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.

-------