
Confluence of the 1.317 -0.0132
Youghiogheny and
Monongahela Rivers
Elrama 1.317 -0.0132
Town of Monongahela 1.306 -0.0131
Donora and Webster 1.323 -0.0133
Near Charleroi 1.317 -0.0132
California and Brownsville 1-308 -0.0131
Maxwell Lock and Dam 1.311 -0.0130
Point Marion 1.323 -0.0133
Morgantown 1.404 -0.0140
Fairmont 1.449 -0.0144
9th Street Bridge 1.323 -0.0133
Cooper's Rock 1.323 -0.0133
8-8
-------
8.2.3 Past Comparisons of Benefit Estimation Methods
Comparisons of the results of benefit estimation methodologies within the
context of a common problem have been quite limited. The first such com-
parison was undertaken by Knetsch and Davis [1966] and involved a bidding
game version of what is now commonly referred to as contingent valuation and
a form of the travel cost model. The survey was based on a sample of 185
users of a forest recreation area in northern Maine. With the iterative bid-
ding game, respondents were asked their willingness to pay (as increased
cost to visit the area). A similar format was used to elicit willingness to
drive to the area. Individuals were also asked the actual distance they
traveled to the site.
Knetsch and Davis compared three approaches for estimating the aggre-
gate benefits from the site. The first used a willingness-to-pay equation
based on the survey results to estimate a willingness-to-pay schedule for the
user population in the area surrounding the site. The two sets of distance
measures were each valued at $.05 per mile and used to derive aggregate
schedules for the user population. The aggregate benefit estimates derived
for each approach provided the basis for comparing the methods:
Contingent valuation $71,461
Willingness to drive $63,690
Travel cost $69,450
Because the contingent valuation approach measures willingness to pay
and travel cost measures the ordinary consumer surplus, the latter would be
expected to exceed the former at an individual level. However, it is difficult
to gauge the expected nature of the differences between the two methods for
these calculations because they involve the aggregate schedule over all indiv-
iduals and relate to changes in the price of the site comparable to a loss of its
availability for this population. As Bockstael and McConnell [1980] observe,
the Willig bounds may not hold where the analysis involves the removal of the
site. They observed that:
it is difficult to find single valued functions, x = f(p,m) [where
x = quantity demanded, p = price and m - income], decreasing in p
and increasing in m, such that:
§x
1. 3m is finite for all values of p and
x
2. the function f(p,m) must tend to zero rapidly enough with in-
creases in p that the integral of f(p,m) will be hounded when
evaluted as p goes to infinity, (p. 61)
Because Knetch and Davis do not present demand equation estimates with
their travel cost findings, it is difficult to evaluate the relationship between
their willingness to pay and consumer surplus estimates on an individual basis.
Their benefit estimates based on the willingness-to-travel responses are diffi-
8-9
-------
Table 8-2. Bishop-Heberlein Comparative
Results for Benefit Approaches3
Average
benefit estimate
Method per permit
I. Actual case offers $ 63
II. Hypothetical responses
(a) willingness to sell $101
(b) willingness to pay $ 21
III. Travel cost ordinary consumer $11 to $45
surplus (variation associated
with valuation of travel time
from 0 to \ median income rate)
These estimates are taken from Table 1 in Bishop and Heberlein
[1979], p. 929.
cult to interpret within the conventional welfare economics framework and thus
cannot be directly associated with either of the other benefit estimates. Thus,
while this study offered the first evaluation of benefit estimation approaches,
it did not permit a detailed comparative analysis of them.
The second comparative analysis was conducted by Bishop and Heberlein
[1979] and was primarily intended to evaluate the relationship between hypo-
thetical and actual responses to willingness-to-sell questions.* Their analysis
was conducted using goose hunting permits for hunters in Wisconsin. Three
samples of hunters were used in their analysis. The first sample received
actual cash offers for their permits (ranging from $1 to $200); a second sample
received questionnaires asking the individual's willingness to pay for (and wil-
lingness to sell) their permits; and a third sample received questionnaires de-
signed to permit the estimation of a traval cost demand equation. Table 8-2
summarizes the Bishop and Heberlein estimates per permit for each of the ap-
*Bishop and Heberlein describe a number of potential biases that might
distinguish hypothetical and actual responses to willingness-to-pay questions.
Some of these problems conform to the definitions used in the papers reporting
contingent valuation survey results. The most directly comparable case is
strategic bias. However, the Bishop-Heberlein approach does not attempt to
induce differential responses from individuals, by giving them, for example,
different information about the uses that will be made of their bids to hypo-
thetical changes. This approach has been the most common method for investi-
gating the potential for strategic bias in the contingent valuation experiments
(see Schulze, d'Arge, and Brookshire [1981]). Rather, their comparison of
actual and hypothetical responses will reflect a composite of any such biases
due to the "framing" of their hypothetical survey instrument and to the dis-
tinction between hypothetical and real conditions.
8-10
-------
proaches considered. Their findings suggest that hypothetical willingness-to-
sell estimates overstate actual responses. Moreover, Bishop and Heberlein
argue that hypothetical willingness to pay and ordinary consumer surplus esti-
mated with the travel cost demand model understate the actual willingness-to-
sell by more than the Willig bounds would imply.
The Bishop and Heberlein results, while limited to a single experiment,
have potentially important implications for the relationship between hypotheti-
cal and actual estimates of willingness to sell. They do not offer as much
guidance on the comparative properties of the benefit estimation methodologies
themselves. The authors' benefit estimates made with the travel cost model
can be interpreted (for one value for the opportunity cost of travel time) as
quite close to the hypothetical willingness to pay. However, because the
selection of an opportunity cost for travel time is treated as judgmental, more
specific conclusions are not possible. Finally, the Bishop-Heberlein research
design (i.e., the selection of independent samples for the hypothetical and
travel cost surveys) did not permit comparison of the hypothetical willingness
to pay and ordinary consumer surplus on an individual basis.
Most recently, Brookshire et at. [1982] provided comparative analysis of
benefit estimation methods, maintaining it as a validation analysis of the con-
tingent valuation methodology. As observed earlier, this reflects the inter-
pretation given to contingent valuation versus indirect benefit estimation
methods by many economists and is somewhat unfortunate. Each of the meth-
ods involved in the Brookshire et at. [1982] comparative evaluation is based
on different assumptions concerning the economic behavior of households and
the role of environmental amenities (i.e., air quality) in their decisionmaking.
Neither method provides the "true" benefit estimates for air quality improve-
ments .
The Brookshire et at. [1982] analysis compares a hedonic property value
model'to a contingent valuation approach for measuring the willingness to pay
for reductions in air pollution. The authors interpret the hedonic model as
providing an upper bound for willingness to pay and argue that the assump-
tions of the model are approximately satisfied for the Los Angeles area. At
issue in their comparison, however, is whether direct questions can be be-
lieved. They demonstrate if each method conforms to its respective assump-
tions, the annual rent differential for pollution should exceed estimates of the
annual willingness to pay.
Using paired areas in Los Angeles selected to be homogeneous with re-
spect to socioeconomic, housing, and community characteristics but with varia-
tion in air pollution, Brookshire et al. [1982] tested two hypotheses:
The rent differential for pollution should exceed estimates of
annual willingness to pay.
Willingness to pay estimated from the contingent valuation sur-
vey bids are different from zero.
The design for the test used a hedonic property model that was estimated
with sales of single-family houses in these areas and the contingent valuation
8-11
-------
experiment conducted with households selected from the same areas. Overall,
the Brookshire et al. [1982] findings supported the presence of positive bids
for air pollution reductions in all areas, as well as the ranking of rent differ-
entials over bids in 10 of the 11 communities. Thus, the Brookshire et al.
[1982] analysis provides the first evidence that benefit estimates derived from
survey procedures fall within the theoretical bounds for willingness to pay.
Nonetheless, the comparison is based on average responses within the selected
communities and not estimates at an individual level.
In summary, past efforts (especially those of Bishop and Heberlein [1979]
and Brookshire et al. [1982]) directed toward comparative evaluations of bene-
fit methodologies are complementary to those available from the comparative
analysis of this study. The comparison of the travel cost and contingent
valuation is especially important because of the ability to compare benefits esti-
mated for the same users.
8.3 A COMPARATIVE EVALUATION OF THE CONTINGENT VALUATION,
TRAVEL COST, AND CONTINGENT RANKING BENEFIT ESTIMATION
METHODS
Mean estimates are provided in Table 8-3 for each component of the
benefits associated with three water quality changes:
Deterioration in water quality leading to the loss of the recrea-
tional use of the area for water-based activities
Improvement in water quality from its present state (boatable
conditions) to fishable conditions
Improvement from boatable to swimmable conditions.
*
The estimates include the option price and its components—user value and
option value. These results are based on different subsets of the Mononga-
hela survey respondents and are measured in 1981 dollars. The contingent
valuation estimates are based on the full sample, excluding protest bids and
those respondents identified as outliers in the survey (i.e., using the Bels-
ley, Kuh, and Welsch [1980] regression diagnostics, as detailed in Chapter 4).
The travel cost estimates were derived for the survey respondents who were
users of sites along the Monongahela River.* Finally, the contingent ranking
estimates relate to those survey respondents who reported complete ranking
information and income. Thus, this group includes some individuals who were
judged outliers in the contingent valuation survey.
*The travel cost results include all survey respondents who were users
of sites along the Monongahela River, whether or not they were identified as
protest bids or Belsley, Kuh, and Welsch [1980] outliers. Table C-18 in
Appendix C provides the regression comparisons of contingent valuation and
travel cost estimates with these individuals deleted from the sample. The
deletion of these respondents does change any of our conclusions.
8-12
-------
00
I
w
Table 8-3. A Comparison of Benefit Estimates for Water Quality Improvements
(1981 Dollars)
Methodology
t. Contingent valuation
Direct question
Payment card
Iterative bidding ($25)
Iterative bidding ($125)
II. Contingent ranking0
Ordered loglt
Ordered normal
III. Generalized travel cost
AWQ = Loss of use
Option User Option
price value value
24.55 6.57 17.98
(19.71)
51.00 6.20 44.82
(19.71)
28.97 2.16 26.81
(6.58)
57.40 12.08 45.31
(36.25)
-
-
82.65
AWQ = Boatable to fishable
Option
price
17.65
29.26
15.95
36.88
60.03
62.12
-
User Option
value value
7.06 10.59
(21.18)
9.72 19.54
(30.88)
1.38 14.57
(4.21)
6.77 30.10
(20.31)
-
-
7.01
AWQ = Boatable to swimmable
Option
price
31.20
42.87
25.09
60.20
108.06
111.81
User
value
13.61
(31.18)
15.92
(51.18)
3.12
(10.53)
13.43
(48.75)
-
-
14.71
Option
value
20.80
26.76
21.64
43.96
-
-
-
aThe numbers in parentheses below the estimated user values report average user values for users only. Since nonusers have a
zero user value, the combined mean understates user values.
bThese estimates are for the combined sample including users and nonusers. It excludes protest bids and outliers detected using
the Belsley, Kuh, and Welsch regression diagnostics.
°These estimates are for the sample of respondents with usable ranks and reported family income. Estimates evaluated at the
intermediate payment level.
dThese estimates are for survey respondents using Monongahela sites and have been converted to 1981 dollars using the consumer
price index.
-------
Table 8-3 clearly illustrates the pairwise comparisons possible with these
three methods. Because contingent valuation provides the most complete set
of estimates, it can be compared to both of the other methods for several com-
ponents of the benefits from a water quality change.
Simple comparisons of the means in Table 8-3 indicate that the relation-
ship between the methods depends on the type of change in water quality
being considered. For example, in the case of user values, contingent valua-
tion estimates would be expected to be less than the travel cost estimates of
ordinary consumer surplus for improvements in water quality. However,
based on the arguments developed in the previous section of this chapter,
these differences would likely be slight. This relationship does not seem to
have been upheld for improvements in water quality when the mean willing-
ness to pay for users (reported in parentheses in Table 8-3) is compared
with the ordinary consumer surplus increments. Three of the four contingent
valuation approaches contrast with this expectation for both of the water
changes. Only the mean for the iterative bidding format with the $25 start-
ing point is less than the ordinary consumer surplus estimate. Moreover, the
differences in some cases are greater than the theoretical arguments would
have implied. Because the largest of these estimates is not associated with the
iterative bidding framework with a $125 starting point, the discrepancy cannot
be attributed to starting point bias. These comparisons are not statistical
tests, and the contingent valuation estimates exhibit considerable variability.
Indeed, the travel cost estimates do fall, for both levels of improvement in
water quality, in the range of estimates provided by the various approaches to
contingent valuation.
The comparison between the means for the contingent valuation and travel
cost estimates is consistent with theoretical expectations for a reduction in
water quality that leads to the loss of the area. In this case, the ordinary
consumer surplus is more than twice the size of the largest of the contingent
valuation estimates. The size of this difference was somewhat unexpected
based on the simple theoretical arguments discussed earlier. Accordingly, it
serves to highlight the potential importance of each methodology's assumptions
in comparing their respective estimates. One explanation of this large differ-
ence arises from an assumption implicit in the travel cost model. The data
required that the travel cost demand model ignore the effects of substitute
sites as determinants of the demand for any one site's services. However,
judging the potential effects of this limitation on the estimates from the gen-
eralized travel cost model are difficult. The model developed in Chapter 7
assumes that each individual considered only site attributes in judging the
degree of substitutability between sites. Indeed, it was based on the assump-
tion that all sites' services could be measured on a common scale reflecting
these attributes. To the extent this assumption is either inappropriate or a
relatively weak approximation of each individual's perceptions of the relation-
ship between sites, there will be two types of effects on the demand model.
First, the omission of variables reflecting the prospective role of these sub-
stitution effects in any site's demand function is a specification error that may
bias estimates of the other variables' effects on demand. Equally important,
the differential accessibility of substitute sites of comparable or higher quality
will tend to mitigate the impact of any deterioration in water quality at a given
8-14
-------
-•uce the incremental benefits from improvements. Thus, it IS diffi-
cult to predict with certainty the impacts of the treatment of the role of sub-
stitutes for benefit estimates derived from the generalized travel cost model.
Nontheless, it does seem reasonable to expect that the use of a model
that ignores the role of substitutes may not seriously affect the benefit esti-
mates associated with the increments to water quality that serve to enhance
the activities supported by a recreation site. By contrast, this judgment is
not as readily accepted for the loss of a site. In this case, the presence of
substitute facilities can be expected to mitigate the loss. Thus, the general-
ized travel cost model (which ignores the role of substitute sites) may over-
estimate the consumer surplus associated with the loss of the use of the
Monongahela River for boating recreation.
The second comparison that can be made is between the contingent valua-
tion and contingent ranking estimates of the option price. Regardless of the
technique used to estimate the random utility function, the contingent ranking
approximation of option price consistently exceeds the contingent valuation
estimates. Because both methods focus on the same benefit concept, the
explanations for it must arise from the assumptions of each approach. The
approximations used to derive the contingent ranking benefit estimates may be
especially important to such an explanation.* However, in the final analysis,
there is little additional information that can be gleaned from a comparison of
means.
The most interesting comparisons of contingent valuation and travel cost
estimates are based on the subsample of users; the most interesting compar-
isons of contingent valuation and contingent ranking are based on the sub-
sample of respondents with complete information on the ranking of water quality
and payment alternatives. Both sets of comparisons use individual benefit
estimates.
The comparison of contingent valuation and travel cost estimates of user
values is presented in Table 8-4. The objective of this comparison is to judge
how the benefit measures derived using the two approaches compared across
individuals. Accordingly, a common set of procedures was used to evaluate
the accuracy of a set of forecasts (see Theil [1961], pp. 31-33, for discus-
sion of this type of application). In this comparison, the contingent valua-
tion measure of user value was regressed on the travel cost estimate. Because
this comparison may be affected by the question format used with the contin-
gent valuation approach, qualitative variables for three of the four modes
were also included as determinants of the level of the contingent valuation
estimates.
*This benefit measure is described as approximate because of its defini-
tion as an increment to the payment required to hold an individual's utility
constant in the presence of a water quality improvement and because of the
theoretical inconsistency in the functional form used for the indirect utility
function (see Chapter 6 for details).
8-15
-------
Table 8-4. A Comparison of Contingent Valuation and
Generalized Travel Cost Benefit Estimates3
flWQ = Loss of area AWQ = Boatable to fishable AWQ = Boatable to swimmable
Model
Test--
Model
TestL
Model
Test1-
Independent variable
Intercept
Travel cost benefit
estimate
Qualitative variables
Payment card
Direct question
Iterative bid ($25)
21.862
(1.371)
.328
(1.169)
-32.640
(-2.551)
-14.602
(-1.270)
-31.817
(-2.549)
-4.357
33.985
(1.900)
-3.670
(-1.204)
51.757
(2.639)
12.957
(0.748)
-11.244
(-0.595)
-1.712
59.574
(2.017)
-2.713
(-1.141)
77.010
(2.359)
21.001
(0.729)
-21.819
(-0.693)
-1.793
R*
n
.099
93
2.42
(0.05)c
.120
93
3.00
(0.02)c
.107
93
2.62
(0.04)c
The numbers in parentheses below the estimated coefficients are t-ratios for the null hypothesis of no
association.
This column reports the t-ratio for the hypothesis that the coefficient for the travel cost variable was 1.55.
The travel cost model measures consumer surplus in 1977 dollars. The contingent valuation experiments
were conducted in 1981. Using the consumer price index to adjust the travel cost benefit estimates to 1981
dollars would require multiplying each estimate by 1.55. Since the estimated regression coefficients (and
standard errors) will correspondingly adjust to reflect this scale change, a test of the null hypothesis that
the coefficient of travel cost was equal to unity is equivalent to a test that is equal to 1.55 when the
travel cost benefit estimates are measured in 1977 dollars and user values estimates (the dependent vari-
able) are in 1981 dollars.
cThis number in parentheses below the reported F-statistic is the level of significance for rejection of the
null hypothesis of no association between the dependent and independent variables.
The analysis was considered for each of three water quality changes:
Deterioration in water quality leading to the loss of the areas
Improvement in water quality from its present state (boatable
conditions) to fishable conditions
Improvement from boatable to swimmable conditions.
The results generally reinforce the earlier judgments from comparing the esti-
mated mean user values from each method. Theory suggests contingent valua-
tion estimates would be less than the ordinary consumer surplus estimates from
the travel cost model for water quality improvements, but these differences
should be rather small. This a priori expectation can be evaluated by testing
the null hypothesis that the intercept for the model is zero. Equally important,
if the two methods provide comparable estimates of user values that closely
8-16
-------
approximate each individual's willingness to pay, the slope parameter for the
travel cost consumer surplus would be expected to be insignificantly different
from unity. Finally, if the question mode does not influence the responses
derived with contingent valuation surveys, the dummy variables for question
mode would likely not be significantly different from zero.
More formally, it has been maintained that the contingent valuation esti-
mates of an individual's willingness to pay for water quality changes "a", CVa/
will be approximately a homogeneous function of the conditional expectation
for the Marshallian consumer surplus, MS (i.e., the predicted consumer sur-
plus from the generalized travel cost model for water quality change "a").
This function will exhibit a slope of unity. This model is to be distinguished
from an errors-in-variables framework in which it would be maintained that
neither benefit measure describes what it is purported to measure. Under
this study's interpretation, the travel cost estimates of consumer surplus play
the same role as the estimates of the conditional expectation of endogenous
variables in a deterministic simulation of an econometric model (see Howrey
and Kelejian [1969] and Aigner [1972]). Hence, large sample evaluations of
the parameters in the model--testing the hypotheses of zero intercept and uni-
tary slope—do provide some guidance as to the relationship between methods.
The results provide some interesting insights for each of these issues.
Considering the relationship between the level of the contingent valuation
estimates and those of the travel cost model, there is some evidence for -a dif-
ference between the levels of the two approaches for improvements in water
quality that contradicts a priori expectations. The intercepts for the equa-
tions associated with both levels of water quality increments (i.e., from boat-
able to fishable and from boatable to swimmable) are positive and statistically
significant at the 90-percent significance level. However, there are at least
two reasons for interpreting these results cautiously. The generalized travel
cost model does not permit the effect of the intercept to be distinguished from
at least one of the questioning formats. In the models reported in Table 8-4,
the intercept reflects the effects of the iterative bidding format with a $125
starting point. Testing whether the sum of the intercept and any one of the
coefficients for other models was nonzero would simply change the format in-
cluded. Ignoring the effects of question format by eliminating these variables
from the models simply reinforces the conclusion that the intercept for these
cases is positive and significantly different from zero.
Thus, there is some evidence to support the conclusion that contingent
valuation methods may overstate willingness to pay for water quality improve-
ments. It is not unambiguous evidence, because the tests are based on large
sample behavior and have been applied using the conventional t-distributions.
These findings are not necessarily at variance with the Brookshire et al.
[1982] conclusions. Their evaluation concluded that contingent valuation esti-
mates fall within the bounds which can be established by theory. It does not
indicate how close the estimates fall to the "true" value of individual willing-
ness to pay. An appraisal suggests that, for increments (improvements) to
water quality, contingent valuation estimates may well overstate the user
benefits.
8-17
-------
The conclusion for reductions in water quality that would be associated
with the loss of the area is less clearcut. In this case, the contingent valua-
tion estimates are less than ordinary consumer surplus, as theory would imply.
However, they are substantially less, and the reasons may be associated with
the travel cost model and not the survey approach to benefit estimation. Based
on the association between estimates across individuals, there is support for
the conclusion that the travel cost model overstates the benefits associated with
avoiding the loss of the area. The slope coefficient is significantly different
from theoretical expectations. Since the travel cost benefits are measured in
1977 dollars, the correct null hypothesis for the slope coefficient when 1977
dollars are not converted to 1981 is that the coefficient equals the adjustment
factor (in this case, 1.55).* For improvements in water quality, the coeffi-
cients are numerically large and have an incorrect sign, but they are not sig-
nificantly different from 1.55.
Thus, for changes in water quality, the models do seem to move together
(with the contingent valuation potentially exhibiting a positive bias in estimat-
ing willingness to pay). The performance of the contingent valuation method
does appear to depend on the mode of questioning used—with the clearest
distinctions found between the payment card and iterative bid with a $125
starting point. While the explanatory power of the model is not high, reflect-
ing the variability in the contingent valuation responses for user values, the
null hypothesis of no association between these measures of user values (along
with the qualitative variables) is clearly rejected at high levels of significance
based on the F-statistics, reported at the bottom of the table.
The second individual level comparison involves estimates of the option
price using contingent valuation and contingent ranking methods. Table 8-5
reports a comparable set of regression models comparing these estimates. How-
ever, two further distinctions are possible in this comparison. Given the
functional form specified for the indirect utility function, the contingent rank-
ing estimate of option price will depend on the level of the payment suggested
to the individual. Consequently, the benefits were calculated at all three
levels and the regressions were replicated for each of them. In addition, two
econometric estimators were used with the contingent ranking models so that
each was also considered. Table 8-5 reports all of the comparisons for two
increments in water quality—improvements from beatable to fishable and from
boatable to swimmable.
^Scaling all the values of an independent variable by k will scale the
ordinary least-squares estimate of the parameter for this variable (in a linear
model) and its estimated standard error by •£= Thus, to test the null hypoth-
esis of unity for such a parameter would imply using
A
b
8-18
-------
Table 8-5. A Comparison of Contingent Valuation and Contingent Ranking Benefit Estimates
oo
_i
ID
Independent variable
ORDERED LOGIT
Intercept
A Payment
Qualitative variables
Payment card
Direct question
Iterative bidding ($25)
R*
n
F
ORDERED NORMAL
Intercept
A Payment
Qualitative variables
Payment card
Direct question
Iterative bidding ($25)
R*
n
F
AWQ
Payment = $50
Model Test
-20.141
(-1.095)
1.209 0.741
(4.279)
-22.486
(-2.424)
-35.267
(-3.751)
-38.045
(-4.067)
.165
184
8.87
(0.0001)
-13.467
(-0.839)
1.073 0.309
(4.554)
-22.642
(-2.457)
-34.934
(-3.745)
-37.541
(-4.014)
.176
184b
9.53 .
(0.0001)°
= Beatable to flshable
Payment = $100
Model Test
-23.647
(-1.223)
1.315 1.016
(4.237)
-22.070
(-2.380)
-34.595
(-3.683)
-37.562
(-4.015)
.164
184
8.77
X0.0001)
-15.565
(-0.940)
1.140 0.554
(4.528)
-22.357
(-2.426)
-34.458
(-3.696)
-37.196
(-4.004)
.175
184b
9.47
(0.0001)
Payment = $175
Model Test
-23.927
(-1.227)
1.330 1.048
(4.214)
-21.960
(-2.367)
-34.425
(-3.665)
-37.446
(-4.001)
.163
184
8.72
(0.0001)
-15.832
(-0.951)
1.151 0.592
(4.516)
-22.286
(-2.418)
-34.344
(-3.683)
-37.116
(-3.994)
.174
184b
9.43
(0.0001)
AWQ =
Payment = $50
Model Test
-25.661
(-0.795)
1.081 0.293
(3.925)
-46.842
(-2.877)
-55.327
(-3.353)
-68.611
(-4.178)
.153
184
8.06
(0.0001)
-15.153
(-0.537)
.962 -0.165
(4.182)
-47.108
(-2.910)
-54.808
(-3.345)
-67.808
(-4.156)
.162
184b
8.63
(0.0001)
Beatable to swlmmable
Payment =
Model
-30.734
(-0.905)
1.170
(3.867)
-46.145
(-2.834)
-54.215
(-3.288)
-67.817
(-4.128)
.151
184
7.94
(0.0001)
-18.212
(-0.626)
1.018
(4.146)
-46.630
(-2.880)
-54.020
(-3.298)
-67.242
(-4.120)
.160
184b
8.54
(0.0001)
$100 Payment
Test Model
-31.032
(-0.906)
0.561 1.183
(3.841)
-45.961
(-2.822)
-53.935
(-3.270)
-67.626
(-4.115)
.150
184
7.88
(0.0001)
-18.559
(-0.634)
0.073 1.028
(4.131)
-46.510
(-2.872)
-53.832
(-3.286)
-67.112
(-4.111)
.160
184b
8.51
(0.0001)
= $175
Test
-
0.594
-
-
-
-
0.113
-
-
-
"These estimates are for the
regression diagnostics.
bThe*« estimates are for the
combined sample Includli
sample of respondents %
ig users and nonusers. It excludes protest bids and outliers detected using the Kuh-Welsch
rlth usable ranks and reported family income.
