RECREATIONAL BENEFITS TRANSFER PROJECT
    V. Kerry Smith,  Principal Investigator
     Department of Economics and Business
        North Carolina State University
        Raleigh, North Carolina  27695
           EPA Cooperative Agreement
              Project # CR813564

                      to

Vanderbilt Institute for Public Policy Studies
            Vanderbilt  University
             Nashville,  Tennessee

-------
                                                         950R89O05

                   RECREATIONAL BENEFITS TRANSFER PROJECT


                             TABLE OF CONTENTS


                                                                 Page
Chapter 1.   Introduction and Overview   	   1.1-1.7
            V.- Kerry Smith

Chapter 2.   Signals or Noise?  Explaining the Variation  ....   2.1-2.40
            in Recreation Benefit Estimates
            V. Kerry Smith and Yoshiaki Kaoru

Chapter 3.   What Have We Learned Since  Hotelling's Letter? ...   3.1-3.10
            A Meta Analysis
            V. Kerry Smith and Yoshiaki Kaoru

Chapter 4.   Nearly All Consumer Surplus Estimates Are Biased .  .   4.1-4.15
            V. Kerry Smith

Chapter 5.   Demands for Data and Analysis Induced By 	   5.1-5.72
            Environmental Policy
            Clifford S. Russell and V.  Kerry Smith

-------
        CHAPTER 1
INTRODUCTION AND OVERVIEW
     V. Kerry Smith

-------
                                                                              J.. 1
                           Introduction and Overview






      This  report summarizes  the research  completed  under EPA  Cooperative




Agreement #  CR813564  entitled "Recreational  Benefits Transfer  Project."   The




objective of -this  research was to review all  travel  cost  recreational demand




models completed between 1970 and 1986 from published and unpublished sources,




including all Master's and Ph.D. essays that could be identified and obtained.




From this literature  a subset of  the  studies was assembled for meta-analysis.




The meta-analysis  sought to develop a statistical summary  of the results from




these demand analyses in order to determine  the  influences  of judgmental and




site-characteristic variables on the consumer surplus estimates derived and to




gauge the effect of these variables on  other measures of demands for recreational




sites.   This project  was  funded under the Innovative Benefit Analysis Program




because  the effort was viewed as exploratory.  The primary research activities




were undertaken jointly with Dr. Yoshiaki Kaoru, currently an Assistant Social




Scientist at Woods Hole Oceanographic Institution.  At  the time, Dr. Kaoru was




a. graduate student in the Department  of Economics at Vanderbilt University.




      As  the papers  prepared under this  agreement  indicate,  the research was




quite successful.  Statistical  summaries  were developed for some 77 different




demand  studies  for   recreational  resources,  and  we  compared the   relative




importance of variables  describing modeling  judgments with characteristics of




the recreational sites and the activities  undertaken at  them.  Four papers were




prepared with partial  support  from this Cooperative Agreement.  Two of the papers




describing our approach have been presented at several universities in the United




States,  as well as at Academia Sinica  in Taiwan and at a National Bureau of




Economic Research  (NBER) Conference on Data  Needs for Economic Policy Making.

-------
                                                                           1.2
One of the papers is to be  published  in a volume from the NBER conference.   The




remainder are currently under consideration  for publication.  Two have received




preliminary indications of publication interest,  pending suggested revisions.




      Rather than rewriting the materials developed from the research papers in




an alternative  technical  format,  this report is organized  into  four chapters




following th,is  introductory  chapter,  which  highlights  the  overall conclusions




of the research.  Chapter 2 presents the first paper prepared from the research.




It  describes the  conceptual  issues  associated with  using meta-analysis  to




summarize estimates  of the consumer  surplus per unit of use across  a diverse




range  of travel  cost demand  studies and  summarizes  the  findings  from  our




analysis.   Chapter 3  focuses  on  a subset  of the  studies  used for  the  meta-




analysis  of  per-unit  benefit measures  and considers  the  feasibility  of




summarizing the estimates for other features of recreation demand (such as the




price elasticity).  We used a subset  here because it was not always possible to




estimate these price elasticities with the  information reported in many of the




recreation demand studies.




      Because consumer surplus and price elasticity estimates are themselves




random variables, Bockstael and Strand [1987] have emphasized the importance of




incorporating their properties as estimators  Into policy analysis.  Our use of




the Newey-West  [1987]  adjusted covariance matrix  in evaluating the effects of




modeling  assumptions was  one  reflection of this  influence. Chapter  4 was an




unanticipated byproduct of the theoretical  analysis of  the properties of our




consumer surplus estimators.  It proposes a new estimator for developing consumer




surplus  estimates  and evaluates  it  with   some  sampling  experiments for  a




particular specification of the travel cost  demand model. This  estimator offers




an alternative to the proposal  recently advanced by Adramowicz et al.  [1989] for




cases with unstable consumer surplus estimates.  Chapter 5 places our findings

-------
                                                                            1.3
in a somewhat more general context,  as part of an evaluation of new data needs




for environmental policy making.