-------
The interpretation of these results is somewhat different from the earlier
comparison with travel cost estimates. In this case, both methods seek to
estimate the same benefit concept. However, they are not independent. Each
survey respondent was asked to engage in both activities—one of four types
of contingent valuation experiment and a contingent ranking. Thus, these re-
sults reflect the consistency in individuals' responses and the potential effects
of how the valuation exercise is undertaken (i.e., requests for bids or ranks).
Despite the fairly substantial differences in the means for the two approaches
as reported in Table 8-3, these results exhibit remarkable consistency. Once
again, the relevant hypotheses are for zero intercept and unitary slope coef-
ficients. Both hypotheses cannot be rejected across all possible variants of
the contingent ranking and changes in water quality. Indeed, the numerical
estimates of the slope coefficient exhibit rather considerable agreement between
the direction of the movements in the two estimates of option price. The esti-
mated coefficients for the question format used are especially interesting.
They indicate that the association between the two approaches depends quite
importantly on the question format, with the iterative bidding format with a
$125 starting, point providing larger estimates than any of the other three
formats.
Overall, these findings suggest that even though the models used to
derive benefit estimates from the contingent ranking models were somewhat
arbitrary (and in some cases inconsistent with a strict interpretation of the
relevant theory), the results move closely with the contingent valuation esti-
mates. Indeed, one of the primary sources of divergence between the two
arises in the format used with the contingent valuation questions.
8.4 IMPLICATIONS
This chapter has developed comparisons of three methods for estimating
the benefits from water quality improvements. Each method has involved a
fairly detailed set of assumptions and, in some cases, a complex model. Over-
all, the results are remarkably consistent across methods for comparable
changes in water quality. While this discussion has been devoted to the types
of discrepancies between each method's estimates, the consistency in these
estimates should be interpreted as offering strong support for the feasibility
of performing benefit analyses for water quality changes. The range of varia-
tion in estimates across methods is generally less than the variation expected
in models seeking to translate the effects of effluent in a water body into the
corresponding measures of water quality parameters.
Nonetheless, this conclusion does not imply that there is not room for
improvement in benefit estimation methods. In most cases, the indirect meth-
ods for benefit measurement, such as the travel cost framework, have been
limited by the data availability. While this study's analysis was greatly en-
hanced by the existence of the Federal Estate Survey, the form of the data
nonetheless imposed limitations on the character of the travel cost demand
models that could be formulated. Survey approaches do not face the same
types of limitations. However, this study's findings do suggest that the ques-
tion format used is an important factor in the benefit estimates derived from
the survey. They also suggest that greater attention to the nature and form
8-20
-------
of the information provided to survey respondents will be needed if this ap-
proach is to seek to develop detailed measures of the components of benefits.
The analysis performed for this study had the advantage of a well-defined
valuation problem that was easily explained and, according to interviewer feed-
back after the survey, readily understood by the survey respondents. Many
of the most complex environmental valuation problems do not share this char-
acteristic and therefore may not have the same successes reported here.
The specific findings of the comparison indicated that contingent valua-
tion methods may overstate the willingness to pay for water quality improve-
ments. Theory would suggest that ordinary consumer surplus should provide
an upper bound for these estimates and this study's findings indicate it does
not. Nonethless, these differences are not substantial and fall within the
range of variation of the contingent valuation estimates across the question
formats. For the case of the loss of the use of the area, the association
adheres to theoretical anticipations. Indeed, there are reasons to believe that
the travel cost estimates overstate the benefits provided by the area.
Comparison between the contingent ranking and contingent valuation esti-
mates indicate a remarkable degree of consistency. While the mean benefit
estimates derived from the contingent ranking framework appear larger than
the contingent valuation estimates, there is not a statistically significant dis-
placement between the two. Moreover, the benefit estimates move in close
agreement across individuals.
8-21
-------
CHAPTER 9
REFERENCES
Aigner, Dennis J., 1972, "A Note on Verification of Computer Simulation
Models," Management Science, Vol. 18, July 1972, pp. 615-19.
Aizen, I., and M. Fishbien, 1977, "Attitude-Behavior Relations: A Theoret-
ical Analysis and Review of Empirical Research," Psychological Bulletin,
Vol. 84, 1977, pp. 888-918.
Allen, P. Geoffrey, Thomas H. Stevens, and Scott A. Barrett, 1981, "The
Effects of Variable Omission in the Travel Cost Technique," Land
Economics, Vol. 57, No. 2, May 1981, pp. 173-80.
Amemiya, Takeshi, 1981, "Qualitative Response Models: A Survey," Journal
of Economic Literature, Vol. 19, No. 4., December 1981, pp. 1483-536.
Anderson, R. J., 1981, "A Note on Option Value and the Expected Value of
Consumer's Surplus," Journal of Environmental Economics and Manage-
ment, Vol. 8, June 1981, pp. 187-91.
Arrow, K. J., and A. C. Fisher, 1974, "Environmental Preservation, Uncer-
tainty and Irreversibility," Quarterly Journal of Economics, Vol. 88, May
1974, pp. 313-19.
Barker, Mary L., 1971, "Beach Pollution in the Toronto Region," in W. R.
Sewell and Ian Burton, eds., Perceptions and Attitudes iin Resource
Management, Ottawa, Canada: Department of Energy, Mines, and Re-
sources, 1971, pp. 37-48.
Becker, Gary. S., 1965, "A Theory of the Allocation of Time," Economic
Journal, Vol. 75, September 1965, pp. 493-517.
Becker, Gary S., 1974, "A Theory of Social Interactions," Journal of Political
Economy, Vol. 82, 1974, pp. 1063-93.
Becker, Gary S., 1981, "Altruism in the Family and Selfishness in the Market
Place," Economica, Vol. 48, February 1981, pp. 1-16.
Beggs, S., S. Cardell, and J. Hausman, 1981, "Assessing the Potential
Demand for Electric Cars," Journal of Econometrics, Vol. 16, September
1981, pp. 1-19.
9-1
-------
Belsley, David A., Edwin Kuh, and Roy E. Welsch, 1980, Regression Diag-
nostics, New York: John Wiley and Sons, 1980.
Berndt, Ernst, R., 1983, "Quality Adjustment in Empirical Demand Analysis,"
Working Paper 1397-83, Sloan Schools, Massachusetts Institute of Tech-
nology, Cambridge, Massachusetts: January 1983.
Binkley, Clark S., and W. Michael Hanemann, 1978, The Recreation Benefits
of Water Quality Improvement: Analysis of Day Trips in an Urban
Setting, Washington, D.C.: U.S. Environmental Protection Agency,
1978.
Bishop, R. C., and T. A. Heberlein, 1979, "Measuring Values of Extra-Market
Goods: Are Indirect Measures Biased?" American Journal of Agricul-
tural Economics, Vol. 6, December 1979, pp. 926-30.
Bishop, Richard C., 1982, "Option Value: An Exposition and Extension,"
Land Economics, Vol. 58, February 1982, pp. 1-15.
Blackorby, Charles, Daniel Primont, and R. Robert Russell, 1978, Duality,
Separability, and Functional Structure: Theory and Economic Applica-
tions, New York: North Holland, 1978.
Bockstael, Nancy, 1982, Discussion comments at the Association of Environmen-
tal and Resource Economists Session, American Economic Association
Annual Meeting, 1982.
Bockstael, Nancy E., and Kenneth E. McConnell, 1980, "Calculating Equivalent
and Compensating Variation for Natural Resource Environments," Land
Economics, Vol. 56, No. 1, February 1980, pp. 56-63.
Bockstael, Nancy E., and Kenneth E. McConnell, 1981, "Theory and Estima-
tion of the Household Production Function for Wildlife Recreation," Jour-
nal of Environmental Economics and Management, Vol. 8, September 1981,
pp. 199-214.
Bockstael, Nancy E., and Kenneth E. McConnell, 1982, "Welfare Measurement
in the Household Production Framework," unpublished paper, Department
of Agricultural and Resource Economics, University of Maryland, College
Park, Maryland, February 1982.
Bohm, Peter, 1971, "An Approach to the Problem of Estimating Demand for
Public Goods," Swedish Journal of Economics, Vol. 73, March 1971,
pp. 55-66.
Bohm, Peter, 1975, "Option Demand and Consumer Surplus: Comment," Amer-
ican Economic Review, Vol. 65, September 1975, pp. 733-36.
Bouwes, Nicolaas W., Sr., and Robert Schneider, 1979, "Procedures in Esti-
mating Benefits of Water Quality Change," American Journal of Agricul-
tural Economics, August 1979, pp. 535-39.
9-2
-------
Brookshire, David S., Ronald G. Cummings, Morteza Rahmatian, William D.
Schulze, and Mark A. Thayer, 1982, Experimental Approaches for Valu-
ing Environmental Commodities, draft report prepared for U.S. Environ-
mental Protection Agency, University of Wyoming, Laramie, Wyoming,
April 1982.
Brookshire, David S., Ralph C. d'Arge, William D. Schulze, and Mark A.
Thayer, 1979, Methods Development for Assessing Air Pollution Control
Benefits, Volume II, Experiments \n Valuing Non-Market Goods: A Case
Study of Alternative Benefit Measures of Air Pollution Control [n the
South Coast Air Basin of Southern California, EPA-600/5-79-001D, U.S.
Environmental Protection Agency, Washington, D.C., 1979.
Brookshire, David S., B. Ives, and William D. Schulze, 1976, "The Valuation
of Aesthetic Preferences," Journal of Environmental Economics and Man-
agement, Vol. 3, December 1976, pp. 325-46.
Brookshire, David S., and Alan Randall, 1978, "Public Policy Alternatives,
Public Goods, and Contingent Valuation Mechanisms," paper presented at
the Western Economic Association Meeting, Honolulu, Hawaii, June 1978,
pp. 20-26.
Brookshire, David S., Mark A. Thayer, William D. Schulze, and Ralph C.
d'Arge, 1982, "Valuing Public Goods: A Comparison of Survey and
Hedonic Approaches," American Economic Review, Vol. 72, March 1982,
pp. 165-77.
Brown, Gardner, Jr., and Robert Mendelsohn, 1980, "The Hedonic Travel
Cost Method," unpublished paper, Department of Economics, University
of Washington, Seattle, Washington, December 1980.
Brown/ William G., and Farid Nawas, 1973, "Impact of Aggregation on the
Estimation of Outdoor Recreation Demand Functions," American Journal of
Agricultural Economics, Vol. 55, May 1973, pp. 246-49.
Burt, O. R., and D. Brewer, 1971, "Estimation of Net Social Benefits from
Outdoor Recreation," Econometrica, Vol. 39, October 1971, pp. 813-27.
Cesario, Frank J., 1976, "Value of Time in Recreation Benefit Studies," Land
Economics, Vol. 52, February 1976, pp. 32-41.
Cesario, Frank J., and Jack L. Knetsch, 1970, "Time Bias in Recreation
Benefit Estimates," Water Resources Research, Vol. 6, June 1970,
pp. 700-04.
Cesario, Frank J., and Jack L. Knetsch, 1976, "A Recreation Site Demand and
Benefit Estimation Model," Regional Studies, Vol. 10, 1976, pp. 97-104.
Cicchetti, Charles J., Anthony C. Fisher, and V. Kerry Smith, 1976, "An
Economic Evaluation of a Generalized Consumer Surplus Measure: The
Mineral King Controversy," Econometrica, Vol. 44, November 1976,
pp. 1259-76.
9-3
-------
Cicchetti, Charles J., and A. Myrick Freeman, III, 1971, "Option Demand and
the Consumer Surplus: Further Comment," Quarterly Journal of Econom-
ics, Vol. 85, August 1971, pp. 528-39.
Cicchetti, Charles J., Joseph J. Seneca, and Paul Davidson, 1969, The De-
mand and Supply of Outdoor Recreation, New Brunswick, New Jersey:
Bureau of Economic Research, Rutgers University, 1969.
Cicchetti, Charles J., and V. Kerry Smith, 1976, The Costs of Congestion,
Cambridge, Massachusetts: Ballinger Publishing Co., 1976.
Clawson, M., 1959, "Methods of Measuring the Demand for and Value of
Outdoor Recreation," Reprint No. 10, Resources for the Future, Inc.,
Washington, D.C., 1959.
Clawson, M., and J. L. Knetsch, 1966, Economics of Outdoor Recreation,
Washington, D.C.: Resources for the Future, Inc., 1966.
Conrad, J. M., 1980, "Quasi Option Value and the Expected Value of Informa-
tion," Quarterly Journal of Economics, Vol. 94, June 1980, pp. 813-20.
Cook, Philip J., and Daniel A. Graham, 1977, "The Demand for Insurance and
Protection: The Case of Irreplaceable Commodities," Quarterly Journal
of Economics, Vol. 91, February 1977, pp. 143-56.
Council of Economic Advisors, 1982, "Economic Report of the President,"
Washington, D.C.: U.S. Government Printing Office, 1982.
Cox, D. R., 1972, "Regression Models and Life-Tables," Journal of the Royal
Statistical Society, Series B, Vol. 34, 1972, pp. 187-202.
Cronin, Francis J., 1982, Valuing Nonmarket Goods Through Contingent
Markets, prepared for U.S. Environmental Protection Agency, Pacific
Northwest Laboratory, Richland, Washington, September 1982.
Davidon, Fletcher Ri, and M. Powell, 1963, "A Rapidly Convergent Descent
Method for Minimization," The Computer Journal, Vol. 6, 1963, pp. 163-
68.
Davis, Robert K., 1963, "Recreation Planning as an Economic Problem,"
Natural Resources Journal, Vol. 3., No. 2, October 1963, pp. 239-49.
Deaton, Angus, and John Muellbauer, 1980, Economics and Consumer Behavior,
Cambridge: Cambridge University Press, 1980.
Deyak, Timothy A., and V. Kerry Smith, 1978, "Congestion and Participation
in Outdoor Recreation: A Household Production Approach," Journal of
Environmental Economics and Management, Vol. 5, March 1978, pp. 63-807"
Ditton, B., and T. L. Goodale, 1973, "Water Quality Perceptions and the
Recreational Users of Green Bay," Water Resources Research, Vol 9,
No. 3, 1973, pp. 569-79.
9-4
-------
Dwyer, J. F., J. R. Kelly, and M. D. Bowes, 1977, Improved Procedures for
Valuation of the Contribution of Recreation to National Economic Develop-
ment, Urbana-Champaign: University of Illinois, 1977.
Feenberg, Daniel, and Edwin S. Mills, 1980, Measuring the Benefits of Water
Pollution Abatement, New York: Academic Press, 1980.
Fisher, A. C., and V. Kerry Smith, 1982, "Economic Evaluation of Energy's
Environmental Costs, with Special References to Air Pollution," Annual
Review of Energy, Vol. 7, 1982, pp. 1-35.
Fisher, Ann, and Robert Raucher, 1982, "Comparison of Alternative Methods
of Evaluating the Intrinsic Benefits of Improved Water Quality," paper
presented at the American Economics Association Annual Meeting, New
York, New York, December 1982.
Fisher, Franklin M., and Karl Shell, 1968, "Taste and Quality Changes in the
Pure Theory of the True Cost-of-Living Index," in J. N. Wolfe, ed.,
Value, Capital, and Growth: Papers in Honour of Sir John Hicks,
Chicago: Aldine Publishing Co., 1968.
Freeman, A. Myrick, III, I979a, The Benefits of Environmental Improvement:
Theory and Practice, Baltimore: Johns Hopkins Press for Resources for
the Future, Inc., 1979.
Freeman, A. Myrick, III, 1979b, The Benefits of Air and Water Pollution
Control: A Review and Synthesis of Recent Estimates, prepared for
Council on Environmental Quality, Washington, D.C., December 1979.
Freeman, A. Myrick, III, 1979c, "Hedonic Prices, Property Values and Meas-
uring Environmental Benefits: A Survey of the Issues," Scandinavian
Journal of Economics, Vol. 81, No. 2, 1979, pp. 154-73.
Freeman, A. Myrick, III, 1981, "Notes on Defining and Measuring Existence
Values," unpublished manuscript, Department of Economics, Bowdoin
College, Brunswick, Maine, June 1981.
Freeman, A. Myrick, III, 1982, "The Size and Sign of Option Value," unpub-
lished paper, Bowdoin College, Brunswick, Maine, 1982.
Giles, D. E. A., 1982, "The Interpretation of Dummy Variables in Semilogar-
ithmic Equations: Unbiased Estimation," Economic Letters, Vol. 10,
1982, pp. 77-79.
Goldberger, A. S., 1968, "The Interpretation and Estimation of Cobb-Douglas
Functions," Econometrica, Vol. 36, July to October 1968, pp. 464-72.
Graham, D. A., 1981, "Cost-Benefit Analysis Under Uncertainty," American
Economic Review, Vol. 71, September 1981, pp. 715-25.
9-5
-------
Greene, William H., 1981, "On the Asymptotic Bias of the Ordinary Least
Squares Estimates of the Tobit Model," Econometrica, Vol. 49, March 1981,
pp. 505-14.
Greene, William H., 1983, "Estimation of Limited Dependent Variable Models by
Ordinary Least Squares and the Method of Moments," Journal of Econo-
metrics, Vol. 21, February 1983, pp. 195-212.
Greenley, D. A., Richard G. Walsh, and Robert A. Young, 1981, "Option
Value: Empirical Evidence from a Case Study of Recreation and Water
Quality," Quarterly Journal of Economics, Vol. 96, November 1981,
pp. 657-74.
Greenley, D. A., Richard G. Walsh, and Robert A. Young, 1983, "Notes on
Mitchell and Carson's Proposed Comment on 'Option Value: Empirical
Evidence From a Case Study of Recreation and Water Quality," unpub-
lished manuscript, Department of Economics, Colorado State University,
Fort Collins, Colorado, January 1983, pp. 1-12.
Gum, R., and W. E. Martin, 1975, "Problems and Solutions in Estimating the
Demand for the Value of Rural Outdoor Recreation," American Journal of
Agricultural Economics, Vol. 57, November 1975, pp. 558-66.
Haspel, Abraham, E., and F. Reed Johnson, 1982, "Multiple Destination Trip
Bias in Recreation Benefit Estimation," Land Economics, Vol. 58, August
1982, pp. 364-72.
Hause, John C., 1975, "The Theory of Welfare Cost Measurement," Journal of
Political Economy, Vol. 83, December 1975, pp. 1145-82.
Hausman, Jerry A., 1978, "Specification Error Tests in Econometrics," Econo-
metrica, Vol. 46, November 1978, pp. 1251-72.
Hausman, Jerry A., 1981, "Exact Consumer's Surplus and Deadweight Loss,"
American Economic Review, Vol. 71, No. 4, September 1981, pp. 662-76.
Hausman, Jerry A., and David A. Wise, 1978, "A Conditional Profit Model for
Qualitative Choice: Discrete Decisions Recognizing Interdependence and
Heterogeneous Preferences," Econometrica, Vol. 42, March 1978, pp 403-
26.
Henry, C., 1974, "Option Values in the Economics of Irreplaceable Assets,"
Review of Economic Studies, Vol. 64, 1974, pp. 89-104.
Hicks, John R., 1943, "The Four Consumers' Surplus," Review of Economic
Studies, Vol. 11, Winter 1943, pp. 31-41.
Hirshleifer, J., 1970, Investment, Interest, and Capital, Englewood Cliffs,
New Jersey: Prentice Hall, 1970.
9-6
-------
Howrey, E. Phillip, and Harry H. Kelejian, 1969, "Simulation versus Analytical
Solutions," in T. H. Naylor, ed., The Design of Computer Simulation
Experiments, Durham, North Carolina: Duke University Press, 1969.
Johnson, Norman L., and Samuel Kotz, 1970, Continuous Univariate Distribu-
tions-1, New York: Houghton Mifflin, 1970. '
Just, Richard E., Darell L. Hueth, and Andrew Schmitz, 1982, Applied Wel-
fare Economics and Public Policy, Englewood Cliffs, New Jersey: Prentice
Hall, 1982.
Keener, Robert, and Donald M. Waldman, 1981, "Maximum Likelihood Regression
of Rank-Censored Data," unpublished paper, Department of Economics,
University of North Carolina at Chapel Hill, Chapel Hill, North Carolina,
October 1981.
Kelejian, Harry H., 1971, "Two Stage Least Squares and Econometric Systems
Linear in Parameters but Nonlinear in Endogenous Variables," Journal of
the American Statistical Association, Vol. 66, June 1971, pp. 373-74.
Klein, R. W., L. G. Rafsky, D. F. Sibley, and R. D. Willig, 1978, "Decisions
with Estimation Uncertainty," Econometrica, Vol. 46, November 1978,
pp. 1363-88.
Knetsch, Jack L., and Robert K. Davis, 1966, "Comparison of Methods for
Recreation Evaluation," in A. V. Kneese and S. C. Smith, eds., Water
Research, Baltimore: Johns Hopkins, 1966.
Krutilla, J. V., 1967, "Conservation Reconsidered," American Economic Review,
Vol. 57, September 1967, 777-86.
Krutilla, J. V., and A. C. Fisher, 1975, The Economics of Natural Environ-
ments , Baltimore: Johns Hopkins Press for Resources for the Future,
Inc., 1975.
Lau, Lawrence, J., 1982, "The Measurement of Raw Materials Inputs," in
V. Kerry Smith and John V. Krutilla, eds., Explorations m Natural
Resource Economics, Baltimore: Johns Hopkins, 1982, pp. 167-200.
Maler, Karl G., 1974, Environmental Economics: A Theoretical Inquiry,
Baltimore: Johns Hopkins Press for Resources for the Future, Inc.,
1974.
Malinvaud, E., 1972, Lectures 'm Microeconomic Theory, Amsterdam: North
Holland, 1972.
McConnell, Kenneth E., and Ivan Strand, 1981, "Measuring the Cost of Time
in Recreation Demand Analysis: An Application to Sport Fishing,"
American Journal of Agricultural Economics, Vol. 63, February 1981,
pp. 153-56.
9-7
-------
McConnell, Kenneth E., and Jon G. Sutinen, 1983, "A Conceptual Analysis of
Congested Recreation Sites," in V. Kerry Smith, ed., Advances in
Applied Micro-Economics, Greenwich, Connecticut: JAI Press, forth-
coming, 1983.
McFadden, Daniel, 1974, "Conditional Logit Analysis of Qualitative Choice
Behavior," in P. Zarembka, ed., Frontiers ]n_ Econometrics, New York:
Academic Press, 1974.
McFadden, Daniel, 1981, "Econometric Models of Probabilistic Choice," in
Charles F. Manski and Daniel McFadden, eds., Structural Analysis of
Discrete Data with Econometric Applications, Cambridge, Massachusetts:
MIT Press, 1981.
McKenzie, G. W., and I. F. Pearce, 1982, "Welfare Measurement—A Synthe-
sis," American Economic Review, Vol. 72, No. 4, September 1982,
pp. 669-82.
Miller, Jon R., and Frederic C. Menz, 1979, "Some Economic Considerations
for Wildlife Preservation," Southern Economic Journal, Vol. 45, January
1979, pp. 718-29.
Mitchell, Robert Cameron, and Richard T. Carson, 1981, An Experiment jn
Determining Willingness to Pay for National Water Quality Improvements,
draft report prepared for U.S. Environmental Protection Agency, Re-
sources for the Future, Inc., Washington, D.C., June 1981.
Mitchell, Robert Cameron, and Richard T. Carson, 1982, "Comment on Option
Value: Empirical Evidence from a Case Study of Recreation and Water
Quality," unpublished paper, Resources for the Future, Inc., Washington,
D.C., 1982.
Morey, Edward R., 1981, "The Demand for Site-Specific Recreational Activities:
A Characteristics Approach," Journal of Environmental Economics and
Management, Vol. 8, December 1981, pp. 345-71.
Muellbauer, John, 1974, "Household Production Theory, Quality and the
'Hedonic Technique,1" American Economic Review, Vol. 64, December 1974,
pp. 977-94.
Nielsen, Larry, 1980, "Water Quality Criteria and Angler Preference for
Important Recreational Fishes," EPA Benefits Project Recreation Working
Paper 3, unpublished, Washington, D.C., Resources for the Future,
Inc., 1980.
Olsen, Randall J., 1980, "Approximating a Truncated Normal Regression With
the Method of Moments," Econometrica, Vol. 48, July 1980, pp. 1099-106.
9-8
-------
Page, Talbot, Robert Harris, and Judith Bruser, 1981, "Waterborne Carcino-
gens: An Economist's View," in Robert W. Crandall and Lester B. Lave,
eds., The Scientific Basis of Health and Safety Regulation, Washington,
D.C.: The Brookings Institution, 1981, pp. 197-228.
Page, Talbot, Robert Harris, and Judith Bruser, 1982, "An Economist's View
of Waterborne Carcinogens," in R. W. Crandall and L. B. Lave, eds.,
The Scientific Basis of Health and Safety Regulation, Washington, D.C.:
The Brookings Institution, 1982.
Pollak, Robert A., 1969, "Conditional Demand Functions and Consumption
Theory," Quarterly Journal of Economics, Vol. 83, February 1969,
pp. 60-78.
Pollak, Robert A., 1971, "Conditional Demand Functions and the Implications
of Separable Utility," Southern Economic Journal, Vol. 37, April 1971,
pp. 423-33.
Pollak, Robert A., and M. L. Wachter, 1975, "The Relevance of the Household
Production Function and Its Implications for the Allocation of Time,"
Journal of Political Economy, Vol. 83, April 1975, pp. 255-77.
Porter, Richard D., 1973, "On the Use of Survey Sample Weights in the
Linear Model," Annals of Economic and Social Measurement, Vol. 2,
February 1973, pp. 141-58.
Rae, Douglas A., 1981a, Visibility Impairment at Mesa Verde National Park:
An Analysis of Benefits and Costs of Controlling Emissions in the Four
Corners Area, prepared for the Electric Power Research Institute, Charles
River Associates, Boston, Massachusetts, 1981.
Rae, Douglas A., I98lb, Benefits of Improving Visibility at Great Smoky
National Park, draft report prepared for Electric Power Research Insti-
tute, Charles River Associates, Boston, Massachusetts, December 1981.
Rae, Douglas A., 1982, Benefits of Visual Air Quality 'm Cincinnati, prepared
for the Electric Power Research Institute, Charles River Associates,
Boston, Massachusetts, 1982.
Rand McNally and Company, 1978, Standard Highway Mileage Guide, Rand
McNally: Chicago, 1978.
Randall, Alan, Orlen Grunewald, Angelos Pagoulatos, Richard Ausness, and
Sue Johnson, 1978, "Reclaiming Coal Surface Mines in Central Appalachia:
A Case Study of the Benefits and Costs," Land Economics, Vol. 54,
No. 4, November 1978, pp. 472-89.
Randall, Alan, John P. Hoehn, and George S. Tolley, 1981, "The Structure
of Contingent Markets: Some Results of a Recent Experiment," paper
presented at the American Economic Association Annual Meeting, Washing-
ton, D.C., 1981.
9-9
-------
Randall, Alan, Berry Ives, and Clyde Eastman, 1974, "Bidding Games for
Valuation of Aesthetic Environmental Improvements," Journal of Environ-
mental Economics and Management, Vol. 1, 1974, pp. 132-49.
Randall, Alan, and John R. Stoll, 1980, "Consumer's Surplus in Commodity
Space," American Economic Review, Vol. 70, June 1980, pp. 449-55.
Ravenscraft, David J., and John F. Dwyer, 1978, "Reflecting Site Attractive-
ness in Travel Cost-Based Models for Recreation Benefit Estimation,"
Forestry Research Report 78-6, Department of Forestry, University of
Illinois at Urbana-Champaign, Urbana, Illinois, July 1978.
Rosen, Sherwin, 1974, "Hedonic Prices and Implicit Markets: Product Dif-
ferentiation in Perfect Competition," Journal of Political Economy, Vol. 82,
January/February 1974, pp. 34-55.
Rowe, Robert D., and L. G. Chestnut, 1981, Visibility Benefits Assessment
Guidebook, prepared for U.S. Environmental Protection Agency, Abt
West, Denver, Colorado, March 1981.
Rowe, Robert D., Ralph C. d'Arge, and David S. Brookshire, 1980, "An
Experiment on the Economic Value of Visibility," Journal of Environmental
Economics and Management, Vol. 7, March 1980, pp. 1-19.
Samuelson, Paul, 1954, "The Pure Theory of Public Expenditure," Review of
Economics and Statistics, Vol. 36, 1954, pp. 387-89.
Saxonhouse, Gary R., 1977, "Regression from Samples Having Different
Characteristics," Review of Economics and Statistics, Vol. 59, May 1977,
pp. 234-37.
Schmalensee, R., 1972, "Option Demand and Consumer Surplus: Valuing
Price Changes Under Uncertainty," American Economic Review, Vol. 62,
December 1972, pp. 813-24.
Schmalensee, R., 1975, "Option Demand and Consumer Surplus: Reply,"
American Economic Review, Vol. 65, September 1975, pp. 737-39.
Schmidt, P.., 1977, "Estimation of Seemingly Unrelated Regressions With Un-
equal Numbers of Observations," Journal of Econometrics, Vol. 5, May
1977, pp. 365-78.
Schulze, W. D., D. S. Brookshire, E. G. Walter, and K. Kelley, 1981, The
Benefits of Preserving Visibility iri the National Parklands of the South-
west, Volume 8 of Methods Development for Environmental Control Bene-
fits Assessment, prepared for U.S. Environmental Protection Agency, Re-
source and Environmental Economics Laboratory, University of Wyoming,
Laramie, Wyoming, 1981.
Schulze, W. D., R. C. d'Arge, and D. S. Brookshire, 1981, "Valuing Envi-
ronmental Commodities: Some Recent Experiments," Land Economics,
Vol. 57, No. 2, May 1981, pp.'151-73. "
9-10
-------
Shephard, Ronald W., 1953, Cost and Production Functions, Princeton:
Princeton University Press, 1953.
Smith, V. Kerry, 1975a, "The Estimation and Use of Models of the Demand for
Outdoor Recreation," in Assessing the Demand for Outdoor Recreation,
Washington, D.C.: National Academy of Sciences, 1975.
Smith, V. Kerry, 1975b, "Travel Cost Demand Models for Wilderness Recrea-
tion: A Problem of Non-Nested Hypotheses," Land Economics, Vol. 51,
May 1975, pp. 103-11.
Smith, V. Kerry, 1983, "Option Value: A Conceptual Overview," Southern
Economic Journal, Vol. 49, January 1983, pp. 654-68.
Smith, V. Kerry, William H. Desvousges, and Matthew P. McGivney, 1983,
"The Opportunity Cost of Travel Time in Recreation Demand Models,"
Land Economics, forthcoming, August 1983.
Smith, V. Kerry, and Raymond J. Kopp, 1980, "The Spatial Limits of the
Travel Cost Recreation Demand Model," Land Economics, Vol. 56, Febru-
ary 1980, pp. 64-72.
Smith, V. Kerry, and J. V. Krutilla, 1982, "Toward Formulating the Role of
National Resources in Economic Models," in V. K. Smith and J. V.
Krutilla, eds., Explorations iin Natural Resource Economics, Baltimore:
Johns Hopkins, 1982, pp. 1-43.
Smith, V. Kerry, and D. Waldman, 1982, "A Comparison of Ordered Logit and
Ordered Probit: Some Monte Carlo Experiment Results," unpublished
manuscript, Department of Economics, University of North Carolina at
Chapel Hill, Chapel Hill, North Carolina, 1982.
Takayama, Akira, 1982, "On Consumer's Surplus," Economic Letters, Vol. 10,
1982, pp. 35-42.
Talhelm, Daniel R., 1978, "A General Theory of Supply and Demand for
Outdoor Recreation in Recreation Systems," unpublished manuscript,
Department of Agricultural Economics, Michigan State University, East
Lansing, Michigan, July 1, 1978.
Thayer, Mark A., 1981, "Contingent Valuation Techniques for Assessing
Environmental Impacts: Further Evidence," Journal of Environmental
Economics and Management, Vol. 8, 1981, pp. 27-44.
Thayer, Mark A., and W. Schulze, 1977, "Valuing Environmental Quality: A
Contingent Substitution and Expenditure Approach," unpublished paper,
Department of Economics, University of Southern California, Los Angeles,
California, 1977.
Theil, Henri, 1961, Economic Forecasts and Policy, Amsterdam: North Holland,
1961.
9-11
-------
U.S. Bureau of the Census, Department of Commerce, 1970, First Count
Summary Tape, File A, 1970.
U.S. Bureau of the Census, Department of Commerce, 1976, "Statistical
Abstract of the United States: 1976," U.S. Government Printing Office,
Washington, D.C, 1976.
U.S. Bureau of the Census, Department of Commerce, 1982, 1980 Census of
the Population and Housing, preliminary data tape, Washington, D.C.,
1982.
U.S. Environmental Protection Agency, Office of Water Regulations and Stand-
ards, 1982, Handbook, Water Quality Standards, draft, U.S. Environ-
mental Protection Agency, Washington, D.C., October 1982.
Varian, Hal R., 1978, Microeconomic Analysis, New York: W. W. Norton and
Co., 1978.
Vaughan, W. J., and C. S. Russell, 1982, Freshwater Recreational Fishing:
The National Benefits of Water Pollution Control, Washington, D.C.:
Resources for the Future, Inc., November 1982.
von Neumann, J., and O. Morgenstern, 1947, The Theory of Games and
Economic Behavior, 2nd Edition, Princeton: Princeton University Press,
1947.
Walsh, R. G., D. G. Greenley, R. A. Young, J. R. McKean, and A. A.
Prato, 1978, Option Values, Preservation Values and Recreational Bene-
fits of Improved Water Quality: A Case Study of the South Platte River
Basin, Colorado, EPA-600/5-78-001, U.S. Environmental Protection
Agency, Office of Research and Development, January 1978.
Water Resources Council, 1979, "Procedures for Evaluation of National Economic
Development (NED) Benefits and Costs in Water Resources Planning
(Level C), Final Rule," Federal Register, Vol. 44, No. 242, December 14,
1979, pp. 72892-977.
Willig, Robert D., 1976, "Consumer's Surplus Without Apology," American
Economic Review, Vol. 66, September 1976, pp. 587-97.
Wilman, Elizabeth A., 1980, "The Value of Time in Recreation Benefit Studies,"
Journal of Environmental Economics and Management, Vol. 7, September
1980, pp. 272-86.
Zellner, A., 1962, "An Efficient Method of Estimating Seemingly Unrelated
Regressions and Tests for Aggregation Bias," Journal of the American
Statistical Association, Vol. 57, June 1962, pp. 348-68.
Ziemer, R., W. N. Musser, and R. C. Hill, 1980, "Recreational Demand
Equations: Functional Form and Consumer Surplus," American Journal of
Agricultural Economics, Vol. 62, February 1980, pp. 136-41.
9-12
-------
APPENDIX A
SAMPLE DESIGN
This appendix provides a justification for the sampling sizes and the sam-
pling protocol employed in the project.
A.1 SAMPLE SIZE JUSTIFICATION
One approach for using the survey information requires that many of the
parameters to be estimated in this study be treated as proportions--for exam-
ple, the proportion of adults who participate in water-related recreation activi-
ties. Accordingly, the proposed sample sizes were determined by computing
the sample size required to estimate proportions of the underlying population
(i.e., households in the Monongahela River basin).
The required sample size depends upon the desired precision of the pro-
portion estimates. The sample size required to produce an estimate, p, within
6 units of the true population proportion, p, with a percent certainty depends
upon 6, p, and a. Obviously, it is desirable to make 6 small and a large.
However, decreasing 6 and increasing a each requires an increase in the re-
quired sample size. Additionally, a 6 value considered precise for large p
values is not necessarily precise for small p values. For example, let 6 =
0.10, pj = 0.85, and p2 = 0.05. Then, pi ± 6 is equal to 0.85 ± 0.10, which
is relatively precise. However, p2 ± 6, which is equal to 0.05 ± 0.10, is not
very precise.
Table A-1 shows the sample sizes needed to detect a specific difference
with power 1 - p. The crucial specific differences for this project were those
in estimated values for the willingness to pay for different levels of water
quality and differences in estimates of option and existence values for the
Monongahela River.
An example using estimated coefficients of variation (which are equal to
the standard error of the estimate divided by the mean estimate, or simply a
method of comparing the variation in the measured benefits) from related stud-
ies, shown in Table A-2, will explain Table A-1. If the coefficient of variation
is equal to 0.2 (as was the case in the Walsh et al. [1978] South Platte River
Basin Study for Denver residents' willingness to pay for existence values), a
sample size of 68 is necessary to detect a 10 percent difference in the mean
value with 95 percent confidence that the difference is different from zero and
a 10-percent chance of not rejecting the null hypothesis (A = 0) when it is
false. If there is little or no variation in the estimates, small differences can
be detected with minimal sample size. However, considerable variation in esti-
mated values will mean that the sample size at 384 may not be able to detect
small differences in the estimates. Thus, when proportions are estimated,
A-1
-------
Table A-1. Sample Sizes Needed to Detect a Specified Difference
With Power 1 - P =======
CV = coefficient of variation (oe/Mc)
Detection
level (A) 0.1 0.2 0.3 0.4 0.5
(a) a = Type I error = 0.05, p = Type II error = 0.1
0.06 M 48 190 428 760 1,189
0.08 p 27 107 241 428 669
0.10 M 17 68 154 274 428
0.15 jj 8 30 68 122 190
0.20 u 4 17 39 68 107
0.25 MC 3 11 25 44 68
(b) a = 0.05, p = 0.25
0.06 MC
0.08 MC
0.10 MC
0.15 MC
0.20 nc
0.25 MC
30
17
11
5
3
2
120
67
43
19
11
7
269
151
97
43
24
16
478
269
192
77
43
28
748
421
269
120
67
43
ao is the common standard deviation for both the treatment and control
responses under the model, and p is the mean response (usage level) for
the control. The sample size is calculated as n = 2(CV/A)2(a* + ai_p)2'
where z is the standard normal variate.
relative precision is often considered as the most appropriate basis for deter-
mining sample size. This is accomplished by requiring that p lie within p6
units of the true p value with a percent certainty for smallest proportion of
interest. In the above example, the estimate of the small p value would
change from 0.05 ± 0.10 to 0.05 ± 0.005, which is a much more precise esti-
mate. Obviously, this method significantly increases the required sample sizes
for small p values.
Table A-3 contains minimum sample sizes for p to be within p6 units of p
with 95 percent certainty (in the sense of repeated sampling) for various
values of p and 6, assuming simple random sampling. The p values to be
estimated in the study are unknown and will probably vary considerably from
one activity to another. Therefore, it is impossible to determine exactly the
appropriate sample size. Based on past work it is reasonable to assume that
A-2
-------
Table A-2. Coefficients of Variation for Selected Benefits Estimates
Study 1a
Measured
benefit CV n
Boatable 0.05 748
water
quality
Fishable 0.05 748
water
quality
Swimmable 0.05 748
water
quality
Study 2b
Measured
benefit CV n
Existence 0.20 88
value
(user)
Existence 0.33 88
value
(user)
Existence 0.63 15
value
(nonuser)
Bequest 0.93 15
value
(nonuser)
Measured
benefit •
Aesthetic
health
Aesthetic
health
Aesthetic
health
Aesthetic
health
Aesthetic
health
Study 3C
CV
and 0.38
and 0.34
and 0.43
and 0.05
and 0.61
n
10
10
9
7
8
f*See Mitchell and Carson [1981]
°See Walsh et al. [1978].
See Brookshire et al. [1979].
Table A-3. Required Sample Size for Estimates of p to be
Within p6 Units of p, Assuming Simple Random Sampling
p\a
0.01
0.05
0.10
0.25
0.35
0.40
0.50
0.75
0.95
0.05
152,127
29,196
13,830
4,610
2,854
2,305
1,537
512
81
0.10
38,032
7,299
3,457
1,152
713
576
384
129
21
0.15
16,903
3,244
1,537
512
317
256
171
57
9
0.20
9,508
1,825
864
288
178
144
96
33
6
0.25
6,085
1,168
553
184
114
92
61
21
4
A-3
-------
most p values will be in the range of 0.35 to 0.40 or higher. Ditton and
Goodale [1973] found that 69.2 percent of the residents in the Green Bay,
Wisconsin, area had engaged in water-related outdoor recreation within the last
year. The 1977 outdoor recreation survey conducted by the Department of the
Interior determined that, with this assumption, a reasonably precise estimate
can be formed by requiring that 6 = 0.20 (i.e., p6 = (0.35)(0.20) = 0.07 or
p6 = (0.40)(0.20) = 0.08). These values of p and 6 produce a required sample
size in the range of 144 to 178. These estimates are based on simple random
sampling and need to be increased because of the effects of a cluster sample
design. That is, the area sampling design requires expansion of the recom-
mended sample size. The recommended sample size also assumed a 20-percent
nonresponse rate. It should be recognized that the proposed sample size will
give less precise estimates for p values below the 0.35 to 0.40 range and more
precise estimates for p values above the range. Since the coefficients of vari-
ations for p shown in Table A-1 are approximately one and one-half times
larger than the coefficients of variations in Table A-2, the recommended sample
size should yield adequate power for detecting differences in the willingness to
pay and option and existence values.
A.2 SAMPLING PROTOCOL
Using 1970 census computer data tapes (more up-to-date data were not
available at the time of the study since the 1980 census computer data tapes
had not been released) for Enumeration Districts and Block Groups (ED/BGs),
noncompact clusters of approximately seven households were constructed. The
1970 data were adjusted by county using preliminary 1980 census data to more
accurately reflect the present. Additionally, the 1970 occupancy rate and the
estimated response rate were taken into account in determining the cluster size.
The clusters were constructed into three groups once they were stratified.
The groups are those households in (1) Pittsburgh, (2) a place other than
Pittsburgh, and (3) not in a place. Fifty-one clusters were selected. The
number of clusters selected from each stratum were proportional to the number
of households in that stratum. For example, since 61 percent of the households
in the five-county area are located in Pittsburgh, 51(0.61) = 31 clusters were
selected from Pittsburgh. The clusters were selected with equal probabilities
within each stratum. Because of the proportional allocation of the sample to
the strata, the probabilities of selection for all clusters were nearly equal.
Because the clusters were contained in ED/BGs, the general physical loca-
tion of the cluster is known. Interviewers were sent to the field to count and
list all households in the ED/BGs that contain the selected clusters. The lists
produced during the counting and listing exercise were used to identify the
specific households in the selected cluster. If the number of households did
not exceed a predetermined number, all households in the cluster were con-
tacted. For those clusters that were too large, the list was used to determine
a subsample of the cluster to be contacted.
Once the households to be contacted were identified, the interviewers
conducted a preliminary visit and compiled a roster of all adults living in the
household. One of the adults was randomly selected (with equal probabilities)
for interview.
A-4
-------
APPENDIX B
SURVEY FORMS AND PROCEDURES
PART 1
HOUSEHOLD CONTROL FORM
Part 1 of this appendix contains the household control form used by field
interviewers to provide assignment and other background information.
B-1
-------
ENlim.it ing Item-fits of W;itor Quality
RTI I'rojiicl. No. 4IU-2222-2
ASSIGNMENT INFORMATION
A. Study No. [.2 [? \2 |2 |
K. Address
OMII No. 2000-0381
Approv.il Expires: 9/30/82
HOUSEHOLD CONTROL FORM
Form No. 01
B. PSU/Sognont No. j_j - LjZI_J_l_l c- Mousing Unit No. [_[ _j | D. Iiilervicwi-
(7-12) (14-16")"
r No.
II.
00
M
(Numhrr/Stri-cl/RFD) (Aparlncnt No.)
KEC(IKI) OF CONTACTS - ENUMERATION AND SAMPLE INDIVIDUAL
Day of
Wi-.-k
l)jtc
Tine
an
pio
an
pm
an
PM
»m
pm
am
pn
am
iim
pin
am
Notes
Result
Code
FI
(City)
DICED
(State) (Zip)
III. CONTACT RESULT CODES (CIRCLE BELOW THE FINAL RESULT CODE FOR EACH TYI'E
OF CONTACT)
Household Enumeration Contact Sample Individual Contact Codes
Codes
01 Enumeration Completed
02 No Enumeration Eligible
at Home
03 Enumeration Respondent
Ureakoff; Partial Data
04 Enumeration Respondent
Refused
05 Language Barrier
*06 Vacant Housing Unit
*07 Not a Housing Unit; e.g., (21-22)
Merged, Demolished,
Group Quarters, Non-
Residential
08 Other (EXPLAIN IN "COMMENTS")
(18-19)
20 Interview Completed (CIRCLE VERSION
ADMINISTERED IN SECTION VI.0)
21 Appointment M:nlc
22 Interview Rre.ikoff; Partial Data
23 Sample Individual not Hume
24 Refusal
25 Language It.irrier
26 Other (EXPLAIN IN "COMMENTS")
IV. Sour(i- ol Information for Result Codes 06, 07
Niimc
Nuinhi-r/Slrccl/RKD
'i:i"iy/.st»fc/x7p
( ) ..__
ifir|>luiiii- Number
V. COMMENTS
-------
VI. HOUSEHOLD ENUMERATION AND SAMPLE INDIVIDUAL SELECTION
Hello, !'• (NAME) with the Research Triangle Institute of North Carolina. We are doing a
household survey for a government agency to study levels of water quality and- some outdoor
recreational activities people take part in both near and on ponds, lakes, streams, and rivers
in the Pittsburgh area. Your household has been randomly selected along with others in this
area to be interviewed. In order to determine who in your household should be interviewed, I
would like to ask a few questions about the residents of your household. I am required to
talk with a household member who is 16 years of age or older. (ASK IF NECESSARY, ARE YOU 16
YEAKS OK OLIIER?)
HOUSEHOLD ROSTKK
1.
2.
0)
co
First, are there any occupied or vacant living quarters other than your own (FOR SINGLE
UNIT STRUCTURE) in this structure or on this property? (FOR MULTI-UNIT STRUCTURES?) in
this unit?
(CIRCLE NUMBER BELOW FOR RESPONSE)
1 YES (ADD TO LIST OF ADDED HOUSING UNITS IF REQUIRED BY MISSED HU RULES)
2 NO
Now, I would like to ask some general questions about you and all of the other people who
live in this household, including friends and roomers. Let's list the people who live
here in order of age, beginning with the oldest first. (ENTER AGES IN DESCENDING AGE
ORDER IN COLUMN B.) 1 have listed ages for persons who are (READ AGES). Is there anyone
else living here now? (IF YES, ENTER AGE(S) AND CORRECT AGE ORDER IN COLUMN B, IF
ASK THE SEX FOR EACH PERSON LISTED AND CIRCLE THE CORRECT CATEGORY IN COLUMN C.
Which person is the head of the household? (WRITE THE WORD "HEAD" IN COLUMN D FOR THE
LINE NUMBER OF THE PERSON CONSIDERED THE HEAD OF HOUSEHOLD.)
PERSONS LISTED ASK THEIR RELATIONSHIP TO THE HEAD OF HOUSEHOLD AND ENTER IN
FOR OTHER
COLUMN 0.
SELECT THE HOUSEHOLD MEMBER TO BE INTERVIEWED FROM AMONG ONLY THOSE PERSONS 18 YEARS OR
OLDER (ELIGIBLE HOUSEHOLD MEMBERS). REFERRING TO THE AGES LISTED IN THE ROSTER,
DETERMINE THE NUMBER OF PERSONS WHO ARE 18 YEARS OR OLDER AND DRAW A LINE ACROSS THE
ROSTER TO SEPARATE THOSE PERSONS FROM THOSE 17 OR YOUNGER. LOCATE ON THE TABLE BELOW THE
ROSTER THE NUMBER OF ELIGIBLE HOUSEHOLD MEMBERS. DIRECTLY BELOW THE NUMBER OF HOUSEHOLD
MEMBERS, FIND THE ROSTER LINE NUMBER SELECTED. CIRCLE THE SELECTED LINE NUMBER ON THE
ROSTER.
(You havc/PEKSON has) been selected as the person to be interviewed. (ASK FOR THE NAME
OF THE PF.RSOU SELECTED AND ENTER HERE) .
PRINT NAME OF SELECTED INDIVIDUAL
IF ENUMERATION RESPONDENT HAS BF.F.N SELECTED. ATTEMPT TO COMPLETE INTERVIEW. IF ANOTHER
PERSON, DETERMINE IF HE/SHE IS AVAILABLE OR WHEN HE/SHE WILL BE.
8. QUESTIONNAIRE VERSION ADMINISTERED (CIRCLE VERSION)
ARC
(Hi CAIi
D
1
A
01
02
03
04
OS
06
07
08
09
10
11
B
AGE
C
SEX
M F
1 2
1 2
1 2
1 2
1 2
1 2
1 2
1 2
1 2
1 2
1 2
1 2
1)
HOUSEHOLD HEAD/RELATIONSHIP
CAKI) 2
i-u, mil*
IIUill CAKI) 1
COL. 80-2
(I9-2U)
(24-28)
(2'J-:i.l)
CM-:«8)
<™-4»
(44-48)
(4-.-5:i)
(M-5K)
(b'J-t.1)
(64-68)
(60-7:,)
CARD 1.
COL. 80
" '
HOUSEHOLD SIZE;
! 2 * H 5
11111
RESPONDENT NO.:
-------
PART 2
COUNTING AND LISTING EXAMPLES
Figure B-1. Sample segment map.
B-4
-------
SEGMENT ID
PREPARED BY M* 6PPLS.2.
REVIEWED Bit
PLACETS Ugu£6i\)£ .rfi
• )
DATE /a ( •j.fal'Zt
DATE
Figure B-2. List unit sketch.
B-5
-------
Page
,f A.
• -./^ ;J
LIST OF HOUSING US1TS * /1 /;
.
Daca Listed 1 & ] "2,£* 1 2T /
(MontiiJ (Day) (Tear)
Sezment So3-H*Z"7 4- Part So.
"..' 1 1
HD
i Mo.
1
2
3
4
St.
Ho.
*«-7
^n
v^jr
9lz<»
6
7 2.-2JO
8
9
1 °
/ I
/ 2
I 3
/ 6
15
J 6
1 7
4-o«J
Zf£-
3/i?
3*7
•3««9
^7
Atoo
i slvot.
f")
,Z.o
^0^
V/o
1H
3 4 j
^S-,^<*/ AJiL£/
,. •• '
fiSrl
2.A/TS -rr
r r9TT?t3«iO fti?S.
i
i
4-^ ^T
1 «
PTZ.ISC.H.LA
1
4
i
IZont ft
PRjfi<.lLCft
,
i
5 T2,"b /. TTZ,^
Iffaav -S/^/oA- U>H StinTrens
S«sJ -TA/s^^f^xT (STZ.Tz,^
*&*t^*t*j,&ru*y BL.PILI^R:
flS^Ot SbOUftiC Oui
-------
PART 3
DEBRIEFING AGENDA
Part 3 of this appendix contains the agenda used during the December
1981 interviewer debriefing session.
Estimating Recreation and Related Benefits of Water Quality
RTI Project 2222-2
DEBRIEFING AGENDA
Thursday, December 10, 1981
Welcome and Introductions
Evaluation of Training
Effectiveness of home study materials
Effectiveness of classroom sessions
Adequacy of training time
Areas encountered in interviewing that should have been covered in
training
Usefulness of specifications and manual
Deficiencies in specifications and manual
Evaluation of Assignment Materials and Procedures
Content and layout of Household Control Form
Accuracy of sample member assignment data (names, addresses,
etc.)
Tracing/locating activities required
Deficiencies in materials and procedures
Obtaining Respondent Cooperation
Gaining access to sample members
Explaining purposes of the survey
Obtaining permission to complete enumeration
Obtaining permission to complete the interview
Intervention by other household/family members
Effectiveness of "Dear Resident" and other informational material
Characteristics of nonrespondents and reasons for nonresponse
Procedures for converting refusing sample members
B-7
-------
Conducting the Interviews
Household enumeration procedures and problems
Usefulness of handout materials
Deficiencies of handout materials
Section-by-section review of all questionnaires
(1) What questions usually worked well and were understood by all
respondents?
(2) What questions frequently were difficult to administer or were
misunderstood by respondents?
(3) What questions appeared to elicit reliable responses with minimal
probing?
(4) What questions frequently yielded "Don't Know" responses?
(5) What questions were respondents reluctant to answer? What
reasons, if any, were stated?
(6) What category of respondents (i.e., disabled, widowed, older
men, etc.) had the most difficulty in responding to the ques-
tions?
(7) What category of respondents were most reluctant to answer
certain questions?
Problems with layout or design of each instrument
Problems in the interview setting
Problems with interview length
Questions or concerns expressed by respondents
Administrative Procedures
Status reporting
Communications with supervisor/central office
Resolution of field problems
Evaluation of callback requirements
Recommendations for Future Similar Surveys
Respondent informational material
Assignment materials and procedures
Contacting, locating, and securing cooperation
Instruments and handouts
Administrative materials and procedures
B-8
-------
PART 4
QUALITY CONTROL PROCEDURES
The quality control procedures used during and after administration of
the survey questionnaire, including both field editing and validation proce-
dures, are described below:
FIELD EDITING
Field interviewers were responsible for conducting a thorough field edit
of each completed survey instrument. Interviewers were provided with an edit
instruction for the instruments to insure that significant edit checks were made.
The importance of. the field editing process and procedures to be followed were
emphasized in the interviewer's manual and received attention as part of inter-
viewer training.