      Several overall conclusions emerged  from  our  research activities.   They




can be categorized into three broad areas.








A.    Conceptual Findings




      Our theoretical analysis of the  issues associated with measuring consumer




surplus suggested that virtually  all consumer surplus estimates will be biased.




This follows because they usually involve nonlinear transformation of estimated




demand parameters.  As a consequence of Jensen's inequality,  the consumer surplus




estimates themselves exhibit bias even if the specification for the demand model




is  correct.    Specification  errors  in  demand analysis  simply compound  the




difficulties raised by the nonlinear transformation.  This  implies that general




purpose strategies designed  to focus  on  estimating  demand models that serve a




variety of purposes  or reliance on  the  existing  literature wherein demand




analyses are developed to serve other  purposes  (test hypothesis,  illustrate new




functional forms or estimators, or highlight the  special features of a particular




data  set)   are  not  necessarily  the  best suited  for environmental  benefit




estimation.  These  objectives may not be consistent with deriving the most robust




benefit measures.  While this general conclusion was  probably recognized by most




researchers in this area, to our  knowledge  this  point has not been specifically




made in the literature.




      This point applies not only to the  literature  on travel cost recreational




demand models,  but to all current techniques  in  use  for measuring recreation




benefits, including the  more  recent random utility models, whether based on logit




or nested  logit  specifications.   In all  cases,  the  benefit measures involve a




nonlinear transformation of  random variables, which in itself will induce bias

-------
                                                                            1.4
in the welfare estimates.  -This suggests  that  new  research  in the area should




consider the  implications  of modeling and  estimation  strategies  specifically




designed to accomplish a broad range of benefit estimation tasks.   Research on




the implications of bottom up versus  top  down  estimates for aggregate benefit




measures, as well as on the development of "transferrable" models for measuring




consumer surplus (as opposed  to  the  demand features of recreational resources),




seems highly appropriate.  Equally  important,  as Chapter 4  illustrates,  it is




possible to develop estimators designed to focus on  consumer surplus measurement




instead  of estimation of demand parameters.   While  this  work  is  largely




illustrative,  it  nonetheless  displays  how  the  performance of  alternative




estimation  strategies  can be  sensitive  to  the features of the  true demand




structure and the objectives of the analysis.




      A further set of  conceptual issues resulting from the research arises from




the meta-analysis.  Here we found strong confirmation for systematic variation




in the consumer surplus estimates per unit of use across a wide range of studies.




This  systematic  variation could  be  attributed to both the features  of the




resources involved and the modeling decisions made in estimating the travel cost




demand models  for  these recreational  resources.   Indeed,  the most important




factors we  found bore  a close correspondence  to the  issues identified in the




literature  as  the most  significant questions   in  modeling  recreation demand.




Thus,  the  empirical analysis  provides strong confirmation  for  the  implicit




research agenda  that has evolved in recreational demand modeling.








B.    Empirical  Findings




      The empirical analysis suggest  that it is possible  to  summarize both the




consumer surplus estimates per unit of use  and the own-price elasticity of demand




for recreation sites  across  a wide range of studies.   These estimated models

-------
                                                                            1.5
include variables  for the features  of the recreation  sites,  as well  as  the




modeling judgments  made in  developing each of  the  demand estimates.   After




adjustment for the panel nature of our sample  data set,  the results display a




remarkable   degree   of  consistency   and  robustness   across   alternative




specifications.  While these  are not predictive equations in the sense that they




provide a mechanism  for predicting  the consumer surplus per unit  of use that




would arise for  each  type of recreation site,  they can be used as approximate




gauges of the plausibility of estimates derived from transfer exercises or from




specific studies for  individual sites.  Perhaps most importantly, they provide




a basis  for judging the degree of  maturity in  travel  cost recreation demand




models.  By appraising the relative importance of judgmental versus theoretically




motivated variables,  this type  of  analysis  evaluates  how much  our current




estimates are influenced by factors that arise from a priori theory versus those




which  represent  analysts'  adjustments  to  take account of  incomplete data or




modeling assumptions required for meta-analysis.








C.    Benefit Transfer Findings




      In addition to the first two categories of results,  the analysis also has




implications for the process  of developing transferrable benefit estimates.  The




most  important  of  these  implications is  the  demonstration  that unifying




principles connect quite diverse estimates across modeling efforts and widely




varying  recreation  sites.   Because these  modeling efforts were  undertaken by




different  investigators  at   very  different  times  with  diverse   amounts  of




information, this is reasonably strong support for a set  of unifying principles




connecting  the per unit valuation measures for a wide  range  of recreational




resources.




      The meta-analysis also forces  the analyst to consider the measure used as

-------
                                                                            1.6
the focus of an empirical summary.  We  considered  two  --  the  consumer surplus




per unit of use and  the own-price elasticity  of demand.   Either could provide




the basis for a benefit transfer analysis used in a policy evaluation.




      Analysts have  tended  to  use  a. unit value  approach  to benefit transfer,




treating the model transfer  task as one involving  the  development or transfer




of a per unit-value appropriate to the policy and then dealing with the number




of people and units  of use  affected by  the  policy  as a separate question.   By




forcing the selection of a metric for summary,  meta-analysis has identified that




consumer surplus per unit of use need  not be the  focus for a benefits analysis.




The  early benefit-cost  analyses  of  Harberger  [1971]  and,  indeed,  current




evaluations  of  the   effects  of  cost-reducing  technological innovations  in




agriculture (following Griliches [1957]  early  methodology for  hybrid corn) rely




on point estimates of demand and supply  elasticities.  We could easily consider




the use of a price elasticities/local approximation approach  to estimating the




benefits from a  policy improving access to a recreation site (i.e. where the




change could be viewed as a price change).




      Equally important,  there  is a general issue of how we wish to  prepare these




summaries.  Chapter 2 suggests that for well-behaved demand functions, we have




little intuition  about the  properties of the  consumer surplus per unit to use




in judging the plausibility of differences  in estimates  across  alternative




studies.  Both conceptual and empirical research is needed here.




      Finally, perhaps the most important conclusion for benefits transfer arises




from the inadequacy of the reporting standards used in most published research.




Because this is unlikely  to change  in the  near future, a reorientation in the




research and data acquisition in support of benefit analysis for policy purposes




is clearly warranted.  More specifically, policy offices need to establish groups




that summarize  in a format  consistent  with the needs of a meta-analysis the

-------
                                                                            1.7
findings of  new empirical studies as  they  are available.   By  establishing  a




consistent protocol for these summaries,  it would  be  possible  to request from




researchers at the time their unpublished  or published reports become available




the companion supplementary information needed for meta-analyses.  Usually these




are summary  statistics for the variables used in the  study,  descriptions  of




transformations, sample characteristics,  clarifications, etc.   When the study




is  recent,  this information is  easily available  from researchers,  does  not




require that they furnish their complete data (which may be planned for use in




future research).  It  is  also a more  manageable enterprise.   After a lapse of




time and the completion of the policy task,  these  requests are less likely to




be responded to and, in most cases, the timing does not permit a response.




      As policy analyses increasingly  rely on using research developed  for other




purposes and research available on the  proverbial "research shelf," it is clearly




essential that analysts set up mechanisms  to "define the shelf and maintain it."




With limited resources and an increasing number of policies  to be  evaluated, EPA




and other mission-oriented agencies have  concluded that they cannot afford to




support research that does not have an  immediate policy relationship. This means




they must choose the most  important questions  for these  limited investments and




rely on information from  the performing community  for all  the rest.  While an




understandable response,  it reinforces the need  for research on how to archive




what  is being  done   so  it can  be  systematically  used  for  future  policy




evaluations.   A meta-analytic  approach  forces  the systematic  collection of




information as it is developed.  Examples  of its use for policy issues (outside




economics) are now making  the popular press.  For example, the July 1, 1989 issue




of The New York Times reported  the results of  a study  indicating  a  narrowing in




the traditional differences in verbal and mathematics aptitude scores between

-------
                                                                            1.8
men and women.  It was based on a. meta-analysis  of different researchers' studies




of these groups' test performances over a number of years.




      Increased availability  of  data,  the extensive  increase  in  contingent




valuation surveys for a wide range of environmental resources (see Mitchell and




Carson  [1989]), and enhancements  in micro-computing  together make this task a




reasonably straightforward data management effort.  Without this effort, benefits




transfer will  remain.a  haphazard  and last-minute  enterprise that is not fully




informed by available research.  As such, it progressively will lose professional




credibility and fail to systematically learn  from past experience.

-------
                                                                            1.9
                                  CHAPTER  1

                                  REFERENCES
Adramowicz, Victor  L. ,  Jerald J. Fletcher, and  Theodore  Grahara-Tomasi.  1989.
      "Functional Form  and the  Statistical  Properties of  Welfare  Measures,"
      American Journal of Agricultural Economics (in press), May.

Bockstael, Nancy E.  and Ivar E. Strand,  Jr.  1987.  "Regression Error and Benefit
      Estimates'^ " Land Economics 63 (Pebruary) :  11-20.

Griliches,  Zvi.   1957.   "Hybrid Corn:   An  Exploration  in  the  Economics of
      Technological Change," Econometrica 25: 501-522.