Field editing by interviewers involved two steps. First, each completed
instrument was scanned for completeness at the conclusion of each interview
while the interviewer was still in the respondent's presence. If any incomplete
or omitted items were detected, the missing data were obtained. Second, inter-
viewers thoroughly edited each completed instrument before submitting their
work. Any omissions or problems noted during this edit were resolved by a
telephone call or, if necessary, a return visit to the respondent by the inter-
viewer. These field edit procedures were especially important as an aid to
insure that high quality and complete data were received from the field.
To insure quality control of the interviewing process, each interviewer's
completed interviews were edited at the Research Triangle Institute (RTI) dur-
ing the fieldwork period. The editor used edit specifications that focus on the
key elements of each document, and interviewers received ongoing assessments
of the quality of their work by telephone. In addition, where graphic instruc-
tion to an interviewer was helpful to explain the nature of an error, photo-
copies were made of questionnaire pages to show interviewers exactly what the
problem was.
VALIDATION
A major quality control procedure involved validation of a random sample
of 10 percent of the interviews conducted. This procedure was accomplished
through telephone contacts with participating sample members. The validation
contact was designed to determine whether the interview actually took place on
or about the date reported; whether the interviewer secured a complete, cur-
rent household roster; whether appropriate sample member selection procedures
were followed; and whether the entire interview schedule was completed. Also,
key items were asked and responses compared with original responses reported
by the interviewer. In addition, the contact elicited other information concern-
ing the interviewer's performance.
B-9
-------
APPENDIX C
SURVEY ANALYSIS: SUPPORTING TABLES
This appendix provides supporting statistical analysis for the option price,
user value, and option value results presented in Chapters 4 and 5. The
tables in general focus on three issues: (1) estimates with outliers excluded;
(2) estimates with protest bids excluded; and (3) t-tests for differences from
zero.
In addition, Table C-16 supports the analysis in Chapter 6. This table
shows benefit estimates from an alternative contingent-ranking specification.
C-1
-------
Table C-1. Student t-Statistics_ of Characteristics for
H : Xj = X2
Characteristic
Ownership or use of a boat
Participation in any outdoor
recreation in the last year
Numerical rating of the
Monongahela
Rating for a particular site
Length of residence
Education
Race
I ncome
Age
Sex
User vs. nonuser
2.471b
10.746b
0.365
5.988b
0.242
1.655
-0.804
1.124
-5.995b
-1.338
Zero vs.
nonzero bids
-1.589
-4.818b
-1.369
-3.205b
0.167
-2.031b
1.699
-1.713
4.942b
-0.347
at-statistics are derived from the results reported in Chapter 4.
Denotes significance at the 0.05 level.
C-2
-------
Table C-2. Estimated Option Price for Changes in Water Quality.:
Effects of Instrument and Type of Respondent—All Respondents
Tvpe of respondent
Change in
water quality
User
X s n
Nonuser
X s n
Combined
X s n
1. Iterative bidding framework--starting point = $25 (Version C)
D to E (avoid) 21.7 18.6 24 23.7 32.9 54 23.1 29.2 78
D to C 15.0 16.4 24 11.9 15.6 54 12.9 15.8 78
C to B. 9.4 13.7 24 5.7 10.7 54 6.9 11.7 78
D to BD 25.4 27.5 24 41.4 51.5 54 20.1 25.3 78
Combined: all levels 47.1 41.8 24 17.7 24.1 54 43.1 48.5 78
2. Iterative bidding framework—starting point = $125 (Version D)
D to E (avoid) 89.5 70.3 22 44.6 84.1 50 58.3 82.4 72
D to C 63.9 53.5 22 29.7 56.0 50 40.1 57.1 72
C to B. 41.8 54.2 22 19.9 51.1 50 26.6 52.6 72
D to BD 111.8 94.4 22 51.3 102.1 50 69.8 103.1 72
Combined: all levels 201.4 149.8 22 95.9 177.6 50 128.1 175.5 72
3. Direct question framework (Version B)
D to E (avoid) 42.6 67.8 23 13.5 35.2 51 22.5 49.2 74
D to C 27.9 42.7 23 9.3 22.3 51 15.1 31.1 74
C to B. 24.0 49.5 23 7.7 22.5 51 12.8 33.8 74
D to B 53.0 84.6 23 17.7 43.5 51 28.7 61.0 74
Combined: all levels 95.7 130.7 23 31.2 77.0 51 51.2 100.6 74
4. Direct question framework: payment card (Version A)
D to E (avoid)
D to C
C to B
D to B°
Combined: all levels
57
46
22
70
127
.1
.0
.5
.6
.7
92.8
71.1
45.3
112.5
159.4
24
24
24
24
24
38.9
15.9
5.6
21.7
60.6
68.8
30.3
17.3
42.5
96.1
51
51
51
51
51
44.7
25.5
11.0
37.3
82.1
77.
48.
30.
75.
123.
1
9
1
4
0
75
75
75
75
75
aThe two respondents who did not complete the questionnaire are excluded.
D to B includes respondents who were willing to give an amount only for fishable
or swimmable water and respondents who were willing to pay some amount to
avoid the decrease in water quality in addition to the improvements in water
quality. ,
C-3
-------
Table C-3. Estimated Option Price for Changes in Water Quality: Effects
of Instrument and Type of Respondent—Protest Bids Excluded
Change in
water quality
User
X s
Type of respondent
Nonuser
n X s n
Combined
X s n
1. Iterative bidding framework—starting point = $25 (Version C)
D to E (avoid) 27.4 16.7 19 28.4 34.2 45 28.1
D to C 18.9 16.3 19 14.3 16.1 45 15.7
C to B 11.8 14.5 19 6.9 11.3 45 8.4
D to B 32.1 27.1 19 21.2 25.0 45 24.5
Combined: all levels 59.5 38.1 19 49.7 52.7 45 52.6
2. Iterative bidding framework—starting point = $125 (Version D)
D to E (avoid)
D to C
C to B
D to B
a
93.8
66.9
43.8
117.1
69.0
52.8
54.7
93.3
21
21
21
21
54.4 90.2
36.2 60.0
24.3 55.5
62.6 109.8
41
41
41
41
67.7
46.6
30.9
81.0
29.9
16.2
12.4
25.9
48.7
85.1
59.1
55.6
107.0
Combined: all levels 210.0 146.4 21 117.0 190.1 41 148.8 180.9
3. Direct question framework (Version B)
D to E (avoid)
D to C
C to B
D to B
a
51.6
33.8
29.1
64.2
71.7
44.9
53.3
89.4
19
19
19
19
40.
25.
Combined: all levels 115.8 135.7 19
18.6
12.8
10.6 25.9
24.4
43.0
49.7
87.8
37
37
37
37
37
29.8
19.9
16.9
37.9
67.7
54.
34.
38,
67.
110.8
4. Direct question framework: payment card (Version A)
64
64
64
64
64
62
62
62
62
62
56
56
56
56
56
D to E (avoid)
D to C
C to B
D to Ba
Combined: all levels
65
52
25
80
146
.2
.6
.7
.7
.0
96.
73.
47.
117.
162.
7
8
7
1
6
21
21
21
21
21
49.6
20.3
7.1
27.6
77.3
74.3
33.0
19.3
46.3
102.6
40
40
40
40
40
55.0
31.4
13.5
45.9
100.9
82.2
52.6
32.9
81.3
129.3
61
61
61
61
61
D to B includes respondents who were willing to give an amount only for fishable
or swimmable water and respondents who were willing to pay some amount to avoid
the decrease in water quality in addition to the improvements in water quality.
C-4
-------
Table C-4. Estimated User Values for Changes in Water Quality^
Effects of Instrument and Type of Respondent--All Respondents
Type of respondent
Change in
water quality
User Combined
X
s n X s
n
1. Iterative bidding framework—starting point = $25 (Version C)
D to E (avoid) 5.2 11.5 24 1.6 6.7 78
D to C 3.3 7.0 24 1.0 4.1 78
C to B. 4.0 7.4 24 1.2 4.4 78
D to BD 8.3 13.5 24 2.6 8.3 78
Combined: all levels 13.5 23.3 24 4.2 14.2 78
2. Iterative bidding framework—starting point = $125 (Version D)
D to E (avoid) 38.0 58.9 22 11.6 36.5 72
D to C 31.1 50.0 22 9.5 30.8 72
C to B. 32.0 52.9 22 9.8 32.4 72
D to BD 69.3 102.1 22 21.2 64.1 72
Combined: all levels 107.3 147.3 22 32.8 94.3 72
3. Direct question framework (Version B)
D to E (avoid) 19.1 37.6 23 5.9 22.5 74
D to C 18.0 37.7 23 5.6 22.3 74
C to B. 11.9 31.6 23 3.7 18.2 74
D to BD 29.9 62.3 23 9.3 36.9 74
Combined: all levels 49.0 81.9 23 15.2 50.4 74
4. Direct question framework: payment card (Version A)
D to E (avoid)
D to C
C to B
D to BD
Combined: all levels
20.2
30.2
16.0
46.7
66.9
35.0
73.2
42.7
113.5
121.3
24
24
24
24
24
6.5
9.7
5.1
14.9
21.4
21.7
43.2
25.0
67.0
74.6
75
75
75
75
75
The two respondents who did not complete the questionnaire are ex-
cluded.
D to B includes respondents who were willing to give an amount only for
fishable or swimmable water and respondents who were willing to pay some
amount to avoid the decrease in water quality in addition to the improve-
ments in water quality.
C-5
-------
Table C-5. Estimated User Values for Changes in Water Quality:
Effects of Instrument and Type of Respondent—
Protest Bids Excluded
Type of respondent
Change in User Combined
water quality X s n X s n
1. Iterative bidding framework—starting point = $25 (Version C)
D to E (avoid) 6.6 12.6 19 2.0 7.4 64
D to C 4.2 7.7 19 1.3 4.5 64
C to B 5.0 8.0 19 1.5 4.9 64
D to Ba 10.5 14.4 19 3.1 9.1 64
Combined: all levels 17.1 25.1 19 5.1 15.6 64
2. Iterative bidding framework—starting point = $125 (Version D)
D to E (avoid) 39.8 59.7 21 13.5 39.1 62
D to C 32.6 50.7 21 11.0 32.9 62
C to B 33.6 53.7 21 11.4 34.7 62
D to Ba 72.6 103.4 21 24.6 68.6 62
Combined: all levels 112.4 148.9 21 38.1 100.7 62
3. Direct question framework (Version B)
D to E (avoid) 23.1 40.4 19 7.8 25.6 56
D to C 21.8 40.6 19 7.4 25.4 56
C to B 14.4 34.4 19 4.9 20.9 56
D to Ba 36.2 67.1 19 12.3 42.1 56
Combined: all levels 59.3 86.9 19 20.1 57.2 56
4. Direct question framework: payment card (Version A)
D to E (avoid)
D to C
C to B
D to Ba
Combined: all levels
23.1
34.5
18.3
53.3
76.4
36.6
77.5
45.3
120.2
127.1
21
21
21
21
21
8.0
11.9
6.3
18.4
26.3
23.8
47.7
27.6
73.9
82.0
61
61
61
61
61
aD to B includes respondents who were willing to give an amount only for
fishable or swimmable water and respondents who were willing to pay some
amount to avoid the decrease in water quality in addition to the improve-
ments in water quality.
C-6
-------
Table C-6. Estimated Option Values for Changes in Water Quality:
of Instrument and Type of Respondent—All Respondents
Effects
Type of respondent
Change in
water quality
User
Nonuser
Combined
n
n
X
1. Iterative bidding framework—starting point = $25 (Version C)
D to E (avoid) 16.5 17.0 24 23.7 32.9 54 21.5
D to C 11.7 13.8 24 11.9 15.6 54 11.9
C to B. 5.4 9.9 24 5.7 10.7 54 5.6
D to Bb 17.1 21.5 24 41.4 51.5 54 17.5
Combined: all levels 33.5 33.2 24 17.7 24.1 54 39.0
29.1
15.0
10,
23,
46.6
2. Iterative bidding framework—starting point = $125 (Version D)
D to E (avoid)
D to C
C to B.
D to BD
Combined: all levels
51.6
32.7
9.8
42.5
69.9
48.2
28.2
66.5
22
22
22
22
44.6
29.7
19.9
84.1
56.0
51.1
51.3 102.1
50
50
50
50
46.7
30.6
16.8
48.6
3. Direct question framework (Version B)
D to E (avoid) 23.5 41.6 23 13.5
D to C 9.9 22.9 23 9.3
C to B. 12.1 28.6 23 7.7
D to BD 23.1 50.5 23 17.7
Combined: all levels 46.7 84.5 23 31.2 77.0
35,
22,
22.
43.
51
51
51
51
51
16.6
9.5
9.1
19.4
36.0
37.
22.
24.
45.
79.1
78
78
78
78
78
79.6 72
53.4 72
45.3 72
92.3 72
94.1 119.8 22 95.9 177.6 50 95.3 161.3 72
74
74
74
74
74
4. Direct question framework: payment card (Version A)
D to E (avoid)
D to C
C to B.
D to BD
Combined: all levels
36.9
15.8
6.5
24.0
60.8
73
25
21
43
115
.9
.4
.1
.6
.2
24
24
24
24
24
38.9
15.9
5.6
21.7
60.6
68.8
30.3
17.3
42.5
96.1
51
51
51
51
51
38.
15.
5.
22.
60.
3
9
9
4
7
70.0
28.7
18.5
42.6
101.8
75
7<
7E
7E
7E
The two respondents who did not complete the questionnaire are excluded.
DD to B includes respondents who were willing to give an amount only foi
fishable or swimmable water and respondents who were willing to pay som»
amount to avoid the decrease in water quality in addition to the improvements ir
water quality.
C-7
-------
Table C-7. Estimated Option Values for Changes in Water Quality:
of Instrument and Type of Respondent--
Protest Bids Excluded
Effects
Type of respondent
Change in
water quality
User
X s n
Nonuser
X s n
Combined
X s n
1. Iterative bidding framework--starting point = $25 (Version C)
D to E (avoid) 20.8 16.6 19 28.4 34.2 45 26.2
D to C 14.7 14.0 19 14.3 16.1 45 14.5
C to B 6.8 10.7 19 6.9 11.3 45 6.9
D to Ba 21.6 22.1 19 21.2 25.0 45 21.3
Combined: all levels 42.4 31.9 19 49.7 52.7 45 47.5
2. Iterative bidding framework--starting point = $125 (Version D)
D to
D to
E (avoid)
C
C to B
D to B
Combined: all levels
54.0
34.3
10.2
44.5
70.7
48.8
28.8
67.4
21
21
21
21
54.4 90.2
36.2 60.0
24.3 55.5
62.6 109.8
41
41
41
41
54.3
35.6
19.5
56.5
30.1
15.4
11.1
24.0
47.3
83.5
56.1
48.4
97.3
98.6 120.9 21 117.0 190.1 41 110.7 169.0
3. Direct question framework (Version B)
D to E (avoid) 28.5 44.4 19 18.6 40.3
D to C 12.0 24.8 19 12.8 25.4
C to B 14.7 31.0 19 10.6 25.9
D to Ba 28.0 54.5 19 24.4 49.7
Combined: all levels 56.5 90.2 19 43.0 87.8
4. Direct question framework: payment card (Version A)
37
37
37
37
37
21.9
12.6
12.0
25.6
47.6
41.6
25.0
27.5
50.9
88.0
64
64
64
64
64
62
62
62
62
62
56
56
56
56
56
D to E (avoid)
D to C
C to B
D to B
Combined: all levels
42
18
7
27
69
.1
.1
.4
.4
.5
77
26
22
45
121
.7
.5
.5
.7
.0
21
21
21
21
21
49.6
20.3
7.1
27.6
77.3
74.3
33.0
19.3
46.4
102.6
40
40
40
40
40
47.0
19.5
7.2
27.5
74.6
75.0
30.7
20.3
45.8
108.3
61
61
61
61
61
aD to B includes respondents who were willing to give an amount only for fishable
or swimmable water and respondents who were willing to pay some amount to avoid
the decrease in water quality in addition to the improvements in water quality.
C-8
-------
Table C-8. Option Price_Student t-Statistics for
H
X = 0
With Outliers and Protest Bids Excluded
Payment card
Level D to E
Level D to C
Level C to B
Total D to B
Total E to B
Direct question
Level D to E
Level D to C
Level C to B
Level D to B
Total E to B
Iterative Bidding $25
Level D to E
Level D to C
Level C to B
Total D to B
Total E to B
Iterative bidding $125
Level D to E
Level D to C
Level C to B
Total D to B
Total E to B
User
4.56
2.62
-
2.49
4.15
2.86
2.92
2.34
3.01
3.91
7.14
5.07
3.57
5.15
6.80
5.73
4.48
2.74
4.54
5.70
Nonuser
4.22
3.94
2.34
3.82
4.81
3.05
2.93
2.26
2.86
3.03
5.20
5.97
3.86
5.63
6.05
4.27
3.27
-
3.32
4.37
Total sample
5.59
4.36
2.85
4.04
6.34
3.86
3.93
3.22
4.03
4.67
7.20
7.81
5.23
7.54
8.48
6.42
5.16
3.28
5.21
6.46
Only those values that are significant at the 0.05 level are reported.
C-9
-------
Table C-9. User Value Student t-Statistics for
H : X = 0
o
With Outliers and Protest Bids Excluded3
User Total sample
Payment card
Level D to E 2.37 2.17
Level D to C
Level C to B
Total D to B
Total E to B 2.29 2.11
Direct question
Level D to E 2.15 2.01
Level D to C
Level C to B
Total D to B
Total E to B 2.71 2.42
Iterative bidding $25
Level D to E 2.28 2.12
Level D to C 2.39 2.21
Level C to B 2.73 2.46
Total D to B 3.18 2.76
Total E to D 2.97 2.62
Iterative bidding $125
Level D to E 2.46 2.23
Level D to C
Level C to B
Total D to B 2.22 2.05
Total E to B 2.46 2.23
aOnly those values that are significant at the 0.05 level are reported.
Table C-10. Option Value Student t-Statistics for Differences
in Means Between Bidding Methods—Outliers and Protest
Bids Excluded
User Total sample
Iterative bidding $25 vs. iterative bidding $125
Level D to E -2.14 -1.97
Total E to B -2.11 -
aOnly those values that are significant at the 0.05 level are reported.
C-10
-------
Table C-11. Regression Results for Option Price Estimates of Water
Quality Changes—Protest Bids Excluded
Water quality change9
Independent variables
Intercept
Sex (1 male)
Age
Education
Income
Direct question
Iterative bidding game ($25)
Iterative bidding game ($125)
User (1 if user)
Willing to pay cost of water pollution
(1 if very much or somewhat)
Interviewer 1
Interviewer 2
Interviewer 3
Interviewer 4
Interviewer 5
Interviewer 6
Interviewer 7
Interviewer 8
Interviewer 9
R2
F
Degrees of freedom
D to E (avoid)
-22.132
(-0.510)
23.756
(2.104)
-0.314
(-0.983)
3.826
(1.244)
0.0006
(1.299)
-31.506
(-2.208)
-22.986
(-1.671)
28.606
(2.028)
12.896
(1.097)
18.719
(1.601)
30.857
(1.325)
7.754
(0.355)
-24.009
(-1.32)
19.348
(0.501)
6.982
(0.316)
36.351
(0.716)
42.280
(1.815)
11.136
(0.510)
49.806
(1.385)
0.281
3.61
166
D to C
-18.171
(-0.627)
5.268
(0.698)
-0.283
(-1.328)
1.968
(0.956)
0.0002
(0.587)
-13.203
(-1.384)
-13.455
(-1.462)
21.775
(2.308)
10.799
(1.374)
23.848
(3.050)
13.435
(0.862)
15.931
(1.091)
21.959
(1.547)
20.235
(0.783)
3.354
(0.227)
50.645
(1.490)
6.505
(0.418)
25.584
(1.750)
30.573
(1.271)
0.248
2.99
166
C to B
4.690
(0.177)
3.989
(0.577)
-0.239
(-1.221)
0.306
(-0.162)
0.0002
(0.892)
0.777
(0.089)
-5.338
(-0.634)
19.461
(2.252)
10.288
1.430
9.538
1.332
15.658
(1.097)
16.379
(1.224)
8.755
(0.674)
32.428
(1.370)
-4.095
(-0.302)
27.450
(0.882)
7.411
(0.520)
14.498
(1.083)
29.078
(1.320)
0.148
1.61
166
Total
all levels
-25.618
(-0.308)
33.597
(1.555)
-0.869
(-1.423)
5.020
(0.853)
0.001
(1.178)
-44.026
(-1.613)
-41.798
(-1.588)
74.029
(2.743)
35.420
1.575
53.944
(2.411)
54.693
(1.227)
34.788
(0.832)
1.571
(0.039)
66.575
(0.900)
4.168
(0.099)
108.924
(1.121)
58.627
(1.315)
46.024
(1.101)
101 .538
(1.476)
0.276
3.51
166
Total:
improvement
only
-3.486
(-0.069)
9.840
(0,744)
-0.555
(-1.485)
1.194
(0.331)
0.0004
(0.815)
-12.520
(-0.749)
-18.813
(-1.168)
45.423
(2.749)
22.523
(1.636)
35.225
(2.572)
23.836
(0.874)
27.034
(1.057)
25.580
(1.029)
47.227
(1.043)
-2.814
(-0.109)
72.572
(1.220)
16.347
(0.599)
34.888
(1.363)
51.732
(1.228)
0.229
2.74
166
-ratios for the null hypothesis of no association.
C-11
-------
Table C-12. Regression Results for User Value Estimates of Water
Quality Changes—Protest Bids Excluded
Water Quality change8
Independent variables
Intercept
Sex
Age
Education
Income
Direct question
Iterative bidding ($25)
Iterative bidding ($125)
Willing to pay cost
Interviewer 1
Interviewer 2
Interviewer 3
Interviewer 4
Interviewer 5
Interviewer 6
Interviewer 7
Interviewer 8
Interviewer 9
R*
F
Degrees of freedom
D to E (avoid)
26.618
(1.408)
-0.567
(-0.115)
-0.328
(-2.512)
0.140
(0.104)
0.000002
(0.010)
-1.694
(-0.271)
-5.195
(-0.860)
6.214
(1.006)
4.790
(0.950)
-10.977
(-1.075)
-5.433
(-0.567)
-9.462
(-1.039)
-11.818
(-0.697)
-12.842
(-1.322)
-10.835
(-0.486)
4.895
(0.482)
-10.016
(-1.044)
-2.618
(-0.166)
0.11
1.26
167
D to C
9.513
(0.422)
-7.465
(-1.273)
-0.231
(-1.485)
0.212
(0.132)
0.0001
(0.594)
-5.944
(-0.796)
-11.770
(-1.635)
-2/406
(-0.327)
9.560
(1.591)
-3.649
(-0.300)
4.711
(0.412)
23.386
(2.153)
1.810
(0.090)
-5.401
(-0.466)
9.970
(0.375)
6.735
(0.557)
6.084
(0.532)
-0.119
(-0.006)
0.12
1.32
167
C to B
9.497
(0.546)
-5.447
(-1.204)
-0.172
(-1.431)
0.253
(0.204)
0.0001
(0.452)
-1.312
(-0.228)
-4.114
(-0.740)
5.525
(0.972)
4.808
(1.037)
-7.453
(-0.793)
-1.321
(-0.150)
8.302
(0.990)
-3.542
(-0.227)
-9.620
(-1.076)
-1.871
(-0.091)
1.162
(0.124)
-4.086
(-0.463)
7.050
(0.485)
0.09
0.99
167
Total
all levels
24.423
(0.630)
-11.303
(-1.122)
-0.455
(-1.698)
-0.041
(-0.015)
0.0003
(0.667)
-8.307
(-0.647)
-15.345
(-1.240)
6.233
(0.492)
14.834
(1.436)
-9.504
(-0.454)
4.240
(0.216)
32.793
(1.756)
-0.471
(-0.014)
-12.998
(-0.653)
7.909
(0.173)
15.612
(0.750)
2.539
(0.129)
6.722
(0.208)
0.1.1
1.26
167
Total:
improvement
only
51.041
(1.023)
-11.870
(-0.915)
-0.783
(-2.270)
0.098
(0.028)
0.0003
(0.522)
-10.001
(-0.605)
-20.541
(-1.289)
12.447
(0.763)
19.624
(1.475)
-20.481
(-0.760)
-1.193
(-0.047)
23.331
(0.970)
-12.289
(-0.275)
-25.840
(-1.008)
-2.926
(-0.050)
20.507
(0.765)
-7.478
(-0.295)
4.105
(0.099)
0.12
1.39
167
aNumbers in parentheses are symptotic t-ratios for the null hypothesis of no association.
C-12
-------
Table C-13. Regression Results for Option Value Estimates of Water
Quality Changes—Protest Bids Excluded
Water Quality change3
Independent variables
Intercept
Sex
Age
Education
Income
Direct question
Iterative bidding ($25)
Iterative bidding ($125)
Willing to pay cost
User
R2
F
Degrees of freedom
D to E (avoid)
-3.931
(-0.105)
18.033
(1.745)
-0.341
(-1.172)
3.202
(1.143)
0.0003
(0.830)
-25.304
(-1.872)
-15.199
(-1.164)
25.841
(1.936)
27.643
(2.655)
-18.682
(-1.770)
0.179
4.22
175
D to C
1.879
(0,091)
7.528
(1.324)
-0.302
(-1.885)
1.595
(1.035)
-0.0001
(-0.477)
-5.552
(-0.747)
0.690
(0.096).
27.909
(3.802)
21.039
3.673
-14.078
(-2.424)
0.217
5.39
175
C to B
23.017
(1.205)
5.259
(1.002)
-0.232
(-1.568)
-0.810
(-0.569)
-.0000
(-0.013)
4.980
(0.725)
0.970
(0.146)
17.004
(2.508)
10.588
(2.001)
-9.307
(-1.735)
0.090
1.92
175
Total :
improvement
only
24.897
(0.684)
12.096
(1.209)
-0.544
(-1.928)
0.888
(0.328)
-0.0001
(-0.324)
-0.257
(-0.020)
0.775
(0.061)
45.796
(3.544)
33.146
(3.287)
-24.071
(-2.355)
0.177
4.18
175
Numbers in parentheses are asymptotic t-ratios for the null hypothesis of no association.