Harberger,  Arnold C.   1971.   "Three Postulates  of Applied Welfare Economics:
      An Interpretive Essay," Journal of Econometric Literature 9 (September):
      785-797.

Mitchell, Robert Cameron and Richard T. Carson.  1989.  Using Surveys to Value
      Public  Goods:   The  Contingent Valuation Method  (Washington,  D.  C. :
      Resources for the Future).

Newey, Whitney K. and Kenneth  D. West.  1987.  "A Simple,  Positive Semi-Definite,
      Heteroskedasticity  and Autocorrelation Consistent  Covariance  Matrix,"
      Econometrica 55 (May):  703-708.

-------
                 CHAPTER 2
SIGNALS OR NOISE?  EXPLAINING THE VARIATION

      IN RECREATION BENEFIT ESTIMATES
              V. Kerry Smith
                    and
              Yoshiaki Kaoru

-------
                                                                          2.1




                                                                June 15,  1989
  Signals  or Noise?   Explaining  the Variation  in  Recreation Benefit Estimates






                      V. Kerry Smith and Yoshiaki Kaoru*








I.    Introduction




      This paper proposes a new method for taking stock  of what we have learned




about the benefits users derive  from  environmental resources. Our approach uses




econometric  methods  to  review  the  literature.    While we  have  applied  this




approach to one class of benefit estimates--empirical studies using the travel




cost method to estimate the demand for specific recreation sites, it has general




relevance for gauging what has been learned by empirical research in many other




areas of economics.1




      The research landscape for benefit estimation has  changed dramatically in




the ten years since Freeman wrote his  influential  overview of the field.  Freeman




described the motivation for his book as a  response to  a gap in the literature




on benefit  estimation.   As he  noted,  by  1979 there  had been "...substantial




research effort devoted to developing a rigorous  and unambiguous definition and




measure of  changes in  welfare at the theoretical  level..." but "...relatively




little  concern  for  translating the  theoretical  concepts and definitions into




usable, operational empirical techniques" (p.  15).  This situation has changed,




especially  for applications  in the  United  States.    Of  the  77  travel  cost




recreation demand studies analyzed in this  paper,  61 were prepared since 1980.




Mitchell and  Carson  identified  over  120 contingent valuation studies, most of




which were completed after 1980.  A similar pattern emerges for hedonic property

-------
                                                                          2.2
value  studies:  • of  the 35  including  information on  air  pollution,  30 .were




available  after 1980.   Certainly the  increased  role  given to  benefit-cost




analyses for evaluating environmental policies  in Executive Order 12291 (issued




in February 1981) has contributed to the dramatic expansion in this literature



(see Smith  [T98.4]  and Office of Policy Analysis, U. S.  EPA for evaluations).




Nonetheless, the available benefit estimates fall  short  of what is needed for




an increasing array of policy related activities (see  Ward and Loomis; Naughton,




Parsons, and Desvousges).  Indeed, the  practice of adjusting the results from




one or more existing studies for a specific type of environmental resource and




using  them  to value  changes  in  another  resource has  become a growing area for




research.   Labeled as "Benefits Transfer," this process usually involves two




steps:   (1)  adjusting or transferring an estimated model  (or set of per unit




benefit  estimates)  from the situation where  it was   developed  to the  new




application; and (2)  developing an aggregate estimate for the relevant population




from per unit estimates and other assumptions.  While judgment plays an important




role in both steps,  it  has been the principal  basis  for the first step.  Many




of the published sources used for benefit estimates in policy analysis were not




designed to  provide  measures of the benefits for  a  change in the quantity or




quality of  a resource.   Rather,  they  were developed to  introduce a new model,




test  a hypothesis,  evaluate  the implications of  specific assumptions,  or




illustrate a "new"  estimator.   Consequently,  they must be adapted for benefit




measurement.  The  nature of  these modifications depends upon both the benefit




estimation task and  the  information reported in the original sources.




       Our findings show a systematic relationship between the estimates and the




features of the empirical models.  We  found that  both the type of recreation




site involved and  the assumptions made  in developing the empirical models were

-------
                                                                          2.3
important to the results.  We classified the variables used to describe models



according  to  whether  they  attempted to  reflect specific  theoretical  issues




associated with individuals'  recreation  decisions or analysts' judgments needed




to estimate a model (e.g., selecting a functional form for the demand model or




making  assumpri-ons  to compensate for inadequate data).    Ideally,  the  latter




variables would  not be  important  determinants  of  the  variation in   benefit



estimates.  We found that they are.




      The specific  factors  found to be significant determinants of  the  real




consumer  surplus  per  unit  of use  have direct  implications for research on



households' recreation decision-making;   for further uses  of the travel  cost




demand  model;  and  for the  practices used in transferring  benefit estimates




derived  from  this  class  of models  to  new  applications.   We  describe these




implications in the last section of the paper, after developing the background




for  this  approach in  Section  II  and describing the  data set as well  as our



results in Section III.