C-13
-------
Table C-14. Regression Results for Option Value Estimates of Water
Quality Changes—Protest Bids and Outliers Excluded
* Independent variables
Intercept
Sex
Age
Education
Income
Direct question
Iterative bidding ($25)
Iterative bidding ($125)
User
Willing to pay' cost
Interviewer 1
Interviewer 2
Interviewer 3
Interviewer 4
Interviewer 5
Interviewer 6
Interviewer 7
Interviewer 8
Interviewer 9
R2
F
Degrees of freedom
D to E (avoid)
-35.228
(-1.019)
5.779
(8.986)
-0.277
(-1.066)
5.306
(2.131)
0.0006
(1.532)
-29.503
(-2.596)
-14.040
(-1.294)
13.018
(1.084)
14.515
(-1.549)
11.346
(1.224)
20.321
(1.100)
-1.272
(-0.075)
-9.319
(-0.563)
-20.891
(-0.656)
13.911
(0.832)
54.899
(1.063)
20.251
(1.070)
19.014
(1.115)
38.062
(0.992)
0.269
2.78
136
Water
D to C
-24.058
(-1.185)
-0.172
(-0.033)
-0.182
(-1.188)
2.890
(1.975)
0.0001
(0.564)
-8.629
(-1.292)
-0.575
(-0.090)
16.697
(2.366)
-8.312
(-1.510)
14.134
(2.595)
6.246
(0.578)
-0.279
(-0.028)
0.349
(0.036)
-5.726
(-0.306)
4.466
(0.454)
76.817
(2.530)
1.467
(0.132)
18.181
(1.814)
43.784
(1.942)
0.294
3.14
136
Quality change9
C to B
0.683
(0.043)
-2.209
(-0.531)
-0.155
(-1.286)
0.148
(0.128)
0.0002
(1.39)
0.786
(0.149)
0.160
(0.032)
4.633
(0.833)
-2.763
(-0.637)
3.666
(0.854)
10.166
(1.189)
6.402
(0.818)
2.596
(0.339)
16.615
(1.123)
2.793
(0.361)
55.478
(2.318)
5.098
(0.582)
15.698
(1.987)
-3.945
(-0.222)
0.129
1.12
136
Total:
improvement
only
-17.021
(-0.547)
-4.046
(-0.500)
-0.326
(-1.390)
3.088
(1.378)
0.0003
(0.863)
-6.927
(-0.676)
-1.138
(-0.116)
23.315
(2.153)
-11.371
(-1.346)
19.901
(2.382)
9.072
(0.545)
-0.745
(-0.049)
-3.135
(-0.210)
2.848
(0.099)
2.562
(0.170)
125.627
(2.698)
2.024
0.119
27.557
(1.792)
30.263
(0.875)
0.253
2.55
136
aNumbers in parentheses are asymptotic t-ratios for the null hypothesis of no association.
C-14
-------
Table C-15. Regression Results for Option Value Estimates of Water
Quality Changes—Protest Bids Excluded
Water Quality change8
Independent variables
Intercept
Sex
Aae
"o
Education
I ncome
Direct question
Iterative bidding ($25)
Iterative .bidding ($125)
User
Willing to pay cost
Interviewer 1
Interviewer 2
Interviewer 3
Interviewer 4
Interviewer 5
Interviewer 6
Interviewer 7
Interviewer 8
Interviewer 9
R2
F
Degrees of freedom
D to E (avoid)
-36.611
(-0.890)
20.914
(1 .953)
-0.257
(-0.849)
4.067
(1.394)
0.0005
(1.252)
-30.187
(-2.230)
-16.969
(-1.300)
24.667
(1.843)
-14.859
(-1.333)
19.183
(1.730)
45.060
(2.039)
13.174
(0.636)
-4.031
(-0.200)
28.659
(0.782)
19.815
(0.944)
46.018
(0.955)
44.117
(1.997)
19.804
(0.955)
50.923
(1.493)
0.241
2.93
166
D to C
-17.778
(-0.770)
9.950
(1.655)
-0.274
(-1.611)
2.067
(1.262)
-0.0000
(-0.010)
-7.565
(-0.996)
-1.014
(-0.138)
26.037
(3.467)
-11.852
(-1.894)
18.577
(2.984)
19.717
(1.590)
11.210
(0.964)
7.156
(0.633)
16.378
(0.796)
8.747
(0.743)
39.722
(1.469)
5.264
(0.425)
18.400
(1.581)
29.468
1.539
0.247
3.03
166
C to B
2.781
(0.132)
7.304
(1.329)
-0.236
(-1.521)
-0.321
(-0.214)
0.0001
(0.610)
1.854
(0.267)
-0.711
(-0.106)
15.358
(2.237)
-7.063
(-1.235)
8.014
(1.409)
25.128
(2.216)
17.693
(1.665)
7.027
(0.681)
34.403
(1.829)
5.520
(0.513)
28.591
(1.156)
10.457
(0.923)
17.741
(1.668)
21.089
(1.205)
0.143
1.54
166
Total:
improvement
only
-9.546
(-0.235)
15.986
(1.514)
-0.511
(-1.711)
1.811
(0.630)
0.0001
(0.206)
-4.781
(-0.358)
-2.224
(-0.173)
42.630
(3.231)
-19.465
(-1.771)
28.340
(2.592)
38.220
(1.754)
22.775
(1.115)
8.697
(0.438)
43.904
(1.215)
10.171
(0.492)
62.895
(1.324)
10.920
(0.501)
30.310
(1.483)
42.740
(1.271)
0.212
2.48
166
aNumbers in parentheses are asymptotic t-ratios for the null hypothesis of no association.
C-15
-------
Table C-16. Benefit Estimates from Contingent Ranking Models
Model/estimator Average Range
Payment = 5 Water quality change: boatable to fishable
Final Model
(specification I)
Ordered logit -8.77 -73.77 to 115.82
Ordered normal -9.90 -157.02 to 287.88
II Payment = 50 Water quality change: boatable to fishable
Ordered logit 51.40 48.51 to 55.41
Ordered normal 72.45 49.06 to 97.79
III Payment = 100 Water quality change: boatable to fishable
Ordered logit 49.56 48.31 to 51.70
Ordered normal 69.39 48.90 to 85.94
IV Payment = 175 Water quality change: boatable to fishable
Ordered logit 49.17 48.26 to 50.94
Ordered normal 68.75 48.86 to 83.67
V Payment = 5 Water quality change: boatable to swimmable
Ordered logit -15.78 -132.78 to 208.48
Ordered normal -17.82 -282.64 to 518.18
VI Payment = 50 Water quality change: boatable to swimmable
Ordered logit 92.52 87.31 to 99.74
Ordered normal 130.40 88.30 to 176.02
V|| Payment = 100 Water quality change: boatable to swimmable
Ordered logit 89.21 86.95 to 93.05
Ordered normal 124.90 88.01 to 154.70
VIII Payment = 175 Water quality change: boatable to swimmable
Ordered logit 88.51 86.87 to 91.69
Ordered normal 123.75 87.95 to 150.60
C-16
-------
Table C-17. Estimated Option Values fcr Water Quality Change:
Effects of Instrument and Type of Respondent--
Protest Bids and Outliers Excluded
Type of respondent
_. . Usera Nonuser
Change in
water quality X s n X s n
1. Iterative Bidding Framework, Starting Point = $25
D to E (avoid) 21.43 16.81 14 28.52 34.16 44
D to C 14.64 12.32 14 14.55 15.47 44
C to B. 8.93 11.80 14 6.48 11.13 44
D to BD 23.57 22.65 14 21.02 23.61 44
2. Iterative Bidding Framework, Starting Point = $125
D to E (avoid) 62.33 67.03 15 37.58 50.96 33
D to C 40.33 49.77 15 25.45 44.90 33
C to B. 14.00 33.60 15 11.21 32.60 33
D to BD 54.33 72.60 15 39.24 68.30 33
3. Direct Question Framework
D to E (avoid) 18.21 31.29 14 17.89 34.42 37
D to C 10.50 26.94 14 10.62 20.74 37
C to B. 9.86 27.14 14 8.73 20.97 37
D to BD 22.14 53.73 14 20.30 39.75 37
4. Payment Card
D to E (avoid)
D to C
C to Bb
D to B°
27.73
15.91
5.00
20.91
30.03
21.19
10.00
27.46
11
11
11
11
49.19
20.47
6.63
28.26
72.69
32.27
18.70
44.87
43
43
43
43
aThese results are based on the narrow definition of users.
D to B represents the sum of bids for the improvements in water quality and
for some individuals the payment to move from Level D to Level B directly.
C-17
-------
Table C-18. A Comparison of Contingent Valuation and Travel Cost
Benefit Estimates—Protest Bids and Outliers Excluded*
AWQ = Loss of area AWQ = floatable to flshable AWQ = Boatable to swimmable
Model
Test"
Model
Test"
Model
Test
Independent variable
Intercept
Travel cost benefit
estimate
Qualitative variables
Payment card
Direct question
Iterative bid ($25)
17.482
(1.022)
.450
(1.475)
-34.502
(-2.335)
-27.039
(-2.062)
-28.803
(-1.993)
3.608
35.422
(1.672)
-4.923
(-1.299)
69.510
(2.883)
17.831
(0.850)
-4.740
(-0.201)
-1.708
58.359
(1.669)
-3.186
(-1.076)
109.632
(2.734)
17.421
(0.499)
-11.500
(-0.293)
-1.600
R2
.117
68
2.09 f
(0.09)c
.158
68
2-96.
(0.03)c
.146
68
2.68 ,
(0.04)c
aThe numbers in parentheses below the estimated coefficients are t-ratios for the null hypothesis of no
association.
This column reports the t-ratio for the hypothesis that the coefficient for the travel cost variable was
1.55. The travel cost model measures consumer surplus In 1977 dollars. The contingent valuation experi-
ments were conducted in 1981. Using the consumer price index to adjust the travel cost benefit estimates
to 1981 dollars would require multiplying each estimate by 1.55. Since the estimated regression coefficients
(and standard errors) will correspondingly adjust to reflect this scale change, a test of the null hypothesis
that the coefficient of travel cost was equal to unity is equivalent to a test that is equal to 1.55 when the
travel cost benefit estimates are measured in 1977 dollars and user values estimates (the dependent vari-
able) are in 1981 dollars.
°This number in parentheses below the reported F-statistic Is the level of significance for rejection of the
null hypothesis of no association between the dependent and independent variables.
C-18
-------
APPENDIX D
SURVEY QUESTIONNAIRES
This appendix contains two parts. Part 1 contains the survey question-
naires as administered during the survey of the Monongahela River basin.
Part 2 contains a brief summary of suggestions for improving the questionnaire
for future use in similar surveys.
D-1
-------
PART 1
SURVEY QUESTIONNAIRE AS ADMINISTERED DURING THE
MONONGAHELA RIVER BASIN SURVEY
OMB » 2000-03*1
Appr
-------
A-3 LEAVE CARD 1 IN FRONT OF RESPONDENT. GIVE RESPONDENT CARD 2, "LIST OF
SITES." Here is a list of recreational sites in the area. GIVE RESPON-
DENT CARD 3, "PICTORIAL MAP." And here is a pictorial map showing the
location of these sites. ALLOW RESPONDENT TIME TO LOOK AT BOTH CARDS.
THESE THREE CARDS SHOULD REMAIN IN FRONT OF THE RESPONDENT THROUGHOUT THE
INTERVIEW.
How many times within the past twelve months did you visit any of the
sites listed on this card or any other recreational site near water?
AS SITES ARE MENTIONED, RECORD SITE CODE AND NUMBER OF TIMES THE SITE WAS
VISITED. THEN ASK: Which activities listed on Card 1 did you partici-
pate in at that site during the last 12 months?
CIRCLE THE ACTIVITY NUMBER(S) IN THE COLUMN ACROSS FROM THE SITE(S) MEN-
TIONED.
IF UNLISTED SITES ARE MENTIONED, ENTER SITE NAME ON LINE AND RECORD
NUMBER(S) OF VISITS AND ACTIVITIES.
Siu *••*•
Hoc U>ud
/
Siu
COOM
Mo. ot
Vliits
CANOEING, KAYAKING. ETC.
01
01
01
01
01
01
01
01
01
01
01
01
01
01
01
01
I
02
02
02
02
02
02
01
02
02
02
02
02
02
02
02
02
SAILING
03
03
03
03
03
03
03
03
03
03
03
03
03
03
03
03
s
8
I
04
04
04
04
04
04
04
04
04
04
04
04
04
04
04
04
|
OS
OS
OS
OS
OS
0]
OS
OS
OS
OS
OS
OS
OS
OS
os
OS
SNIMUNG, SUNBATHING
06
06
06
06
06
06
06
06
06
06
06
06
06
06
06
06
I
07
07
07
07
07
07
07
07
07
07
07
07
07
07
07
07
PICNICKING
Ot
Ot
01
01
Ot
Ot
Ot
Ot
01
01
01
OS
Ot
Ot
01
01
BIRD/HIUILIFG OBSI-RV/WB
09
09
09
09
09
09
09
09
09
09
-09
09
09
09
09
09
OHER MAUING/JOGGING
10
10
:o
10
10
10
10
10
10
10
10
10
10
10
10
1C
WITUDIR
11
11
11
11
11
11
11
11
11
11
11
11
11
11
11
11
lURSHBAOC RIDING
12
12
12
12
12
12
12
12
12
12
12
12
12
12
12
12
SNUN1I
13
13
13
13
13
13
13
13
13
13
13
13
13
13
13
13
HIKING OR BACKPACKING
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
I
IS
IS
IS
IS
IS
IS
IS
IS
IS
IS
IS
IS
IS
IS
IS
IS
on tat arm** SPORTS
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
OFF-KOAD DRIVING/RIDINB
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
i
11
It
It
11
It
11
It
u
11
11
It
11
11
It
11
11
SICmSPGING
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
It
D-3
-------
A. RECREATIONAL ACTIVITIES
A-l a. First, do you own or have the use of any kind of boat? CIRCLE
NUMBER.
YES 01 (GO TO A-l.b.)
NO 02 (GO TO A-2)
b. Which of the following describes the boat you use most often? READ
ANSWER CHOICES AND CIRCLE NUMBER.
SAILBOAT 01
INBOARD 02
OUTBOARD 03
CANOE 04
OTHER (SPECIFY) . . . . . .05
A-2 The next few questions we would like to ask deal with outdoor recrea-
tional activities which people take part in near lakes and rivers in this
area; that is, the activities shown on this card. GIVE RESPONDENT CARD
1, "ACTIVITY CARD". Please look carefully over the list of activities,
keeping in nind that all the activities listed refer to activities near
lakes or rivers. ALLOW RESPONDENT TIME TO LOOK AT THE LIST.
Within the past 12 months, that is since last November, did you take part
in any of the activities listed? CIRCLE NUMBER.
YES 01 (GO TO A-3)
NO 02 (GO TO B-l)
(23)
(24)
D-4
-------
B. BENEFITS MEASURES
B-l The next group of questions is about the quality of water in the Mononga-
hela River. Congress passed water pollution control laws in 1972 and in
1977 to improve the nation's water quality. The states of Pennsylvania
and West Virginia have also been involved in water quality improvement
programs of their own. These programs have resulted in. cleaner rivers
that are better places for fishing, boating, and other outdoor activities
which people take part in near water. We all pay for these water quality
improvement programs both as taxpayers and as consumers.
In this study we are concerned with the water quality of only the Monon-
gahela River. Keep in mind that people take part in all of the activi-
ties on Card 1 both on and near the water.
Generally, the better the water quality, the better suited the water is
for recreational activities and the more likely people' will take part in
outdoor recreational activities on or near the water. Here is a picture
of a ladder that shows various levels of water quality. GIVE RESPONDENT
CARD 4, "WATER QUALITY LADDER".
The top of the ladder stands for the best possible quality of water. The
bottom of the ladder stands for the worst possible water quality. On the
ladder you can see the different levels of the quality of the water. For
example: (POINT TO EACH LEVEL — Z, D, C, B, A — AS YOU READ THE STATE-
MENTS BELOW.)
Level "E" (POINTING) is so polluted that it has oil, raw sewage and
other things like trash in it; it has no plant or animal life and
smells bad.
Water at level "D" .is okay for boating but not fishing or swimming.
Level "C" shows where the water is clean enough so that game fish
like bass can live in it.
Level "B" shows where the water is clean enough so that people can
swim in it safely.
And at level "A", the quality of the water is so good that it would
be possible to drink directly from it if you wanted to.
Now, think about the water quality of the Monongahela River on the
whole. In terms of this scale from zero to ten, how would you rate
the water quality of the Monongahela River at the present time?
POINT TO THE ZERO-TO-TEN SCALE ON THE LADDER AND CIRCLE NUMBER.
00 01 02 03 04 05 06 07 08 09 10 (GO TO B-l.b.)
DON'T KNOW 11 (GO TO B-2)
Card S
1-22
Dup.
(S3-S4)
b. Is your rating for a particular site on the river? CIRCLE NUMBER.
*ES 01 (GO TO B-l.c.)
NO 02 (GO TO B-2)
(25)
D-5
-------
c. On the map, please show me which river site your rating applies to.
Site Code:
IF NOT ON LIST OF RECREATIONAL SITES, SPECIFY:
(26-27)
B-2 Another important purpose of this study is to learn how much the quality
of water of the Honongahela River is worth to the people who live in the
river basin. In answering this question, there are three ways of think-
ing about water quality that might influence your decision. GIVE RESPON-
DENT CARD 5, "VALUE CARD". The three ways are shown on this card.
a. One, you might think about how ouch water quality is worth to you
because you use the river for recreation. POINT TO PART I OF VALUE
CARD AND GIVE RESPONDENT TIME TO READ THAT PART.
How important a factor is your actual use of the river in making a
decision about how much clean water is worth to you? CIRCLE NUMBER.
VERT IMPORTANT 01
SOMEWHAT IMPORTANT .... 02
NEITHER IMPORTANT NOR
UNIMPORTANT 03
NOT VERY IMPORTANT .... 04
NOT IMPORTANT AT ALL ... 05
(28-29)
b. Another way you might *M"fc about how much clean water is worth to
you is that it is worth something to you to know that a clean water
river is being maintained for your use if you should decide, in the
futUre, that you want to use it. POINT TO PART II OF VALUE CARD AND
GIVE RESPONDENT TIME TO READ THAT PART. For example, you might buy
an advance ticket for the Steelers or Pirates just to be able to go
to a home game if you later decide you want to go. Likewise, you
might pay some amount each year to have a clean water river avail-
able to use if you should decide to use it.
In deciding how much clean water is worth to you, how important a
factor is knowing that a clean water river is being maintained for
your use, if you should decide to use it? CIRCLE NUMBER.
VERY IMPORTANT 01
SOMEWHAT IMPORTANT .... 02
NEITHER IMPORTANT NOR
UNIMPORTANT 03
NOT VERY IMPORTANT .... 04
NOT IMPORTANT AT ALL ... 05
(30-31)
D-6
-------
A third thing you might think about in deciding how much clean water
is worth to you is the satisfaction of knowing that a clean water
river is there. POINT TO PART III OF VALUE CARD AM) GIVE RESPONDENT
TIME TO READ THAT PART. lor example, you might be willing to pay
something to maintain a public park even though you know you won't
use it. The same thing could be true for clean water in the Monon-
gahela; that is, you might pay something just for the satisfaction
of knowing that it is clean and that others can use it.
In deciding how much clean water is worth to you, how important is
knowing that a clean water river is being maintained? CIRCLE
NUMBER.
VERY IMPORTANT ...... 01
SOMEWHAT IMPORTANT .... 02
NEITHER IMPORTANT NOR
UNIMPORTANT 03
NOT VERY IMPORTANT .... 04
NOT IMPORTANT AT ALL ... OS
INTRODUCTION TO QUESTION B-3
Now, we would like for you to think about the relationship between im-
proving the quality of water in the Monongahela River and what we all have to
pay each year as taxpayers and as consumers. We all pay directly through our
tax dollars each year for cleaning up all rivers. We also pay indirectly each
year through higher prices for the products we buy because it costs companies
money to clean up water they use in making their products. Thus, each year,
we are paying directly and indirectly for improvements in the water quality of
the Monongahela River.
I want to ask you a few questions about what amount of money you would be
willing to pay each year for different levels of water quality in the Mononga-
hela River. Please keep in mind that the amounts you would pay each year
would be paid in the form of taxes or in the form of higher prices for the
products that companies sell.
We are talking about different levels of water quality for only the
Monongahela River, with water quality at other sites on Card 2 staying the
same as it is now.
I also want you to keep in mind the recreational activities that you now
do and that you might do in the future on the Monongahela River or at other
sites. That is, keep in mind the first two parts of the value card. (POINT
TO THE VALUE CARD, CARD 5.) Your actual use or possible use can involve
activities in the water or near the water, or both, as we talked about earlier.
We know that for the Monongahela River as a whole the current water
quality is at level "D", but that it may vary at different points along the
river. At level "D" it is clean enough for boating, but not clean enough for
catching game fish or for swimming.
(32-33)
HAVE REMINDER CARD READY.
ASKED.
RECORD DOLLAR AMOUNTS GIVEN FOR EACH PART
D-7
-------
B-3 a. This payment card shows different yearly amounts people might be
willing to pay for different levels of water quality. HAND RESPON-
DENT CARD 6, "PAYMENT CARD," AND ALLOW RESPONDENT TIME TO LOOK AT
IT.
What is the most it is worth to you (and your family) on a yearly
basis to keep the water quality in the Monongahela River from
slipping back from level "D" to level "E", where it is not even
clean enough for boating? Please pick any amount on the card, any
amount in between, or any other amount you think is appropriate.
DOLLARS
/IF ANY AMOUNT, GO TO B-3.b.;\
\IF ZERO DOLLARS, ASK -| . I
T
(34-36)
Would it be worth something to you (and your family) to raise the
water quality level from level "D" to a higher level? CIRCLE
NUMBER.
YES
NO
01
02
(GO TO B-3.b.)
(GO TO B-3.e.)
(In addition to the amount you just told me,) What is the most that
.you would be willing to pay each year in higher taxes and prices for
products that companies sell to raise the water quality from level
"D" to level "C", where game fish can live in it and it is improved
for other activities?
DOLLARS
DOLLARS
» T0 B-3
DOLLARS> co T0 B
• c.; \
-3.dJ
(37-39)
How much more than (READ AMOUNT FROM b.) would you be willing to pay
each year in higher taxes and prices for products that companies
sell to raise the water quality from level "C" to level "B", where
it is clean enough for swimming and it is improved for other activi-
ties?
DOLLARS (GO TO B-4)
(40-42)
D-8
-------
What is the most that you would be willing to pay each year in
higher taxes and prices for products that companies sell to raise
the water quality from level "D" to level "B", where it is cleaa
enough for swimming and it is improved for other activities?
DOLLARS
:IF ANY AMOUNT IN a., GO TO B-4; >
IF ZERO DOLLARS IN a. AND:
• ANT AMOUNT IN d., GO TO B-4.d.;
. ZERO DOLLARS IN d., GO TO B-3.eJ
(43-45)
We have found in studies of this type that people have a lot of
different reasons for answering as they do. Some people felt they
did not have enough information to give a dollar amount, some did
not want to put dollar values on environmental quality, and some
objected to the way the question was presented. Others gave a zero
dollar amount because that was what it was worth to them.
Which of these reasons best describes why you answered the way you
did? REPEAT REASONS IF NECESSARY AND CIRCLE NUMBER.
NOT ENOUGH INFORMATION
01
DID HOT WANT TO PLACE
DOLLAR VALUE 02
OBJECTED TO WAY QUESTION >• (GO TO B-6)
WAS PRESENTED 03
THAT IS WHAT IT IS WORTH . 04
OTHER (SPECIFY) ...... 05 J
(4S-47J
D-9
-------
B-4 REFER TO REMINDER CARD.
AMOUNTS ON CARD.
DO NOT ASK QUESTIONS CORRESPONDING TO ZERO
b.
In answering the next question(s), keep in mind your actual and
possible future use of the Monongahela. You told me in the last
section that it was worth $ (AMOUNT) to keep the water quality from
slipping from level "D" to level "E". How much of this amount was
based on your actual use of the river?
Tou (also) told me that you would be willing to pay $(AMOUNT) to
raise the water quality from level "D" to level "C". POINT TO
LEVELS "D" AND "C". How much of this amount was due to your actual
use of the river?
(48-50)
(S1-S3)
Tou (also) told me that you would be willing to- pay $ (AMOUNT) to
raise the water quality from level- "C" to level "B". POINT TO
LEVELS "C" AND "B". How much of this amount was due to your actual
use of the river?
$ (GO TO B-5)
(S4-S6)
d. Tou told me in the last question that you would be willing to pay
$ (AMOUNT) to raise the water quality from level "D" to level "B".
POINT TO LEVELS "D" AND "B". How much of this amount was due to
your actual use of the river.
(S7-S9)
B-5 REFER TO REMINDER CARD. You have said that you would be willing to pay
$(AMOUNT) to keep the level of water quality from slipping from level "D"
to level "E" and you said that you would be willing to pay $(b. PLUS c..
OR d.) to raise the level from level "D". This is a total of (READ TOTAL
$ AMOUNT).
Let's think about another way that the quality of water in the Mononga-
hela River could affect your recreation on or near water. I would like
you to think about how the river being closed for certain activities for
different periods of time would change the (READ TOTAL S AMOUNT) you
would be willing to pay per year. Suppose the government is considering
relaxing the water pollution control laws, but not totally removing them.
This would mean that the overall quality of the water in the Monongahela
River would decrease to level "E" where it would be closed some weekends
for activities on or in the water like boating, fishing and swimming and
you would not know it was closed until the day you wanted to go. The
area, however, would remain open all weekends for activities near the
water, like jogging or hiking or picnicking.
D-10
-------
If the water pollution laws were relaxed to the point that the water
quality would decrease to level "E" and the area would be closed 1/4
of the weekends of the year for activities on or in the water but
would remain open for activities near the water, how much would you
change this (READ TOTAL $ AMOUNT) to keep the area open all weekends
for all activities?
$ DOLLAR CHARGE
(eo-sz)
b. If the area would be closed for activities on or in the water 1/2 of
the weekends, how much would you change this (READ TOTAL $ AMOUNT)
to keep the area open all weekends for all activities?
DOLLAR CHANGE
(SZ-6S)
e. If the area would be closed for activities on or in the water 3/4 of
the weekends, how ouch would you change this (READ TOTAL $ AMOUNT)
to keep the area open all weekends for all activities?
DOLLAR CHANGE
(66-ea)
B-6 a. If the water quality in the Monongahela River were improved from
level "D" to level "B", where it is clean enough for swioming and it
is improved for other activities, how would this affect your annual
use or future use of sites along the river? CIRCLE NUMBER.
INCREASE USE BT MORE THAN 5 VISITS PER TEAR ... 01
INCREASE USE BT 1 TO 5 VISITS PER TEAR 02
NO CHANGE IN USE 03
DECREASE USE ALONG THE MONONGAHELA RIVER 04
DON'T KNOW. 05
(69-70)
b. How would this change from "D" to "B" in the Monongahela River
affect your annual use or future use of other recreational sites
near water, but not along the Monongahela River? CIRCLE NUMBER.
DECREASE USE VI MORE THAN S VISITS PER TEAR ... 01
DECREASE USE BY 1 TO 5 VISITS PER TEAR 02
NO CHANGE IN USE 03
INCREASE USE Q4
DON'T KNOW 05
(71-72)
D-11
-------
B-7 Up to now we have talked about water quality based on your use and pos-
sible future use of the Monongahela River. Let's again think about the
third part of the value card. That is, it is worth something just to
know a river with clean water is there without actually using it or
planning to use it. We want you to think only in terms of this satis-
faction which excludes any use by you of the river. With this in mind,
suppose the government were to remove the water pollution laws entirely.
This would mean lower taxes and would allow companies to produce their
products at lower prices. But, it would also mean that during most of
the rest of your lifetime the Monongahela River would be at level "E" and
would not be usable for recreational activities. The change could be
reversed in your lifetime but it would cost a great deal of money.
a. What is the most that you (and your family) would be willing to pay
each year in the form of higher taxes and prices for the goods you
buy for keeping the river at level "D" where it is okay for boating,
even if you would never use the river?
s IH ANY AMOUNT, GO TO B-7.b.;\
* VIF ZERO DOLLARS, GO TO B-8) /
b.
Suppose the change could not be reversed for an even longer period
of time than your lifetime. How much more than (READ AMOUNT FROM a.)
would you (and your family) be willing to pay per year to keep the
river at level "D", even if you would never use the river?
B-S
GIVE RESPONDENT THE FOUR CARDS FROM THE CARD SET 7- I would now like you
to look at these cards which show different combinations of levels of
water quality and amounts in higher taxes and prices it would cost every
family each year to have the indicated water quality levels.
a. First, I would like you to rank the combinations of water quality
levels and amounts you might be willing to pay to obtain those
levels in order from the card, or combination, that you most prefer
to the one you least prefer. I would like you to do this based only
on your use and possible use in the future of the Monongahela River.
That is, keeping in mind only Farts I and II of the value card.
POINT TO VALUE CARD - PARTS I AND II. RECORD RANKING OF CARDS BY
CIRCLED WATER QUALITY LEVELS AND DOLLAR AMOUNTS.
RANKING
Most Preferred
2nd
3rd
Least Preferred
WATER
QUALITY
LEVEL
$ AMOUNT
$
$
$
S
(73-7S)
(76-78J
Col.
80 - S
Card e
1-22
Dup.
(24-2?)
(28-31)
(32-351
(36-39)
D-12
-------
Now, I would like you to repeat this procedure but assume this tine
that you will not use the river now or in the future. That is,
think about only Part III of the value card. POINT TO VALUE CARD -
PART HI. RECORD RANKING OF CARDS BY CIRCLED WATER QUALITY LEVELS
AND DOLLAR AMOUNTS.
RANKING
Host Preferred
2nd
3rd
Least Preferred
WATER
QUALITY
LEVEL
$ AMOUNT
$
$
$
$
(40-43)
(44-47)
(48-51)
(52-65)
Col.
80 " S
D-13
-------
C. BACKGROUND DATA
I have a few more questions that will help out research staff analyze the
results of the study properly.
C-l How long have you lived in the Monongahela River basin area? CIRCLE
NUMBER.
LESS THAN 1 YEAR
1 TEAR OR LONGER BUT
LESS THAN 3 YEARS . . .
3 YEARS OR LONGER BUT
LESS THAN 5 TEARS . . .
5 TEARS OR LONGER ....
01
. 02
. 03
. 04
C-2
Now I am going to read some phrases that describe, different kinds of
interests people have. As I read each one, please tell me how ouch the
phrase is like you; that is, a lot like you, somewhat like you, a little
like you, or not at all like you. CIRCLE ONE NUMBER ON EACH LIKE.
REPEAT ANSWER CHOICES AS NECESSARY.
SOME A NOT NO
A LOT WHAT LITTLE AT ALL OPINION
a. AN OUTDOORS PERSON .
b. AN ENVIRONMENTALIST.
01 . . 02 .
01 . . 02 .
03 . . . 04 .
03 ... 04 .
05
05
c. SOMEONE WHO IS AGAINST
NUCLEAR POWER FOR
ELECTRIC PLANTS 01 . . 02
d. SOMEONE WHO IS CONCERNED
ABOUT WATER POLLUTION. ... 01 .. 02
e. SOMEONE WHO IS WILLING TO
PAT THE COST REQUIRED TO
CONTROL WATER POLLUTION. . . 01 . . 02
03 ... 04 ... 05
03 ... 04 ... 05
03 ... 04 ... 05
C-3 Which of the following best describes your present status? READ CHOICES
AS NECESSARY AND CIRCLE NUMBER.
EMPLOYED FULL-TIME 01
EMPLOTED PART-TIME 02
RETIRED 03
NOT EMPLOTED 04
A HOUSEWIFE 05
A STUDENT 06
OTHER (SPECIFT) 07
(GO TO C-5)
(GO TO C-4)
Cord ?
1-22
Dup.
(24-2S)
(26-Z?)
(28-29)
(30-31)
(32-33)
(34-35)
(36-3?)
D-14
-------
C-4 Have you ever been employed?
YES
HO
CIRCLE NUMBER.
. . 01 (GO TO C-5)
. . 02 (GO TO C-6)
(38-39)
C-5
a. What kind of work (do/did) you do; that is,
called?
what (is/was) your job
(40-42)
b.
e.
What (do/did) you actually do in that job?
your main duties and responsibilities?
What (are/were) soae of
What kind of an organization (do/did) you work for? (PROBE: What
do they nake, what do they do?) BE SURE TO NOTE IF RESPONDENT IS AN
EMPLOYEE OF GOVERNMENT AT ANY LEVEL, INCLUDING THE SCHOOL SYSTEM.
(43-4S)
(46-48)
d. How many hours (do/did) you work at your job in a usual week?
NUMBER OF HOURS WORKED IN A WEEK
(49-50)
C-6 What was the last grade of regular school that you completed — not
counting specialized schools like secretarial, art, or trade schools?
CIRCLE NUMBER.
NO SCHOOL 01
GRADE SCHOOL (1-8) 02
SOME HIGH SCHOOL (9-11) . . 03
HIGH SCHOOL GRADUATE (12) . 04
SOME COLLEGE (13-15). ... 05
\ COLLEGE GRADUATE (16) ... 06
POST GRADUATE (17+) .... 07
NO RESPONSE/REFUSED .... 08
(51-52)
D-15
-------
C-7
ASK ONLY IT NOT OBVIOUS. How would you describe your racial or ethnic
background? READ CHOICES AND CIRCLE NUMBER.
WHITE OR CAUCASIAN 01
BLACK OR NEGRO 02
OTHER (SPECFIY) 03
C-8 Here is a list of income categories. HAND RESPONDENT CARD 8. Would you
call off the code number of the category that best describes the combined
income that you (aad all other members of your family) received during
1980. Please be sure to include wages and salaries, or net income from
your business, and pensions, dividends, interest, and any other income.
CIRCLE NUMBER.
UNDER $5,000 01
$5,000 - $9,999 02
$10,000 - $14,999 03
$15,000 - $19,999 04
$20,000 - $24,999 05
$25,000 - $29,999 06
$30,000 - $34,999 07
$35,000 - $39,999 08
$40,000 - $44,999 09
$45,000 - $49,999 10
$50,000 AND OVER 11
NOT SURE/REFUSED 12
C-9 There is a possibility that my supervisor would like to call you to
verify your participation in this study. What is the telephone number
where you can be reached?
(S3-S4)
(SS-S6)
Col.
80-7
TELEPHONE NUMBER: (.
Thank you for participating in this study.
INTERVIEW STOP TIME:
AM / PM
D-16
-------
OMB I 2000-0381
Approval Expim: 9/20/S2
ESTIMATING BENEFITS OF WATER QUALITY
QUESTIONNAIRE
Form No. 02
(1)
A. Study No.
(2-8)
Housing
Unit No.
I. IDENTIFICATION INFORMATION
B. PSU/Segment No. | | "
Interviewer i—r
I-D
(15-17)
- Sample Individual
Roster Line No.
(8-13)
ID Ho.
(Skip)
F. Questionnaire Version
II. INTRODUCTION
B
(22)
IF THE ENUMERATION RESPONDENT IS ALSO THE SELECTED SAMPLE INDIVIDUAL,
CONTINUE TOUR INTRODUCTION TO THE STUDY BY READING THE SECOND PARAGRAPH BELOW.
IF THE SAMPLE INDIVIDUAL IS SOMEONE OTHER THAN THE ENUMERATION RESPONDENT,
READ THE ENTIRE INTRODUCTION BELOW.
Hello, I'll (NAME) from the Research Triangle Institute in North Carolina.
We are doing a study for a government agency to study levels of water quality
and some outdoor recreational activities people take part in both near and on
ponds, lakes, streams aad rivers in the Honoagahela River Basin. You have
been randomly selected to participate in the study.
Your participation is entirely voluntary and you may refuse to answer
any questions. Because only a small number of people are being selected for
the study, the participation of each person selected is extremely important.
Most of the questions have to do with your attitudes and opinions and there
are no xight or wrong answers. The information which you provide will be kept
strictly confidential and will be used only for overall statistical results.
If you would like, we will send you a summary of the results of the study.
CHECK APPROPRIATE BOX BELOW AND IF "YES" PRINT RESPONDENT'S MAILING
ADDRESS.
RESULTS REQUESTED: YES Q NO Q
Mailing
Address
Number/Street/RFD
Apt. No.
City/State
ZIP
INTERVIEW START TIME:
AM / PM
D-17
-------
B-3 a. What is the most it is worth to you (and your family) on a yearly
basis to keep the water quality in the Honongahela River from
slipping back from'level "D" to level "E", where it is not even
i?
S DOLLARS (jfSolS
T, GO TO B-3.b.;\
ARS, ASK -i . /
^ Would it be worth something to you (and you
water quality level from level "D" to a
NUMBER.
YEs oi (GO TO B-3.b.
HO 02 (GO TO B-3.C.
r family) to raise the
higher level? CIRCLE
)
)
(34-36)
b.
(In addition to the amount you just told me,) What is the most that
you would be willing to pay each year in higher taxes and prices for •
products that companies sell to raise the water quality from level
"D" to level "C", where game fish can live in it and it is improved
for other activities?
DOLLARS
/IF ANY AMOUNT, GO TO B-3.C.; \
\IF ZERO DOLLARS, GO TO B-3.d./
c.
How much more than (READ AMOUNT FROM b.) would you be willing to pay
each year in higher taxes and prices for products that companies
sell to raise the water quality from level "C" to level "B", where
it is clean enough for swimming and it is improved for other activi-
ties?
DOLLARS (GO TO B-4)
(37-39)
(40-42)
D-18
-------
d. What is the most that you would be willing to pay each year in
higher taxes and prices for products that companies sell to raise
the water quality from level "D" to level "B", where it is clean
enough for swimming and it is improved for other activities?
DOLLARS
;IF ANY AMOUNT HI a., 00 TO B-4; X
IF ZERO DOLLARS IN a. AND:
• AN? AMOUNT IN d., CO TO B-4.d.;
• ZERO DOLLARS IN d., GO TO B-3.e./
(43-45)
We have found in studies of this type that people have a lot of
different reasons for answering as they do. Some people felt they
did not have enough information to give a dollar amount, some did
not want to put dollar values on environmental quality, and some
objected to the way the question was presented. Others gave a zero
dollar amount because that was what it was worth to them.
Which of these reasons best describes why you answered the way you
did? REPEAT REASONS IF NECESSARY AND CIRCLE NUMBER.
NOT ENOUGH INFORMATION
01
DID NOT WANT TO PLACE
DOLLAR VALUE ...... 02
OBJECTED TO WAY QUESTION
WAS PRESENTED ...... 03
THAT IS WHAT IT IS WORTH . 04
OTHER (SPECIFY) . ..... 05 J
(46-47)
(GO TO B-6)
D-19
-------
OMB * 1000-0381
Approval Expire*: 9/20/82
A. Study No.
ESTIMATING BENEFITS OF WATER QUALITY
QUESTIONNAIRE
Form No. 02
(1)
I. IDENTIFICATION INFORMATION
_] B. PSU/Segnent No. || ~
(2-6)
(8-13}
Housing
Unit No.
D.
(15-17)
Interviewer
ID No.
_ Sample Individual
Roster Line No.
(19-20)
(Skip)
F. Questionnaire Version
II. INTRODUCTION
C
(22)
IF THE ENUMERATION RESPONDENT IS ALSO THE SELECTED SAMPLE INDIVIDUAL,
CONTINUE TOUR INTRODUCTION TO THE STUDY BY READING THE SECOND PARAGRAPH BELOW.
IF THE SAMPLE INDIVIDUAL IS SOMEONE OTHER THAN THE ENUMERATION RESPONDENT,
READ THE ENTIRE INTRODUCTION BELOW.
Hello, I'a (NAME) from the Research Triangle Institute in North Carolina.
We are doing a study for a government agency to study levels of water quality
and some outdoor recreational activities people take part in both near and on
ponds, lakes, streams and rivers in the Monongahela River Basin. You have
been randomly selected to participate in the study.
Your participation is entirely voluntary and you may refuse to answer
any questions. Because only a small number of people are being selected for
the study, the participation of each person selected is extremely important.
Most of the questions have to do with your attitudes and opinions and there
are no right or wrong answers. The information which you provide will be kept
strictly confidential and will be used only for overall statistical results.
If you would like, we will send you a summary of the results of the study.
CHECK APPROPRIATE BOX BELOW AND IF "YES" PRINT RESPONDENT'S MAILING
ADDRESS. _. _
RESULTS REQUESTED: YES Q NO Q
Mailing
Address
Number/Street/RFD
Apt. No.
City/State
INTERVIEW START TIME:
ZIP
AM / PM
D-20
-------
B-3 a. To you (and your family), would it be worth $25 each year in higher
taxes and prices for products that companies sell to keep the water
quality in the Monongahela River from slipping back from level "D"
to level "E"? CIRCLE NUMBER.
YES
HO
01
02
IF YES, INCREASE THE DOLLAR AMOUNT IN
$5 INCREMENTS UNTIL A "NO" ANSWER IS
GIVEN. E.G., "Would it be worth $30
each year to keep water quality from
slipping from level 'D' to level
•E1?" ETC. WHEN A "NO" ANSWER IS
GIVEN, RECORD DOLLAR AMOUNT OF LAST
"TES" ANSWER.
1
IF NO, DECREASE THE DOLLAR AMOUNT IN
$5 INCREMENTS UNTIL A "YES" ANSWER IS
GIVEN. E.G., "Would it be worth $20
each year to keep water quality—from
slipping from level 'D' to level '£'?"
ETC. WHEN A "YES" ANSWER IS GIVEN,
RECORD DOLLAR AMOUNT.
DOLLARS
/IF
IF
\AS!
AKY AMOUNT, GO TO B-3.b.; \
ZERO DOLLARS IS FINAL AMOUNT,]
ASK —i /
(34-3S)
1
Would it be worth something to you (and your family) to raise the
water quality level from level "D" to a higher level? CIRCLE
NUMBER.
YES 01 (GO TO B-3.b.)
NO . 02 (GO TO B-3.e.)
D-21
-------
b.
Would you (and your family) be willing to pay (an additional) $25
each year in higher taxes and prices for products that companies
sell to raise the water quality from level "D" to level "C", where
game fish can live in it and it is improved for other activities?
CIRCLE NUMBER.
YES
NO
01
02
IF YES, INCREASE THE DOLLAR AMOUNT IN
S3 INCREMENTS UNTIL A "NO" ANSWER IS
GIVEN. E.G., "Would you be willing
to pay $30 (acre) each year to raise
the water quality from level 'D' to
level 'C'?" ETC. WHEN A "NO" ANSWER
IS GIVEN, RECORD DOLLAR AMOUNT OF
LAST "YES" ANSWER.
IF NO, DECREASE THE DOLLAR AMOUNT IN
$5 INCREMENTS UNTIL A "YES" ANSWER IS
GIVEN. E.G., "Would you be willing to
pay $20 (more) each year to raise the
water quality from level 'D' to level
'C'?" ETC. WHEN A "YES" ANSWER IS
GIVEN, RECORD DOLLAR AMOUNT.
/IF ANY AMOUNT, GO TO B-3.C.;
DOLLARS I IF ZERO DOLLARS IS FINAL AMOUNT,
\GO TO B-3.d.
(37-39)
c. Would you (and your family) be willing to pay an additional $25 each
year in higher taxes and prices for products that companies sell to
raise the water quality from level "C" to level "B", where you can
swim in it and it is improved for other activities? CIRCLE NUMBER.
YES
NO
01
02
IF YES, INCREASE THE DOLLAR AMOUNT IN
$5 INCREMENTS UNTIL A "NO" ANSWER IS
GIVEN. E.G., "Would you be willing
to pay $30 more each year to raise
the water quality from level 'C' to
level 'B'?" ETC. WHEN A "NO" ANSWER
IS GIVEN, RECORD DOLLAR AMOUNT OF
LAST "YES" ANSWER.
1
IF NO, DECREASE THE DOLLAR AMOUNT IN
$5 INCREMENTS UNTIL A "YES" ANSWER IS
GIVEN. E.G., "Would you be willing to
pay $20 more each year to raise the
water quality from level 'C' to level
'B'?" ETC. WHEN A "YES" ANSWER IS
GIVEN, RECORD DOLLAR AMOUNT.
DOLLARS (GO TO B-4)
(40-42)
D-22
-------
Would you (and your family) be willing Co pay $25 each year in
higher taxes and prices for products that companies sell to raise
the water quality from level "D" to level "B", where you can swim in
it and it is improved for other activities? CIRCLE NUMBER.
YES
01
02
IF TES, INCREASE THE DOLLAR AMOUNT IN
$5 INCREMENTS UNTIL A "NO" ANSWER IS
GIVEN. E.G., "Would you be willing
to pay $30 each year to raise the
water quality from level 'D' to level
•B'?" ETC. WHEN A "NO" ANSWER IS
GIVEN, RECORD DOLLAR AMOUNT OF LAST
"TES" ANSWER.
1
IF NO, DECREASE THE DOLLAR AMOUNT IN
$5 INCREMENTS UNTIL A "TES" ANSWER IS
GIVEN. E.G., "Would you be willing to
pay $20 each year to raise the water
quality from level 'D' to Level 'B1?'1
ETC. WHEN A "TES" ANSWER IS GIVEN,
RECORD DOLLAR AMOUNT.
DOLLARS
;IF ANT AMOUNT IN a., GO TO B-4; \
IF ZERO DOLLARS IN a. AND:
• ANT AMOUNT IN d., GO TO B-4.d.; ,
• ZERO DOLLARS IN d., GO TO B-3-e.,/
(43-45)
e. We have found in studies of this type that people have a lot of
different reasons for answering as they do. Some people felt they
did not have enough information to give a dollar amount, some did
not want to put dollar values on environmental quality, and some
objected to the way the question was presented. Others gave a zero
dollar amount because that was what it was worth to them.
Which of these reasons best describes why you answered the way you
did? REPEAT REASONS IF NECESSARY AND CIRCLE NUMBER.
NOT ENOUGH INFORMATION
01
DID NOT WANT TO PLACE
DOLLAR VALUE 02
OBJECTED TO WAT QUESTION >>(GO TO B-6)
WAS PRESENTED 03
THAT IS WHAT IT IS WORTH . 04
OTHER (SPECIFT) 05 -
(46-47)
D-23
-------
am i 2000-03)1
Approval Expires: 9/20/42
A. Study No.
_ Housing |-
Unit Ho. |_
ESTIMATING BENEFITS 07 WATER QUALITY
QUESTIONNAIRE
Form No. 02
(1)
I. IDENTIFICATION INFORMATION
~ I I B. PSU/Segment No. | | -
Interviewer i—r
(2-6)
(8-1!)
(15-17)
Sample Individual
' Roster Line No.
D.
ID No.
(19-20)
(Skip)
F. Questionnaire Version
II. INTRODUCTION
D
C22J
IF THE ENUMERATION RESPONDENT IS ALSO THE SELECTED SAMPLE INDIVIDUAL,
CONTINUE YOUR INTRODUCTION TO THE STUDY BY READING THE SECOND PARAGRAPH BELOW.
IF THE SAMPLE INDIVIDUAL IS SOMEONE OTHER THAN THE ENUMERATION RESPONDENT,
READ THE ENTIRE INTRODUCTION BELOW.
Hello, I'm (NAME) from the Research Triangle Institute in North Carolina.
We are doing a study for a government agency to study levels of water quality
and some outdoor recreational activities people take part in both near and on
ponds, lakes, streams and rivers in the Monongahela River Basin. You have
been randomly selected to participate in the study.
Your participation is entirely voluntary and you may refuse to answer
any questions. Because only a small number of people are being selected for
the study, the participation of each person selected is extremely important.
Most of the questions have to do with your attitudes and opinions and there
are no right or wrong answers. The information which you provide will be kept
strictly confidential and will be used only for overall statistical results.
If you would like, we will send you a summary of the results of the study.
CHECK APPROPRIATE BOX BELOW AND IF "YES" PRINT RESPONDENT'S MAILING
ADDRESS. _
RESULTS REQUESTED: YES Q HO Q
Mailing
Address
Number/Street/RFD
Apt. No.
City/State
INTERVIEW START TIME:
ZIP
AM / PM
D-24
-------
B-3 a. To you (and your family), would it be worth $125 each year in higher
taxes and prices for products that companies sell to keep the water
quality in the Monongahela River from slipping back from level "D"
to level "E"? CIRCLE KUMBER.
YES
NO
01
02
IF TES, INCREASE THE DOLLAR AMOUNT IN
$10 INCREMENTS UNTIL A "NO" ANSWER IS
GIVEN. E.G., "Would it be worth $135
each year to keep water quality from
slipping from level 'D' to level
'E'?" ETC. WHEN A "NO" ANSWER IS
GIVEN, RECORD DOLLAR AMOUNT OF LAST
"TES" ANSWER.
1
IF NO, DECREASE THE DOLLAR AMOUNT IN
$10 INCREMENTS UNTIL A "TES" ANSWER IS
GIVEN. E.G., "Would it be worth $115
each year to keep water quality from
slipping from level 'D1 to level 'E'?"
ETC. WHEN A "TES" ANSWER IS GIVEN,
RECORD DOLLAR AMOUNT.
DOLLARS
'II ANT AMOUNT, GO TO B-3.b.;
IF ZERO DOLLARS IS FINAL AMOUNT,
1
.)
(34-36)
Would it be worth something to you (and your family) to raise the
water quality level from level "D" to a higher level? CIRCLE
NUMBER.
TES 01 (GO TO B-3.b.)
NO ......... 02 (GO TO B-3.e.)
D-25
-------
b. Would you (and your family) be willing to pay (an additional) $125
each year in higher taxes and prices for products that companies
sell to raise the water quality from level "D" to level "C", where
game fish can live in it and it is improved for other activities?
CIRCLE KUMBER. /
YES
01
02
IF YES, INCREASE THE DOLLAR AMOUNT IN
$10 INCREMENTS UNTIL A "NO" ANSWER IS
GIVEN. E.G., "Would you be willing
to pay $135 (more) each year to raise
the water quality from level 'D' to
level 'C'?" ETC. WHEN A "NO" ANSWER
IS GIVEN, RECORD DOLLAR AMOUNT OF
LAST "YES" ANSWER.
IF NO, DECREASE THE DOLLAR AMOUNT IN
$10 INCREMENTS UNTIL A "YES" ANSWER IS
GIVEN. E.G., "Would you be willing to
pay $115 (more) each year to raise the
water quality from level 'D' to level
'C'?" ETC. WHEN A "YES" ANSWER IS
GIVEN, RECORD DOLLAR AMOUNT.
(IF ANY AMOUNT, GO TO B-3.C.; \
IF ZERO DOLLARS IS FINAL AMOUNT,) (37-39)
GO TO B-3.d. /
Would you (and your family) be willing to pay an additional $125
each year in higher taxes and prices for products that companies
sell to raise the water quality from level "C" to level "B", where
you can swim in it and it is improved for other activities? CIRCLE
NUMBER.
YES
NO
01
02
IF YES, INCREASE THE DOLLAR AMOUNT IN
$10 INCREMENTS UNTIL A "NO" ANSWER IS
GIVEN. E.G., "Would you be willing
to pay $135 more each year to raise
the water quality from level 'C' to
level 'B'?" ETC. WHEN A "NO" ANSWER
IS GIVEN, RECORD DOLLAR AMOUNT OF
LAST "YES" ANSWER.
IF NO, DECREASE THE DOLLAR AMOUNT IN
$10 INCREMENTS UNTIL A "YES" ANSWER IS
GIVEN. E.G., "Would you be willing to
pay $115 more each year to raise the
water quality from level 'C' to level
'B'?" ETC. WHEN A "YES" ANSWER IS
GIVEN, RECORD DOLLAR AMOUNT.
DOLLARS (GO TO B-4)
(40-42)
D-26
-------
d. Would you (and your family) be willing to pay $125 each year in
higher taxes and prices for products that companies sell to raise
the water quality from level "D" to level "B", where you can swim in
it and it is improved for other activities? CIRCLE NUMBER.
YES
NO
01
02
F YES, INCREASE THE DOLLAR AMOUNT IN
10 INCREMENTS UNTIL A "NO" ANSWER IS
IVEN. E.G., "Would you be willing
o pay $135 each year to raise the
ater quality from level 'D' to level
B'?" ETC. WHEN A "NO" ANSWER IS
IVEN, RECORD DOLLAR AMOUNT OF LAST
YES" ANSWER.
1
IT NO, DECREASE THE DOLLAR AMOUNT IN
$10 INCREMENTS UNTIL A "YESV ANSWER IS
GIVEN. E.G., "Would you be willing to
pay $115 each year to raise the water
quality from level 'D' to level 'B1?'1
ETC. WHEN A "YES" ANSWER IS GIVEN,
RECORD DOLLAR AMOUNT.
DOLLARS
;IF ANY AMOUNT IN a., GO TO B-
IT ZERO DOLLARS IN a. AND:
• ANY AMOUNT IN d., GO TO B
• ZERO DOLLARS IN d., GO TO
" }
-4.d.; I
B-3.e./
(42-45)
We have found in studies of this type that people have a lot of
different reasons for answering as they do. Some people felt they
did not have enough information to give a dollar amount, some did
not want to put dollar values on environmental quality, and some
objected to the way the question was presented. Others gave a zero
dollar amount because that was what it was worth to them.