II.   The Role for Statistical Methods  in Developing a Research Synthesis




A.    Background




      The use of statistical methods to  develop a research synthesis has a long




history.  Most  of these applications have  involved controlled experiments in




psychology, education, or the health sciences. They have focused on consistently




aggregating the results from different controlled experiments.  In these cases,




the methods are motivated by the desire to avoid the subjective nature of most




research reviews.   At best,  the  conventional literature review summarizes the




presence or absence of statistically significant effects  and,  in some cases,




compares  the  size  of estimated effects.   While many  of  these  studies  have

-------
                                                                             2.4
attempted to draw some "bottom line"  conclusions about what is  known (as  Light




and  Pillemer observed),   these appraisals  often  violate  simple  statistical




principles  in distilling  an admittedly complex  array of work.  Moreover,  to




develop this type of summary,  the reviewer usually must adapt the multiple (and




often complex) features of  the studies  to fit  some comparable  format in  order



to propose a consensus judgment.




      Because  empirical   research  in  economics  is  usually  not  based  on




experimental data and may well report multiple models applied to a single data




base, our proposed methodology is different from that used in most meta analyses




(see  Cordray).    It must reflect  both the  modeling judgments  (made  because




controlled  experiments are  usually  impossible)  and  the  interdependent  panel




nature  of any sample of  research results.   .Fortunately,  both  issues can be



addressed with existing econometric methods.




      Moreover,  the rationale for using an econometric framework  for synthesizing




the benefit  estimates for environmental resources  is more  general.  Empirical




models  are  combinations  of  prior  theory and analyst judgment.   That judgment




combines at least four elements:  the problem or issue the empirical model seeks




to  address  (e.g.,  test  a  hypothesis  or  estimate  a  specific parameter or




quantity); the economic theory of behavior  assumed to be relevant to  the problem;



the  data  available to estimate the  model;  and  the  learning that accompanies




evaluating the joint effects of functional  specification,  variable construction,




and  the results  from prior  model  formulations in relationship to the existing




literature.




      The last of these, sometimes referred to as specification searches or  data




mining, has been widely criticized in the  recent econometric  literature.  We do




not  intend  to  repeat that discussion here.    Rather,  by viewing  models as

-------
                                                                          2.5
approximations, we have  further motivation  for using statistical summaries of

the results from existing models  to evaluate the  importance of such compromises

for the findings.



B.    A Simple Model for Describing Recreation Demand Structure

      The  travel  cost  recreation demand model can be described as  a derived

demand for a  recreation  site  that contributes to each individual's production

of a recreational activity providing utility  (see Deyak and Smith or Bockstael

and McConnell).  As a rule,  the specification for these models has been largely

a semantic exercise to assist  in  isolating  the relevant arguments for a  travel

cost demand model.2 We propose using this framework to describe  the components

of  modeling  decisions  that  may explain  the variation  in  consumer surplus

estimates across travel cost demand studies.

      Consider a simple  utility  function specified in terms of  the activities

a person wants to consume, Z,'s,  as in equation (1).



           U - U(Z,, Z,	Zt)                                    (1)



      Each Z, is assumed to be produced by  combining market goods, x,,'s; time,

t;  and  non-marketed commodities, y^'s,  as  in equation  (2).   Of course,  some

activities may not use some inputs.



            Z, - £,  (x,,,..., x,,,  t,, y,	  y.,)                (2)

            where x,,,... .x,,,   -   the amounts of the n marketed commodities used
                                  in the production  of Z,.
                          t,  -   the amount of an individual's time used in the
                                  production of Z,.

-------
                                                                          2.6
                  YD.--- Ymi  =   che  amounts  of the  nonmarketed
                                 commodities  used in the
                                 production of Z,.


To formally derive the implications of  this model  for travel cost demand models

we need  to  specify an individual's  budget and  time constraints  to individual

decisions.

      With  each movement  away  from this  fairly general  description  of  the

household's choice problem, the analyst imposes more structure on the problem.

This structure  can arise  from observing how  households make decisions or from

introspection.   Assumptions  about the  constraints  or  features of the utility

function can also focus attention on specific aspects of decision-making because

these assumptions are  considered to be important  to the problems being addressed.

Finally, in most cases, available information dictates a set  of compromises that

defines the structure of the model.

      Developing a set of hypotheses for the factors  that might  influence benefit

estimates  from  travel  cost  models  involves consideration of five  types  of

decisions:

      (1)   specifying the types of recreation  sites;

      (2)   defining  a recreation site, its usage,  and the site quality;

      (3)   modeling  the opportunity cost of  time:

      (4)   describing  the role  of  other sites  in producing the recreation

            service flows;

      (5)   linking  the specification  of  the  demand model  to  an underlying

            behavioral model.



We use this general specification to consider how the answers  provided for each

-------
                                                                          2.7
issue affect one or more aspects of existing travel cost studies.


      Describing the reasons for variation  in consumer surplus estimates across

sites requires  us to  consider  the rationale for  all  economic models.   Most

economic models assume that individuals share common behavioral functions with

constant parameters, except for a  set  of distinguishing features  (such as age

or education).   This perspective implies that individuals have the same demand

function for a commodity or service.   However, it recognizes  that  price and

income differences, as well as differences in demographic characteristics, can

lead  to  differences in  the actual  quantity  each person  will demand  of any


specific commodity.

      In principle,  the  same  argument applies  to  the measure  we  have used to

summarize the travel cost demand estimates  across studies--the consumer surplus

(CS) per unit  of use  (v).  Unfortunately,  conventional theory does  not offer

clear guidance on the  properties we might expect for this measure, given well-

behaved demand functions.   This is easily  seen by describing it more formally


in terms of  a  demand function,  say g  (P,  I, d, q), with P the travel cost, I


the income, d demographic or taste variables,  and q quality measures.  CS/v can


be defined formally by (3):


                                   PC
      CS/v - L(P0> Pc,  I, d, q) - /   [g(p,  I, d, q)/g(P,, I,  d, q] dp  (3)
                                   P.

            where P0 - current price
                  Pe - choke price



The estimates of consumer surplus from  the literature are generally for specific


sites or derived from regional travel cost  models hypothesized to  describe sets


of  sites  in  the  same geographic  region.   To estimate  CS/v  requires some

specification  of the variables hypothesized to influence  L  (.).   We used the

-------
                                                                         2.8
information reported in each study to estimate the consumer surplus per unit that




was representative  for a typical user of  the  site  and sample relevant  to each



model.








1.    Types gf-Recreation Sites




      The first issue implies that we need a way to define  the different types



of services provided by different recreation sites.  Moreover,  the classification




cannot stop here.   An individual's valuation  of  a  site's services  will depend




on how these services are used.  The household production framework recognizes




a site  demand as a derived  demand.  Thus, both sources  of  variation  must be




considered. Unfortunately, our experience with such taxonomies is quite limited.




      Clawson and Knetsch classified recreation  sites into three categories--




user-oriented, intermediate,  and resource-based. The first type of site included




city and county parks, golf courses, tennis courts,  swimming pools, playgrounds,




etc.  Intermediate sites were federal and state reservoirs and parks that provide




hiking, camping, fishing, boating, and hunting.  The last category had national




markets because their physical characteristics  were important to the recreational




activities  they supported.   In the Clawson-Knetsch taxonomy, these attributes




contributed to the fishing or hiking activities in ways that cause recreationists




to perceive these activities as distinctive from the same activities undertaken




in state parks.




      Our specification attempts to reflect the Clawson/Knetsch perspective, but




is  forced  by  each  study's  site description  to   be  fairly  rough.    Our




classification  allows a  site to satisfy more  than.one  feature simultaneously.




A  site with  a  lake may simultaneously be a  state  park  allowing  hiking and




camping.   We  have  also attempted to identify the  primary activities analysts

-------
                                                                          2.9
indicated were  associated with   each  site.   While some overlap is inevitable,

the association between them is not perfect.

      Equally important,  these activity variables may also reflect the influence

of  analysts'  comparative  evaluations   of   the   consumer  surplus  estimates.

Evaluation of.empirical models  can involve comparing a model's consumer surplus

estimates with  results.from the past literature  to gauge their plausibility.

Because most of the commonly accepted estimates of per-unit values have been for

recreational activities,  these  variables' contribution to our models also may

reflect the effects of informal screening rules for model selection.  Examples

of the activity-based sources for recreation value estimates include the Water

Resource Council  estimates  of unit  day  values,  the Sorg/Loomis review for the

Forest Service RFA process, and (most recently) the Walsh et al. update of the

Sorg/Loomis precis of the benefits per day of specified  recreational activities.



2.    What is a Recreation  Site and How Do We Measure the Use of it?

      The early travel cost literature treated sites as well-defined entities.

Because  the  travel  cost  model  arose from  Harold Hotelling's  suggestion to

consider the visitation patterns from concentric zones  around a specified site,

this can hardly be surprising (see also Clawson).  More recently,  in applications

to marine  recreational fishing in  areas  with a  large  array of similar sites

(e.g., estuaries)  or where  policy requires  a coordinated treatment of a large

number of similar sites (e.g,  the effect of  acid rain on the Adirondack Lakes),
;
the definition  of what a  site  is has  been less clear-cut.  In response to the

difficulties  posted by  developing separate site  demand models  under these

conditions, several studies  pool data across  sites, arguing that their parameters

were  approximately constant (Sutherland [1982b]) or that site characteristics

-------
                                                                          2.10
could be explicitly incorporated into  the model  (Vaughan and Russell;  Smith and




Desvousges  [1985]).