Which of these reasons best describes why you answered the way you
did? REPEAT REASONS IT NECESSARY AND CIRCLE NUMBER.
NOT ENOUGH INFORMATION
01
DID NOT WANT TO PLACE
DOLLAR VALUE 02
OBJECTED TO WAY QUESTION
WAS PRESENTED 03
THAT IS WHAT IT IS WORTH . 04
OTHER (SPECIFY) ...... 05
(46-47)
TO B-6)
D-27
-------
PART 2
SUGGESTIONS FOR IMPROVING THE QUESTIONNAIRE
FOR FUTURE USE
Any survey questionnaire can be improved based on the additional infor-
mation learned in the execution of the survey. This questionnaire is not an
exception. One of the most significant changes would amend the word "addi-
tional" to the introduction of Question B-7 to clarify that the bid amount is in
addition to the amounts previously bid. It is also unclear whether the supply
uncertainty dimension added in this question is effectively expressed. This
could be improved with a couple of clarifying sentences.
The introduction to Question B-5 could be improved by better explaining
how water quality might be worsened only for some weekends. For example,
a sentence describing "the effect of higher water temperatures in the summer
months could reduce water quality only in that part of the year" might clarify
the supply uncertainty that is intended in this question.
The explanation and introduction to the contingent ranking format is too
brief. While this may be minimized by the respondent's familiarity with water
quality from the other contingent valuation questions, it would require expan-
sion for an application as an independent format. This introduction could also
explain in more detail the relation between water quality levels and the amounts
paid.
There is a slight difference in wording between Versions A and B and C
and D as a result of a word processing error. The phrase "where it is not
even clean enough for boating" was inadvertently omitted from Question B-3a
in Versions C and D. The water quality ladder would have shown that E was
not suitable for boating, so the potential bias here is likely small but, none-
theless, could be avoided in future use.
Finally, some changes might be useful in the visual aids. The cards in
the contingent ranking should be same size as the other aids to make them
easier to handle. For consistency with the other aids, the value card could
have been done in bolder print to make it stand out. There is some debate
that a visual aid describing the payment vehicle might have made it clearer to
people how they currently pay for water quality. On the other side of this
argument is the thinking that this may actually increase the respondents' con-
fusion .
In summary, the questionnaire performed well for most of the key ques-
tions, but some relatively minor changes might have made it even better. The
question responses most affected by the change are the existence value re-
sponses in Question B-7.
D-28
-------
APPENDIX E
TECHNICAL WATER QUALITY MEASURES:
AN ECONOMIST'S PERSPECTIVE
E.1 INTRODUCTION
A discussion of water quality measurement should define the term water
quality, including descriptions of the various attributes that determine quality.
Although seldom together, several disciplines have repeatedly explored this
issue, and a significant amount of literature is relevant to the questions that
arise in benefit estimation. This appendix discusses several of these ques-
tions.
E.2 AN OVERVIEW OF TECHNICAL WATER QUALITY MEASURES
E.2.1 Introduction
The following sections briefly describe technical measures of water qual-
ity. Sections E.2.2 and E.2.3 discuss freshwater systems, focusing on their
characteristics and their ability to assimilate effluents. Section E.2.4 dis-
cusses commonly used parameters, noting their importance in an ecosystem,
their measurement, and the ability of individuals to perceive their changes.
E.2.2 Water Quality in Freshwater Systems
Freshwater areas are intricate systems differing in attributes and causal
relationships. Freshwater system descriptions are complicated by climate,
geography, land use, water management, and existing plants and animals.
Because these particular characteristics are usually unknown, actual physical
relationships cannot be determined. Descriptions are further complicated when
scientific analysis cannot measure deleterious long-term or synergistic effects
in a natural setting.
Freshwater systems are either lentic systems, which contain standing
water such as lakes, or lotic systems, which contain running water such as
streams and rivers. However, classifying a system as lentic or lotic can be
difficult when natural impoundments, dams, and reservoirs occur in either.
In addition, while the basic nutrient cycles are the same for both systems,
life cycles and pollution effects differ considerably.
The scope of this project limits discussion only to lotic systems. Im-
poundments are considered due to their general dynamic nature. However,
because the unique lentic system characteristics sometimes appear in natural
and manmade impoundments, problems common to both system types are also
discussed.
E-1
-------
E.2.3 Assimilative Capacity
The ability of a lotic system to assimilate effluents determines actual pol-
lution levels. Assimilative capacity is usually defined with respect to the ab-
sence of deleterious effects with a given level of discharge into a receiving
water. However, any materials discharged into the water have an effect. The
major problem is one of identifying and measuring these changes and of deter-
mining when they become deleterious. An effluent's effect on the environment
is influenced by time period, amount of available oxygen, plant nutrients, and
locational characteristics.
Daily and seasonal variation in the speed of nutrient cycling are major
determinants of an effluent's effect on water quality. Lotic systems derive
most of their nutrients from soil runoff, causing primary productivity to vary
seasonally. As land nutrient and groundwater levels vary, so does the lotic
environment's assimilative capacity. Available sunlight is the primary source
of daily variation, with the peak rate of photosynthesis in the afternoon hours
causing peak levels of dissolved oxygen.
Assimilative capacity is commonly measured by the availability of dissolved
oxygen. Because all aquatic animal life depends on dissolved oxygen, low dis-
solved oxygen levels may cause a reduction in species diversity and number.
Some effluents reduce dissolved oxygen because they change the rate of photo-
synthesis, the solubility of oxygen, and the diffusion of atmospheric oxygen or
they increase aerobic bacteria activity.
Existing plant nutrients also determine the effect of effluents. Each eco-
system has a defined nitrogen-phosphorus ratio, and all organisms within the
system can use nutrients only in this ratio. When an effluent increases nutri-
ent levels, a natural growth limit is eliminated, resulting in excessive plant
growth, which eventually decomposes and decreases dissolved oxygen.
Long-term changes in assimilative capacity occur due to an aging process.
As erosion takes place, headwaters tend to migrate upstream, as will plant and
animal communities. Erosion is also responsible for increases in suspended
solids, which deteriorate and affect the composition of the river bottom over
time.
E.2.4 Water Quality Parameters
The capacity of a water system to accommodate uses may be defined by a
series of hydrological, physical, chemical, and biological parameters. These
parameters are relevant in explaining the effects of an effluent on the equilib-
rium and existing conditions. Both relative and absolute measurements are
important in evaluating parameters. No single parameter can be used as an
adequate measure of water quality, yet in many cases focusing on one param-
eter is dictated by data limitations. Several types of parameters describe
water quality, and a brief discussion of each follows.
E-2
-------
Hydrological Parameters
Hydrological parameters determine the level of physical, chemical, and
biological parameters. These parameters characterize the atmosphere and
catchment area, and care is required in placing the analysis in a particular
hydrological process. Consideration should therefore be given to climate,
properties of air, precipitation, erosion, and vegetation.
Most studies that attempt to measure water quality do not explicitly con-
sider hydrological parameters. Care is taken only to place measurements in a
particular season. For example, flow is often described as important but not
considered directly. This treatment can be explained by a lack of data on
how often hydrological parameter changes occur and their synergistic effect
on the level of other parameters. A possible methodology to include these
parameters would be to use water quality modeling. This technique, however,
requires large amounts of information and time.
Physical Parameters
Physical parameters are commonly used water quality measures. However,
their values vary significantly due to seasonal and diurnal patterns and site-
specific characteristics. Readings may not be applicable to wide areas due to
these variations. These parameters include the following:
Turbidity is caused by the presence of suspended solids. These solids
are usually a variety of substances influenced by man-made and
natural occurrences. Increases in suspended solids will affect the
level of photosynthesis as transparency is decreased. Also, as
settling occurs, eggs and larva may be suffocated, affecting fish
reproduction and species diversity. Water turbidity is usually
measured by a Seechi disk. This disk is lowered into the water
until it disappears, and the resulting depth is recorded. Alterna-
tively, the Jackson Turbidity Unit can be used. Regardless of the
measurement technique, individual perceptions of turbidity are
thought to be generally correlated with measured levels, explaining
its common use in water quality studies. Unfortunately, little is
known of how sensitive individuals are to small turbidity changes
and what importance this has in their decisionmaking.
Color is important in determining both transparency and aesthetics of
water. Water may contain a variety of compounds that change the
amount of sunlight allowed in a water column, resulting in a change
in the photosynthesis rate. Color is usually determined by visual
comparison to a group of standard colors. The use of this param-
eter in water quality studies is rare due to the lack of consistent
measurement over time and among sites. The link between color and
individual perceptions is also not well known.
Temperature is a major determinant of the level of biological and chemical
activity because temperature changes also cause a change in the
equilibrium of a water system. Lotic systems are greatly affected by
E-3
-------
atmospheric temperature and usually do not contain any thermal
stratification. For these reasons organisms are usually tolerant of
large temperature changes. When impoundments occur in the lotic
environment, temperature stratifications do occur, inhibiting the
availability of dissolved oxygen at certain levels. Temperature read-
ings are taken at various depths with a reversing thermometer or
bathythermograph. Simple temperature readings are not a good in-
dicator of water quality. A more appropriate measure would be de-
viation from the norm caused by man-made and natural infractions.
A change in temperature is usually perceived through indirect
changes such as algae growth, changes in fish population, and
physiological disturbances in swimmers.
Odor and taste measure the presence of industrial discharges, micro-
scopic organisms, and vegetation. These factors are usually the
result of industrial discharge or aquatic decomposition. The meas-
urement of odor is determined by concentration levels of various
compounds in a sample. Effects of odor are difficult to measure
because perceptions vary depending on the individual and distance
to the water.
Chemical Parameters
Chemical parameters characterize natural and man-made components of a
particular water sample. Reported results are often misleading because the
parameters may not be measured from a desired area. The choice of parame-
ters and sample sites usually is based on pollutants expected due to regional
and man-made characteristics. Also, cause and effect relationships are not
precisely known in the scientific community nor are changes well perceived by
individuals. Thus, we cannot determine exact relationships between parame-
ters and water quality. Usually only the direction of change in water quality
is known. Common chemical parameters are as follows:
Dissolved oxygen measures the intensity of organic decomposition and the
ability of self purification. Dissolved oxygen is necessary for res-
piration of plants and animals and aerobic decomposition. Concen-
trations of dissolved oxygen are increased with photosynthesis and
atmospheric reaeration. Decreases are caused by nitrification, bio-
logical oxygen demand, and benthal oxygen demand. Many species
are not tolerant of low levels of dissolved oxygen, and offensive
odor may also occur as decomposition occurs without the presence of
oxygen. Dissolved oxygen is expressed in terms of mg/liter or per-
cent saturation. Extensive work has been completed on fish popula-
tions and levels of dissolved oxygen. These controlled experiments
relate fish reproduction rates to minimum dissolved oxygen require-
ments for various species.
Total dissolved solids represent the concentration of nondegradable wastes
in a water sample. These solids may be toxic to the surrounding
food chain, but little is known about this relationship. Concentra-
tions are usually in terms of mg/liter.
E-4
-------
pJ-[ is an index of the acidic-basic relationship of various mineral and
basic substances. Under natural conditions, pH ranges from 5.0 to
8.6 on a scale of 1 to 14. Heavily polluted water may cause a low
pH (i.e., an increased concentration of acid). Existing plant and
animal life may not be tolerant of severe pH changes. A pH change
generally results in a smaller variety of organisms. Recreation use
of water usually requires a pH in the range occurring in natural
conditions. However, swimming may require a narrow range of 6.5
to 8.3. Individual perceptions of pH are sensitive only to large
changes, though a change may be perceived through eye irritation
or touch.
Nitrates are formed by the biochemical oxidation of ammonia. Some strat-
ification occurs naturally, resulting in surface waters having higher
concentrations. Increased concentration may indicate fecal pollution
in the preceding period. The concentration of nitrates may also
indicate the rate of self purification of a water system. Nitrates
are usually reported as mg/liter.
Metals present in a lotic environment can be caused by soil drainage.
Therefore, seasonal changes will affect the concentration of metals
present. Industrial sources of metals include mine pit discharge,
ore enriching factories, and iron and steel factories. The effects
of several metals such as copper, lead, and mercury are commonly
studied and well known. The effects of other metals such as chrom-
ium, cadmium, cobalt, and nickel are not as well known. Concentra-
tions are usually reported as mg/liter. Severe concentrations may
inhibit development if they are passed to higher members of the food
chain.
Surface active agents represent a variety of man-made compounds. These
agents or surfactants are usually found in detergents. Concentra-
tions result in the normal breakdown of organic material. More
noticeable effects are a bitter taste, a soapy and kerosene odor,
and the presence of foam. Concentrations usually are measured in
terms of mg/liter.
Pesticides are any substance designed to destroy plant or animal organ-
isms. These compounds enter the water indirectly from runoff and
drainage or by direct application. Agriculture is the dominant
source of pesticide contamination. Many pesticides have a cumulative
effect, causing increased concentrations at higher levels of the food
chain. As concentrations increase, the natural development of
organisms will be altered. Pesticides include a wide variety of com-
pounds and are usually described in mg/liter. Even though their
diversity usually precludes their use as a measure of water quality,
pesticides are considered an important indicator of water quality.
E-5
-------
Biological Parameters
Biological parameters reveal the quality, size, and type of animal and
plant populations within a water system. Data readings vary significantly with
the season and flow velocity, but these parameters may give a reliable picture
of the average situation since organisms cannot rapidly adapt to change.
Individuals do not directly perceive changes in these parameters but notice
them through such effects as odor, algae, and resulting illness. These factors
are most important to direct contact uses but also apply to secondary recrea-
tion. The two important biological parameters are as follows:
Biological oxygen demand measures the rate of oxygen consumption in a
system due to organic decomposition. High levels of organic waste
cause an increase in the biological oxygen demand and a resulting
decrease in available dissolved oxygen. These rates will differ
depending on the state of the matter being decomposed. Since
temperature controls the rate of organic activity, it also greatly
influences oxygen demand. Biological oxygen demand is generally
measured as the amount of oxygen removed from a sample in a 5-day
period and is an important part of most water quality determinations.
However, sample readings may not be comparable due to changes in
assimulative capacity. For example, a reading may have a large
value and yet have little effect on water quality due to characteris-
tics such as large available dissolved oxygen and strong flow.
Microbiological parameters determine the presence of waterborne disease.
The parameters would include bacteria, viruses, and algae. Both
bacteria and viruses may be excreted in the feces of infected ani-
mals. The most common parameter of fecal contamination is the test
for coliform bacteria expressed as number of bacteria per liter.
Limits are currently set on fecal coliform depending on the use of
the river. The presence of bacteria and viruses does not affect
the appearance of the water. Except at high levels, algae is not
toxic but may indicate overfertilization of the system by man or
other mammals. Algae may be considered a pollutant since it is
readily noticed in water.
E.3 ISSUES IN DETERMINING WATER QUALITY
E.3.1 Introduction
Several issues arise in attempts to define water quality, the most impor-
tant of which involve the uses of a water system as they affect quality and
the selection of an appropriate site. A discussion of these two issues follows,
including a brief description of how they relate to this study.
E.3.2 Water Quality and Use
Water quality is directly dependent on current and future uses of a site.
Common use categories are drinking, swimming, fishing, boating, and indus-
trial This list is an obvious simplification as it does not recognize the attnb-
E-6
-------
utes desirable for each use. The use of water for drinking, for example, may
occur within a wide range of attributes given various levels of water treat-
ment. The inability to define these attribute ranges causes oversimplification
when water quality is measured over various uses.
Uses of a water system are related to each other in a spatial and tempo-
ral sense. As the level of one use changes, the benefits derived from compet-
ing uses will also change. This relationship is not well defined because it
depends on several variables, including the particular uses considered, char-
acteristics of the area, and the time frame considered. In some instances,
the relationship may depend on differential preferences of the potential users
(e.g., teenagers and young families may desire a .crowded beach while honey-
mooners and older people may prefer an uncrowded beach), and, in extreme
cases, uses may be completely independent or mutually exclusive.
To ensure the same uses at each site in the travel cost approach, this
study used only U.S. Army Corps of Engineer areas. Using only these areas
eliminates part of the problem of defining uses, but it does not account for
competing uses. Ideally, more consideration should be given to variation in
uses between sites and their relationship to each other.
E.3.3 Water Quality Within an Area
Water quality is related to the physical boundaries of the study area in
two ways: boundaries determine both the physical attributes and the scien-
tific parameters to consider. In turn, physical attributes determine the uses
allowed and the interrelationship between uses. For example, the presence of
a dam increases the damage caused by an industrial effluent on fish popula-
tions.
The determination of the appropriate scientific parameters is subject to
the continuous nature of water quality. As these measurements vary between
measuring sites, the problem becomes more complex. Quality of water to a
user is determined by the immediate and surrounding area. How to incorpor-
ate these readings is not clear. Consideration should be given to uses in-
volved, as well as the physical relationship between areas. This issue is
clouded by incomplete data when water quality is actually measured.
Data availability ultimately constrains the determination of the study area.
The locations of existing monitoring sites are based on a variety of concerns
such as location of fisheries, effluents present, and convenience. Quite often
the measurements obtained do not conform to the desirable study requirements.
Hence, the use of these data may bias results depending on site proximity to
the study area and the use being considered.
E.4 MEASUREMENT OF WATER QUALITY
E.4.1 Introduction
A useful measure of water quality would be a universal number or index
that can compare uses and scientific parameters. Both individual perceptions
E-7
-------
of parameters and scientific measures of parameters could be used individually
or to compare to an index. However, assigning the appropriate weights to
each measure is a difficult task. A brief discussion of advantages to various
methods to describe water quality follows.
E.4.2 Human Perceptions and Water Quality Measurement
Individual perceptions play an important role in water quality determina-
tion, but consistent measurement of perceptions is a major problem. Studies
have shown that perceptions usually vary with questionnaire design information
provided and sample population. Binkley and Hanemann [1978] found that
respondents base evaluations of water quality on incorrect information. Ditton
and Goodale [1973] found that respondents tended to describe areas closest to
their residence, which causes large variations in the water quality rating over
the entire study area. Moreover, changes in other site attributes limit the
ability to draw general conclusions as to the effects of changes in water qual-
ity alone. On the other hand, Bouwes and Schneider (1979) found reasonably
good correlation between perceptions and the scientifically based lake condition
index.
Some differences in perceptions have been attributed to characteristics of
the respondents. Barker [1971] found that users of an area tend to rate
water quality more favorably than nonusers. Ditton and Goodale [1973] deter-
mined that swimmers' perceptions of water quality differed from fishermen's,
both in terms of their ratings of water quality and the relative importance of
individual features.
E.4.3 Technical Water Quality Measurement
Scientifically measured parameters are usually good indicators of water
quality changes. Unlike individual perceptions, the technical water quality
tests are usually comparable over time and between sites. Determination of
important parameters is difficult, however, since most scientific information is
obtained only through controlled experiments. Changes in water quality caused
by parameters are difficult to determine because particular site characteristics
must be known to determine an expected change, even in the short run. In
addition, long-term and synergistic effects also usually cannot be determined
because of poor information.
£.4.4 Water Quality Indexes
An ideal water quality measure would combine scientifically measured
parameters, individual perceptions, and alternative uses of an area. Unfor-
tunately, these measures require considerable information, and their compo-
nents may vary between sites. In lieu of complete information, many studies
have used approaches that rely on individual parameters or indexes to deter-
mine water quality. While most studies have used one or more individual
parameters without determining their relative importance, other studies have
used the index approach to solve several of the problems noted above. Thus,
while far from perfect, the index approach does represent a tractable method
of relating water quality to users, perceptions, and scientific judgment.
E-8
-------
E.4.4.1 The National Sanitation Foundation Index
The ideal measure of water quality would incorporate scientific parame-
ters, public perception of the water, and potential uses of the water. As an
attempt to incorporate these considerations, the National Sanitation Foundation
(NSF) index is a constructive approach to several problems in water quality
measurement. A composite of nine parameters, the NSF index was developed
through several questionnaires given to individuals with water quality experi-
ence. Respondents first selected parameters they felt were important to water
quality. Followup contacts were then made to give the previous group
responses to the respondents and to allow them to change their initial
responses. A rating of these parameters in terms of water quality and syner-
gistic effects was then developed based on these responses. The final param-
eters chosen included dissolved oxygen, fecal coliform density, pH, 5-day bio-
logical oxygen demand, nitrates, phosphates, temperature, turbidity, and total
solids.
The next step in developing the NSF index required the development of
water quality curves for each parameter. These curves represent the expect-
ed result of parameter concentrations on water quality and must be combined
with the relative weights derived from the respondents' rankings of each
parameter. These quality curves and weights constitute the final components
of the index. More details on this index can be found in EPA [1982].
Researchers have applied the NSF index in a number of studies. The
U.S. Environmental Protection Agency (EPA) applied the NSF index to the
Kansas River basin to determine its effectiveness, including an appraisal of
sampling and computing difficulties. The Kansas River basin, a wide, shallow
river of moderate velocity, has light industry and receives treated municipal
wastes from over 40 cities and towns. EPA calculated two forms of the NSF
index with almost 600 water samples from over 26 sites. Calculated index
values were consistent with researchers' attitudes toward the various reaches
of the river..
The index calculations were also used to examine several other concerns.
For example, the correlations between several variables were measured to test
the validity of substituting parameters when certain data do not exist. The
study determined that suspended solids can be substituted for turbidity and
total coliform for fecal coliform.
The NSF index provides a scientifically based method of linking changes
in water quality to the effects of those changes. The index, however, does
not provide a linkage to individual perceptions of water quality changes and
cannot differentiate threshold values for specific uses like fishing or swimming.
E.4.4.2 Resources for the Future Water Quality Ladder
A significant problem with the NSF index is that it does not take into
account potential uses for a particular body of water. At Resources for the
Future (RFF), Vaughan in Mitchell and Carson [1981] used a variation of the
NSF index to determine minimum levels of water quality for various uses.
Specifically, Vaughan's index used five NSF index parameters chosen on the
E-9
-------
Table E-1. Water Quality Classes by Parameter and Index Values
Measurable water quality characteristics
Water quality
use designation
Acceptable for
drinking with-
out treatment
Acceptable for
swimming
Acceptable for
game fishing
Acceptable for
rough fishing
Acceptable for
boating
Fecal
coliform
(tt/100 mL)
0
200
1,000
1,000
2,000
Dissolved
oxygen
(mg/L)a
7.0 (90)
6.5 (83)
5.0 (64)
4.0 (51)
3.5 (45)
5-Day
BOD
(mg/L)
0
1.5
3.0
3.0
4.0
Turbidity
(JTU)
5
10
50
50
100
pH
7.25
7.25
7.25
7.25
4.25
Ladder
value
9.5
7.0
5.1
4.5
2.5
Numbers in parentheses are percent saturation at 85° F.
basis of judgment and data availability: fecal coliform, dissolved oxygen, bio-
logical oxygen demand, turbidity, and pH. As shown in Table E-1, Vaughan
associated specific parameter levels with five use designations. He then used
a truncated version of the NSF index to place each minimum use designation
on an index value range from 0 to 10 with the final index values for each use
classification shown in Table E-1.
The RFF index provides a valuable link between various parameters and
use designations. Even though the parameter choice may be somewhat arbi-
trary, the parameters neatly map into desirable attributes for a particular use.
However, the RFF index does not account for differing individual perceptions
that may be easily incorporated with further research. The RFF ladder is
used in this study, as shown in Figure 4-5.
E.5 SUMMARY
The questions involved in defining water quality are complex, and there
are no clear answers. Water quality studies must jointly determine the param-
eters to be considered, the uses to be considered, and the definition of the
site to be studied. In addition, each of these issues has many aspects, such
as how to define the relationship between uses, and each is subject to the
constraint of available data. To date, very little has been done to measure
water quality between sites or over time. One exception would be the RFF
and NSF indexes, which measure various aspects of water quality and weigh
them using informed judgment. Further research in this direction could lead
to an index that incorporates individual perceptions and unique characteristics
of an area.
E-10
-------
APPENDIX F
TRAVEL COST: SUPPORTING TABLES
This appendix contains tables displaying data that support the travel cost
analysis presented in Chapter 7. Tables F-1 through F-4 provide additional
data for the benefits calculations. Table F-5 shows the tailored models that
were estimated for selected sites.
F-1
-------
Table F-1. Distribution of Benefit Estimates (Consumer Surplus Loss
Avoided) for Loss of Use of the Monongahela River by Income
Levels for 33 Sites for Individual Users
Benefit
Income
(1981 dollars)
0 - 5,000
5,000 - 10,000
10,000 - 15,000
15,000 - 20,000
20,000 - 25,000
25,000 - 30,000
30,000 - 35,000
35,000 - 40,000
40,000 - 45,000
45,000 - 50,000
50,000 and above
Total
0
0
1
1
2
1
2
1
0
0
0
0
8
0-
10
1
1
0
2
1
1
0
0
0
0
0
6
10-
20
0
3
1
0
1
2
0
0
0
1
0
8
20-
30
0
1
1
1
1
0
0
1
0
1
1
7
30-
40
0
1
0
3
1
0
0
0
0
0
0
5
estimate (1977 dollars)3
40-
50
0
0
0
0
0
1
0
0
0
0
0
1
50-
60
2
1
0
1
0
2
0
2
0
2
_g
10
60-
100
0
0
0
0
1
0
0
1
1
0
0
3
100-
150
1
1
2
2
0
0
3
1
0
0
J_
11
150-
200
0
0
1
0
0
0
0
0
0
0
0
1
200
and
above
0
0
0
1
0
1
0
0
0
0
1_
3
Total
4
9
6
12
6
9
4
5
1
4
_3
63
aTo convert to 1981 dollars, multiply the endpoints of the benefit scale by 1.55.
F-2
-------
Table F-2. Distribution of Benefit Estimates (Consumer Surplus Loss
Avoided) Due to Loss of Use of the Monongahela River by Survey User
Income for 33 Sites—Includes Multiple Visits
Benefit
1 ncome
(1981 dollars)
0
5,000
10,000
15,000
20,000
25,000
30,000
35,000
40,000
45,000
50,000
Total
- 5,000
- 10,000
- 15,000
- 20,000
- 25,000
- 30,000
- 35,000
- 40,000
- 45,000
- 50,000
and above
0-
10
0
0
0
0
1
1
0
0
0
0
0
2
10-
20
0
0
0
0
0
0
0
0
0
0
0
0
20-
30
0
0
0
1
1
1
0
0
0
0
0
3
30-
40
1
0
0
1
1
1
1
0
1
0
0
6
estimate (1977 dollars)3
40-
50
2
0
0
2
0
1
2
0
0
0
0
7
50-
60
0
0
0
0
1
4
0
0
0
0
0
5
60-
70
4
2
0
2
0
4
1
1
1
0
_0
15
70-
80
3
5
5
7
2
8
2
1
0
4
_2
39
80-
90
4
4
3
5
0
2
1
1
1
0
_0
17
Total
10
11
8
18
6
22
7
3
3
4
_2
94
To convert to 1981 dollars, multiply the endpoints of the benefit scale by 1.55.