      The variables used to measure an individual's demand  for a site's services




are also important in distinguishing the available models.  This specification




is another example of a  decision where an a priori selection of a "best" measure




is not always apparent.  What  is  apparent,  however,  is that  price measurement




must be  coordinated  with quantity measure.  Some  quantity measures  can imply




nonlinearities in  the individual's budget  constraint.   Defining use  typically




involves two considerations--the treatment  of on-site time per trip and the time




horizon  for decision making.   From  the perspective  of  a  season,  if ytl  in




equation (2) represents  the use of recreation  site k,  we  might ask if the use




of this site is to be measured as total time at the site or if trips and time-




on-site per trip should  be distinguished.  For many activities, the "production"




of a day of recreation  is comparable to that of  a longer stay.   Longer trips




simply allow more of the activity  (service flow) to be  produced.   For other




activities, this  is  not a reasonable assumption.   Price  per unit of use will




have both  fixed  and  variable  components if use-per-trip is not held constant.




Thus, the  measure  selected for quantity will be important to the existence of




a conventional Marshallian demand function.1  On the basis of these arguments,




we define  variables that describe  the  measurement of  use  (i.e., days versus




trips) and the treatment of on-site-time in the models.









3.     Opportunity Cost of Time




      There are  a variety of potential  specifications for the constraints to




household  utility maximization--technology, income, and time constraints.  The




full income concept, following Becker's original usage, links time and monetary

-------
                                                                          2.11
constraints by defining income in terms of earnings and other sources of income.




Time is assumed to be  freely substituted  in any  use, so  all  uses  of time have




the same  opportunity  costs  (the  wage rate).   Alternatively,  we might specify




different  opportunity costs,  using  the  wage  rates for part-time work (see




Bockstael, Strand, and Hanemann) .   Yet another possibility specifies different




time constraints and maintains that not all types of time can be substituted (see




Smith, Desvousges, and McGivney) .




      Each formulation will 'have  quite different implications for the implicit




price  estimated  for  the  use  of  a  recreation  site.    In this  example,  use




corresponds to one trip to the  site.  In general, an individual's implicit price,




k, to use a recreation site for a fixed amount of time would be defined as:
      k - cd£ + Atf£                                                (4)




      where:  d£ = round trip distance to site £.




               c - vehicle operating cost per mile




              i - travel time for one round- trip  to site £




               A - shadow price for travel time








This  implicit  price  would   vary by  the  location  of  the  individual  and,




potentially, by whether  vehicle  costs  were  shared.   In the full income model,




A is the wage;  the Cesario/Knetsch proposal treats it as a fixed fraction of the




wage;  the Bockstael  et al.  framework maintains  that A  will depend  on the




definition of the marginal time unit for each person  and by the degree of  control




he (or she) has over time allocations.  In the different "types of time" model,




A becomes  a nonlinear function  of  the wage rate and  other parameters  of the




individual's decision  process.

-------
                                                                          2.12
      With detailed  information  on  the time constraints, wage  rates,  and job




opportunities  for individuals,  it  would be  possible  to  test these  models.




Unfortunately, the available information generally falls short of implementing




any of these  frameworks.   In  fact,  the early  models  based on origin zone data




preclude serious  consideration of any  of  these approaches.   Consequently, the




literature offers a selection of approximations.  Because wage rates are often




unknown, they must be estimated.




      The time horizon relevant  for  decision  making  is itself an issue.  This




has become  especially relevant to comparisons  of  recreation models developed




using a random utility framework.  In several cases, these models seek to explain




decisions on  single  trips, as  if each decision  was independent of  what has




happened  earlier.   This  formulation  implici.tly  compresses  the  time horizon




underlying a model of individual choice, .because in most instances it describes




the problem from  a single-trip perspective.  Opportunity costs must be treated




differently in this  context,  because  the choices for time  uses may be more




limited with  this compressed decision horizon.




      Our measures of site usage and individual time allocation decisions are




exceptionally  limited.   Because  of these limitations,  analysts  have usually




proposed informal rules,  such as maintaining that  opportunity costs  are between




one-fourth and one-half the level of the wage  rate (Cesario  and Knetsch).  Our




analysis defines variables that describe how past studies measured the  wage rate




and how they  described the opportunity  cost of travel  time.