F-3
-------
Table F-3. Distribution of Benefit Estimates (Consumer Surplus
Increment) Due to Water Quality Improvement: Beatable
to Fishable by Survey User Income for 33 Sites--
Includes Multiple Visits
Benefit estimate (1977 dollars)
Income
(1981 dollars)
0
5,000
10,000
15,000
20,000
25,000
30,000
35,000
40,000
45,000
50,000
Total
- 5,000
- 10,000
- 15,000
- 20,000
- 25,000
- 30,000
- 35,000
- 40,000
- 45,000
- 50,000
and above
0-10
0
0
0
1
3
5
7
3
3
4
_2
28
10-20
1
0
0
5
3
17
0
0
0
0
_0
26
20-30
7
11
8
12
0
0
0
0
0
0
_0
38
30-40
2
0
0
0
0
0
0
0
0
0
0
2
Total
10
11
8
18
6
22
7
3
3
4
_2
94
aTo convert to 1981 dollars, multiply the endpoints of the benefit scale by
1.55.
F-4
-------
Table F-4. Distribution of Benefit Estimates (Consumer Surplus Increment)
Due to Water Quality Improvement: Beatable to Swimmable
by Survey User Income for 33 Sites--Includes Multiple Visits
Income
(1981 dollars)
0 - 5,000
5,000 - 10,000
10,000 - 15,000
15,000 - 20,000
20,000 - 25,000
25,000 - 30,000
30,000 - 35,000
35,000 - 40,000
40,000 - 45,000
45,000 - 50,000
50,000 and above
Total
Benefit estimate (1977 dollars)3
0-10
0
0
0
0
1
3
3
3
3
4
_2
19
10-20
0
0
0
2
2
11
4
0
0
0
_0
19
20-30
1
0
0
3
3
8
0
0
0
0
_0
15
30-40
2
0
5
13
0
0
0
0
0
0
_g
20
40-50
4
11
3
0
0
0
0
0
0
0
_0
18
50-60
3
0
0
0
0
0
0
0
0
0
0
3
Total
10
11
8
18
6
22
7
3
3
4
_2
94
aTo convert to 1981 dollars, multiply the endpoints of the benefit scale by
1.55.
F-5
-------
Table F-5. Regression Results of Tailored Models for Selected Sites
i
o>
Site
Site number
Lock and Dam No. 2 302
(Arkansas River
Navigation System), AR
Beaver Lake, AR 303
Blakely Mt. Dam, Lake 307
Ouachlta, AR
Intercept
2.63
(7.12)
2.39
(8.24)
2.25
(8.06)
1.94
(6.02)
2.25
(9.41)
1.69
(9.09)
1.70
(11.98)
1.48
(13.17)
1.48
(12.03)
1.74
(16.32)
1.58
(5.59)
1.53
(5.87)
1.69
(9.71)
1.28
(5.95)
1.88
(9:22)
T+M cost
-0.012
(-2.20)
-0.012
(-2.08)
-0.013
(-2.29)
-0.013
(-2.46)
-0.010
(-1.67)
-0.007
(-12.46)
-0.007
(-12.10)
-0.007
(-13.04)
-0.007
(-12.85)
-0.006
(-11.75)
-0.008
(-5.14)
-0.008
(-5.18)
-0.008
(-5.09)
0.008
(-5.23)
-0.007
(-4.89)
Income
8.7 x 10~5
(-1.37)
-1.8 x 10~5
(-1.08)
-1.6 x 10"5
(-0.86)
-1.5 x 10~5
(-0.88)
-8.5 x 10~6
(-0.45)
-1.2 x 10"5
(0.68)
-3.8 x 10~6
(-0.84)
-2.3 x 10~6
(-0.51)
-4.0 x io"e
(-0.91)
-1.9 x 10~6
(-0.43)
6.6 x 10~6
(0.24)
-6.3 x 10"S
(-0.79)
-7.9 x 10~6
(-1.01)
-9.8 x 10~6
(-1.31)
-7.0 x 10~6
(-0.92)
Income
squared Age Sex RECIMP Day R2
3.1 x 10"9 0.17
(1.12)
-0.003 0.15
(-0.50)
0.062 0.14
(.037)
0.378 0.20
(1-62)
-0.209 0.18
(-1.30)
1.9 x 10~9 0.43
(0.51)
-0.003 0.44
(-0.92)
0.212 0.45
(2.32)
0.191 0.44
(1.82)
-0.310 0.46
(-3.18)
-3.2 x 10~10 0.24
(-0.53)
0.005 0.24
(0.84)
0.048 0 24
(0.31)
O.S55 0.31
(3.00)
-0.275 0.26
(-1.56)
DF
37
37
37
37
37
222
222
222
222
222
87
87
87
87
87
F-
ratlo
2.51
2.12
2.07
3.04
2.67
57.02
57.37
60.05
58.83
62.83
9.13
9.31
9.05
12.93
10.07
DF = Degrees of freedom.
"t-values of no association are shown In parentheses. RECIMP Is a binary variable that Is 1 If the respondent considers recreation to be Impor-
tant. Day Is a binary variable that Is 1 If the respondent stayed 1 or more days..
-------
Table F-5. (continued)
Site
Site number Intercept
Cordell Hull Dam and 310 1.97
Reservoir, TX (9.96)
1.58
(7.74)
1.65
(10.41)
1.87
(9.14)
1.88
(14.25)
Dewcy take, KY 312 0,26
(6.64)
0.16
(0,55)
0.08
(Ol36)
0; 43
(2117)
0.54
(2l79)
Grapevine Lake, TN 314 1.54
(7,16)
2.16
(13:76)
1.74
(13196)
1.44
(8.05)
1.80
(15.13)
T+M cost
-0.014
(-5.94)
-0.015
(r6.28)
-0.014
(-6.33)
-6.014
(-5.93)
-6.013
(-5.63)
-p. 002
(-2.74)
-0.003
(-3:19)
-0.003
<-3.67)
-0.002
(-2.91)
-0.002
(-1185)
-0.007
(-8.78)
-0.006
^-7.89)
-0.007
(-8.85)
-0.007
(-9.26)
-0.009
(-6.62)
Income
-1.4 x 10"5
(-0.63)
2.8 * 10"6
(0.33)
2.4 x 10~6
(0.29)
5.6 x 10"8
(0.01)
1.4 x 10~6
(0.17)
3.6 x 10~5
(1-01)
1.9 x 16'5
(1.91)
2.5 x 10~5
(2.74)
2.0 x 10~5
(1.99)
1.9 i 10"5
(1.96)
1 .5
3.5 x 10 3
(1.74)
7.6 x 10~5
(1.59)
8.0 x 10~6
(1.59)
9.4 x 10~6
(1.92)
9.2 x 10~6
(1.77)
Income
squared Age Sex RECIMP Day Rz
3.6 x 10~10 0.34
(0.67)
0.007 0.36
(1.74)
0.311 0.37
(2.29)
-0.021 0.34
(-0.11)
-0.208 0.35
(-1.35)
-3.1 x 10~10 0.18
(-0.47)
0.009 b.21
(1.24)
0.498 0.31
(2.89)
-0.018 0.18
(-0.10)
-0.359 0.24
(-1.78)
-10
-5.4 x 10 '" 0.48
(-1-36)
-0.013 0.52
(-3.14)
0.109 0.47
(1.00)
0.392 0.50
(2.47)
-0.296 0.44
(-2.36)
DF
100
100
100
100
100
42
42
42
42
42
88
88
88
88
88
F-
ratio
17.10
18.40
19.52
16.88
17.79
3.16
3.70
6.46
3.08
4.37
26.94
31.95
26.41
29.60
22,88
DF = Degrees of freedom.
at-values of no association are snown In parentheses. RECIMP Is a binary variable that
tant. , Pay Is a binary variable that ls.1 .Ifrt.he, resppndent stayed 1 or more days.
Is 1 if the respondent considers recreation to be impor-
-------
Table F-5. (continued)
Site
Sit* number Intercept T+M cost
Greers Ferry Lake, AR 315 1.49
(8.04)
1.61
(10.63)
1.45
(12.25)
1.15
(6.69)
1.76
(15.29)
Grenada Lake, MS 316 1.91
(7.47)
1.81
(7.07)
2.06
-H (11.44)
OS 1.28
(4.31)
2.03
(13. on
Lake Washington Ship 320 2.69
Canal, WA (3.27)
1.10
(2.20)
0.81
(2.11)
1.00
(2.01)
Melvern Lake, KS 322 1.87
(3.93)
-0.006
(-8.91)
-0.006
(-9.09)
-0.006
(-8.97)
-0.007
(-9.34)
-0.006
(-8.90)
-0.010
(-4.37)
-0.009
(-4.31)
-0.009
(-4.16)
-0.010
(-4.62)
-0.008
(-3.57)
-0.004
(-4.16)
-0.003
(-2.98)
-0.004
(-3.81)
-0.004
(-3.64)
EQUATION 5
.-0.008
(-1.60)
Income
Income squared Age Sex
7.3 x 10~6 2.6 x 10~11
(0.35) (0.05)
9.6 x 10~6 -0.004
(1.60) (-1.14)
8.4 x 10~6 0.054
(1.42) (0.53)
9.0 x 10~6
(1.55)
1.0 x 10~5
(1.92)
2.1 x 10~5 -1.6 x 10~9
(0.41) (-0.63)
-5.0 x 10'6 0.005
(-0.32) (1.13)
-1.0 x 1(f5 -0.049
(-0.65) (-0.32)
-9.6 x 10~6
(-0.67)
-1.8 x 10~6
(-0.12)
-1.6 x 10~4 4.3 x 10~9
(-2.06) (2.36)
1.6 x 10~5 -0.005
(0.73) (-0.52)
1.9 x 10~5 0.234
(0.94) (0.92)
1.7 x 1(f5
(0.81)
IS NOT OF FULL RANK BECAUSE ALL VISITS
-6.7 x 10~5 1.6 x 10~9
(-1.36) (1.5)
RECIMP Day R2
0.28
0.28
0.28
0.372 0.29
(2.39)
-0.494 0.35
(-4.89)
0.22
0.23
0.22
0.806 0.30
(2.98)
-0.419 0.28
(-2.51)
0.35
0.26
0.27
-0.079 0.26
(-0.24)
WERE DAY VISITS.
0.11
DF
213
213
213
213
213
72
72
72
72
72
39
39
39
39
41
F-
ratlo
27.07
27.66
27.20
29.70
38.06
6.76
7.14
7.36
10.36
9.26
6.95
4.57
4.83
4.48
1.69
DF = Degrees of freedom.
"t-values of no association are shown In parentheses. RECIMP Is a binary variable that is 1 If the respondent considers recreation to be Impor-
tant. Day Is a binary variable that Is 1 If the respondent stayed 1 or more days.
-------
Table F-5. (continued)
Site
Site number Intercept
Melvern Lake, KS (con.) 322 1.10
(2.39)
1.36
(4.30)
0.96
(2.38)
1.47
(4.39)
Millwood Lake, AR 323 1.48
(4.82)
0.83
(2.35)
0.98
(4.68)
1.30
(4.60)
T 1.51
10 (8.13)
Mississippi River Pool 324 2.12
No. 3, MN (3.90)
1.01
(1.89)
1.40
(4.21)
0.94
(2.41)
1.32
(4.05)
Mississippi River Pool 325 1.23
No. 6, MN (3.16)
1.24
T+M cost
-0.009
(-1.72)
-0.008
(-1.69)
-0.007
(-1.46)
-0.008
(-1.62)
-0.008
(-3.96)
-0.009
(-4.45)
-0.009
(-4.59)
-0.008
(-3.94)
-0.007
(-3.47)
-0.005
(-4.22)
-0.006
(-4.53)
-0.006
(-4.44)
-0.006
(-4.97)
-0.006
(-4.56)
-0.007
(-4.31)
-0.007
Income
4.8 x 10"6
(0.36)
5.9 x 10~6
(0.43)
5.1 x 10~6
(0.39)
7.0 x 10~6
(0.52)
1.1 x 10~5
(0.39)
2.1 x 10~5
(2.57)
1.7 x 10~5
(2.29)
1.7 x 10~5
(2.03)
1.9 x 10~5
(2.34)
-7.2 x 10~5
(-1.63)
4.8 x 10~6
(0.55)
4.4 x 10~6
(0.50)
3.1 x 10~6
(0.37)
4.5 x 10~6
(0.51)
3.1 x 10"5
(0.88)
1.4 x 10~5
Income
squared Age Sex RECIMP Day R2
0.005 0.07
(0.57)
-0.135 0.07
(-0.49)
0.380 0.09
(1.21)
-0.303 0.08
(-1.02)
1.2 x 10~10 0.25
(0.20)
0.013 0.30
(1.96)
0.691 0.39
(3.41)
0.166 0.25
(0.59)
-0.333 0.28
(-1.50)
1.4 x 10~9 0.38
(1.78)
0.008 0.34
(0.74)
-0.143 0.34
(-0.73)
0.539 0.38
(1.69)
0.036 0.34
(0.18)
-3.5 x 10~10 • 0.22
(-0.51)
0.005 0.23
DF
41
41
41
41
49
49
49
49
49
45
45
45
45
45
66
66
F-
ratio
1.01
0.98
1.42
1.26
5.41
7.10
10.54
5.55
6.39
9.20
7.89
7.88
9.05
7.63
6.34
6.46
DF - Degrees of freedom.
"t-values of no association are shown In parentheses. RECIMP Is a binary variable that Is 1 if the respondent considers recreation
tant. Day Is a birtary variable that Is 1 If the respondent stayed 1 or more days.
to be impor-
-------
Table F-5. (continued)
Site
Site number Intercept
Mississippi River Pool 335 (4.19)
No. 6, NIN (con.) 1 45
(6'.21)
0.98
(3.43)
1.42
(6.74)
Ozark Lake, AR 331 1.53
(5.06)
1.64
(5.25)
1.71
(7.57)
i. 42
(5.11)
Tt 1.80
.!» (9.04)
o
Phllpott Lake, VA 333 1.61
(5-17)
2.26
(6.85)
2.01
(9.05)
1.40
(3.61)
1.92
(10.03)
Pine River, NIN 334 0.19
(0.50)
0.69
(2.69)
T+M cost
(-4.31)
-0.007
(-4.05)
-0.007
(-3.88)
-0.007
(-4.15)
-0.005
(-4.45)
-0.005
(-4.40)
-0.004
(-4.19)
-0.005
(-4.58)
-0.003
(-3.15)
-6.009
(-4.66)
-6.009
(-4.39)
-0.008
(-3.98)
-0.009
(-4.64)
-0.007
(-3.61)
-0.001
(-0.90)
-6.002
(-1.36)
1 ncome
(1.58)
1.3 x 10~5
(1.50)
1.0 x 10~5
(1.20)
1.4 x 10~5
(1.53)
1.3 x 10~5
(0.32)
-8.6 x 10~5
(-0.63)
-1.0 x 10~4
(-0.73)
-7.1 x 10"6
(-0.53)
-2.0 x 10"5
(-1.15)
4.2 x 10~5
(1.10)
-8.6 x 10~7
(-0.006)
-1.5 x 10~6
(-0.12)
5.5 x 10~6
(0.40)
3.4 x 10~6
(0.27)
5.0 x 10~5
(1.64)
-6.6 x 10"6
(-0.95)
Income
squared Age Sex RECIMP Day Rz
(0.78)
-0.074 0.22
(-0.36)
0.537 0.27
(2.08)
-0.040 0.22
(-0.21)
-6.2 x 10~10 0.32
(-0.58)
0.001 0.31
(0.09)
-0.96 0.32
(-0.47)
0.285 0.33
(1.18)
-0.541 0.37
(-2.15)
-1.3 x 10~9 0.39
(-1.22)
-0.011 0.40
(-1.41)
-0.232 0.39
(-1.26)
0.449 0.40
(1.48)
-6.483 0.46
(-2.43)
1.1 x 10~9 0.08
(-1.90)
0.004 0 04
(0.62)
DF
66
66
66
48
48
48
48
48
34
34
34
34
34
71
71
F-
ratio
6.29
8.08
6.25
7.46
7.30
7.40
7.98
9.54
7.28
7.53
7.33
7.64
9.60
2.16
1.04
DF = Degrees of freedom.
"t-values of no association are shown In parentheses. RECIMP Is a binary variable that Is 1 If the respondent considers recreation
tant. Day Is a binary variable that is 1 If the respondent stayed 1 or more days.
to be Impor-
-------
Table F-5. (continued)
Site
Pine River, MN (con:)
Proctor Lake, TX
Sardls Lake, MS
Whitney Lake, TX
Site
number Intercept
334 0.82
(4.51)
0.53
(2.36)
1.07
(3.42)
337 2.13
(8.57)
1.81
(6.57)
1.99
(12.86)
1.94
(7.54)
2.06
(11.79)
340 2.07
(13.95)
1.91
(13.52)
1.84
(18.39)
1.12
(7.57)
1.88
(20.69)
344 1.50
(8.68)
1.40
(8.79)
1.34
(11.61)
1.23
(9.22)
1.83
(14.50)
T+M cost
-0.002
(-1.06)
-0.002
(-1.25)
-0.002
(-1.31)
-0.013
(-6.48)
-0.013
(-7.50)
-0.014
(-7.86)
-0.013
(7.41)
-0.013
(-7.09)
-0.004
(-3.95)
-0.003
1-3.07)
-0.003
(-3.14)
-0.003
(-3.93)
-0.003
(-3.50)
-0.003
(-1.70)
-0.002
(-1.75)
-0.003
(-1.83)
-0.003
(-1.74)
-0.003
(-2.09)
Income
-6.5 x 10~6
(-0.92)
-8.2 x 10~6
(-1.19)
-5.3 x 10~6
(-0.75)
-6.8 x 10~6
(-0.25)
3.7 x 10"6
(0.53)
-3.0 x 10~7
(-0.04)
1.3 x 10"6
(0.20)
1.2 x 10~6
(0.17)
-2.9 x 10~5
(-1.78)
3.3 x 10~6
(0.57)
4.5 x 10~6
(0.81)
4.2 x 10~6
(0.80)
7.5 x 10~6
(1.32)
-7.7 x 10~6
(-0.45)
3.3 x 10~6
(0.73)
2.9 x 10~6
(0.64)
1.6 x 10~6
(0.35)
3.4 x 10~6
(0.81)
Income
squared Age Sex RECIMP Day R2
-0.092 0.04
(-0.20)
0.363 0.08
(1.91)
-0.291 0.05
(-0.99)
1.5 x 10~10 0.54
(0.32)
0.005 0.55
(1.11)
0.273 0.57
(1-81)
0.139 0.54
(0.61)
0.010 0.54
(0.05)
8.6 x 10~10 0.07
(2.18)
-0.003 0.05
(-0.97)
-0.057 0.05
(-0.68)
0.767 0.18
(5.58)
-0.208 0.08
(-2.61)
2.3 x 10~10 0.02
(0.67)
0.0003 0.02
(0.09)
0.160 0.03
(1.50)
0.271 0.04
(2.19)
-0.601 0.15
(-5.61)
OF
71
71
71
48
48
48
48
48
201
201
201
201
201
198
198
198
198
198
F-
ratlo
0.93
2.17
1.25
18.61
19.43
20.89
18.61
18.54
5.13
3.79
3.63
14.38
5.84
1.30
1.15
1.91
2.78
11.81
DF = Degrees of freedom.
8t-values of no association are shown in oarentheses. RECIMP is a binary variable that is 1 If the respondent considers recreation to be impor-
tant. Day Is a binary variable that is 1 if the respondent stayed 1 or more days.
-------
APPENDIX G
ALTERNATIVE REGRESSION MODELS
This appendix provides a detailed listing of the alternative specifications
of regression models. Listings are given for both the survey and travel cost
models.
G-1
-------
Table G-l. Independent variable combinations used In option price, user value, and option value
regressions. Dependent variables are dollar bids given for changes In water quality.
&
i
ro
Sex
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Age
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Education
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Income
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Dummy Bidding
Variables vs. Non-
to Denote Bidding Length
Survey Game of
Version Dummy Residence
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X X
X
X X
X
Attitude
Index1
X
X
X
X
X
X
X
X
X
X
X
Attitude
towards
Cost
Dummy
X
X
X
X
X
X
X
X
X
User
Dummy
X
X
X
X
X
X
X
X
X
X
X
X
X
Dummy
Water variables
Quality to
Rating denote
Dummy* Interviewer
X
X
X
X
X
X
X X
X
X
X
Dummy
variables
to
denote
Pro-
fession
X
X
X
X
X
X
X
Dummy
Variables
to
denote SIC
Industry
X
X
X
X
X
X
(continued)
-------
Table G-l (continued)
Sex
X
X
X
X
X
Age
X
X
X
X
X
Education
X
X
X
X
X
Dummy
Variables
to Denote
Survey
Income Version
X
X
X
X
X
Bidding
vs. Non-
Bidding Length
Game of
Dummy Residence
X
X
X
X
X
Attitude
Index1
X
X
Attitude
towards
Cost
Dummy
X
X
User
Dummy
X
X
Dummy
Dummy variables
Water variables to
Quality to denote
Rating denote. Pro-
Dummy^ Interviewer fesslon
X
X
X
X
Dummy
Variables
to
denote SIC
Industry
1Th1s Index was constructed by adding responses to various attltudlnal questions.
2See question number B-l-b In the survey questionnaire.
O
i
-------
Table G-2. Independent variable combinations used In all 43 outdoor recreation survey sites.
Dependent variable LN (visits).
Site
Site and
1/3 and 1/3 Recrea-
Travel Travel Travel Travel Site tlon- Day
Travel On- and and and and and Impor- Travel
Time Mile Site Mile Mile Mile Mile Travel tance In- Income Day Cost
Cost Cost Cost Cost Cost Cost Cost Cost Dummy come Squared Dummy1 Dummy2 Age
Day
Hour Site
Site Cost
Hour Cost Race Slope
Sex Dummy3 Dummy* Dummy5 Dummy6
X
X
X
X
X
X
X
X
X
X
X
X •
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X X
X
X
X
X X
X
X
X
X X
X
X
X
X X
X
X
X
X X
X
X
X
X X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
. X
X
X
X
X
X
X
X
X
X
X
X
X X
X X
X
x x
X X
X XX
X X
X
X X
X X
X
X
X
X
X
X
X
X
X
X
(continued)
X
X
X
X
X
X
X
X
X
X
X
-------
Table G-2 (continued)
Site
Site and
1/3 and 1/3 Recrea-
Travel Travel Travel Travel Site tlon- Day
Travel On- and and and and and Impor- Travel
Time Mile Site Mile Mile Mile Mile Travel tance In- Income Day Cost
Cost Cost Cost Cost Cost Cost Cost Cost Dummy come Squared Dummy1 Dummy2 Age
Day
Hour Site
Site Cost
Hour Cost Race Slope
Sex Dummy3 Dummy4 Dummy5 Dummy8
O
X X
X
X X
X
X X
X
X X
X
X X
X
X X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X X
X
X
X
X X
X
X X
X X
X X
X X
X X
X
X X
X X
X X
X X
X X
X
X X
X X
X X
X X
X X
X
X X
X X
X X
X X
X X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X X
X
X
X
X
X
X
X
X
X
k
X
X
X
X
X
X
X
X
X
X
X •
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
-------
Table G-2 (continued)
O)
Site
Site and
1/3 and 1/3 Recrea-
Travel Travel Travel Travel Site Uon- Day
Travel On-* and and and and and Impor- Travel
Time M11e Site Mile Mile Mile Mile Travel tance In- Income Day Cost
Cost Cost Cost Cost Cost Cost Cost Cost Dummy come Squared Dummy1 Dummy2 Age
Day
Hour Site
Site Cost
Hour Cost Race Slope
Se* Dummy3 Dummy4 Dummy5 Dummy6
X
X
X X
X
X
X X
X
X
X X
X
X
X X
X
X X
X
X
X
X
X
X
X X
X X
k x
X
X
X
X
X
X
X
X
X
x :
X
X
* ' "' '[•
1 Intercept dummy equal to one If the respondent stays one:-;or more days and zero otherwise.
'Slope dummy calculated by multiplying day by travel cost*
3Intercept dummy equal to one If stayed less than one hour.
4Slope dummy calculated by multiplying hour by site cost.
Intercept dummy equal to one If white and zero otherwise.
8Slope dummy calculated by multiplying day by onslte cost;
-------
Table 6-3, Independent variable combinations used as tailored models for a subsample of the 43
outdoor recreation survey sites. Dependent variable is LN (visits).
Travel
Time
Cost
X
X
X
X
X
X
X
X
X
1/3
On Travel & Travel 4
Mile Site Mile Mile
Cost Cost Cost Cost
X
X
X
. X
' X
X
X
. x
X
X
X X
X
; X
• - • , x
X
X
X x
X
X
X X
- x x
X
x
x -.-
X
X
- x
x
x
• x
x nx
-X-
X . ..
X,
' X x
Recreation
Importance
Dummy
X
X
,
X
X
X
X
, X
.. X
X
X
X
X
Income Day
Income Squared Dummy
X
X
X X
X X
X
X X
X X
XXX
X
t
X
X
X . X
X
X
,
1
X
x
;' v
X
, x
X
Hf f x
x
, x x
,
x .X
x x
x
X , :. • ' '
X X , X
X
X X i
Day Travel
Time Plus
Mile Cost Camping
Dummy ' Age Sex Dummy
X
X
X
X . X
X X
X . X
X
.X , X
X
X
X
X X
X
X
x , •• "
X
X
,x
X
-------
Table 6-3 (continued)
O
oo
Travel
Time
Cost
X
X
X
X
X
X
X
X
Mile
Cost
X
X
X
X
X
X
X
X
On Travel &
Site Mile
Cost Cost
X
X X
X
X
X
X
X
X
X X
X
1/3
Travel & Recreation
Mile Importance
Cost Dummy
X
X
X X
X X
X X
X X
X
X
X
X
X
X
Income
Income Squared
X X
X
X
X
X X
X
X
X
X
X
X
X X
X X
X
X
X
X
X
X
X
X
X
.X
X
X
Day
Dummy
X
X
X
X
X
X
X
X
X
X
X
X
Day Travel
Time Plus
Mile Cost
Dummy i Age
X
X
X
X
X
X
X
X X
X
X
X
X
X
X X
Camping
Sex Dummy
X
X
X
X X
X
1Intercept dummy equal to one Is respondent engaged In camping.
------- |