4.    The Treatment  of Substitute Sites




      On  theoretical grounds,  we have  little to debate about the relevance of




substitute  prices  for  modeling  the   demand  for  any  commodity,   including

-------
                                                                          2.13
recreation sites.   However,  this  is not  the  issue  that must be  addressed  in




implementing a  travel  cost recreation demand model.   As a rule, micro  level




surveys include information on the respondents' judgments about their "next best




alternative."4   Thus,  in practice,  the  issue of  including substitute prices  is




not clear-cut..'.  It requires  determining  what sites  are actually available and




how potential users perceive alternative  sites.   As  Rosenthal [1987] observed,




collinearity between price measures can yield the appearance of a small role for




substitute prices.4  For the  most  part, past efforts can be grouped  into three




alternatives:  (a)  excluding any consideration of substitutes (and  this has been




the majority of the work); (b)  formulating arbitrary  indexes of  existence  of




substitutes using a diverse array of specifications (each with little connection




to micro theory); and (c) including a selection  of substitute prices.  Based on




this diversity in practice, we have  defined  a variable  to reflect the treatment




of substitutes.








5.    The Behavioral Framework and the Empirical Model




      The specification of any estimating model introduces implicit restrictions




that affect how any sample of actual choices is described.  Economists working




with recreation demand modeling are beginning  to question how  these implicit




restrictions should be selected.    For  example,  Kealy and Bishop,  Bockstael,




Hanemann, and Strand, and other authors argue that these specifications should



follow  from  a well-defined behavioral model,  based on a  specific  functional




specification for either the direct or the indirect utility function.  From these




authors' perspectives  the  "leap of  faith" that  often separates  the  theory and




empirical sections  of  applied papers  is  inappropriate.  A contrasting view of




the process might  suggest  that  because our  information is incomplete, we have

-------
                                                                          2.14
no  reason  to believe  a  complete behavioral description  will  be better  than




starting with a "reduced-form" approximation.




      An examination  of  the  results  from existing studies cannot answer  this




question, because  we  do not  know the truth.   Nonetheless,  by  examining  the




influence of the. demand specification for the consumer surplus estimates, we can




determine whether an  answer is  important.   To  examine  the importance  of these




types  of judgments,  we  grouped the variables used to  describe each  set  of




estimates  into  two classes--one set  reflecting  different  (but economically




plausible) maintained hypotheses and a second describing analyst decisions where




either  the  economic  theory does not provide  guidance or limitations  in  the




available data require assumptions.   By testing whether the second set provides




significant  determinants of  the consumer surplus estimates, we  can gauge the




importance of these more arbitrary modeling decisions.








III.  Results




      Our analysis  is based on  a review  of published articles  in a wide array




of  journals  that  included  travel cost demand  models,  government reports,  and




unpublished  papers,  as well as  Masters  and Ph.D. theses  from  1970-1986.   We




identified  the  studies by surveying all issues of the  relevant journals;  by




contacting economists who have  developed travel cost demand models, government




agencies (e.g., the Fish and Wildlife Service,  Office of  Policy Analysis in the




Department  of  Interior,  Forest  Service  Regional  Offices  and others)  and the




chairpersons  of  departments  of  agricultural economics and   economics  for




unpublished  papers and  graduate  student  Masters  and  Ph.D.  essays;  and  by




reviewing the University of Michigan microfilm  listings for the abstracted Ph.D.




dissertations in resource economics.  We have attempted  to exclude double entries

-------
                                                                          2.15
for unpublished Ph.D.  theses and subsequently published articles.




      We  have  reviewed approximately  200  studies  to  determine  if they  had




empirical  estimates  for  travel cost  recreation demand  models and  provided




sufficient information to  estimate  the Marshallian consumer surplus per unit of




use.   The  results  reported here  relate  to  77 studies  with  either  benefit




estimates  or  sufficient  information to derive  them.   The Appendix  lists  the



studies and the range of consumer surplus estimates in real terms for those with




sufficient information to  be included in our final empirical models (columns 6,




7, and 8 in Table 2).   Using all 77  studies, there are 734 observations for our




analysis.   However,  as  we discuss  further  below,  there  is not  complete




information on all  variables.    Several studies are  responsible  for multiple




observations because  they reported results that varied:    the  demand models'




functional form; the maintained assumptions; estimators; and definition for the



recreation sites. Consequently, our sample resembles a panel data set and this




feature must be reflected in how we analyze these data.




      Our empirical model hypothesizes  that the variation in benefit estimates




arises  from the  theory  underlying these demand  analyses  together with  the




practical issues that we identified earlier to be addressed in implementing it.




The  variables  used to  explain  the  estimates  of  benefits  can be classified




according  to  features  implied  by:  the  assumptions  inherent in the behavioral




model underlying  the travel cost framework, including the definitions for the




measures  for quantity  and own  price,  as  well as the treatment of substitutes




(designated here by a vector of variables, X*);  the  specifications used for the




estimated  demand  function (designated by  a vector  of variables, X,,);  and the




econometric estimator used  for  the model  (designated by a vector, X,).




      Equation (3) above  defined consumer surplus  per unit  of use for a given

-------
                                                                          2.16
recreation site.  In formulating hypotheses concerning the effects of each class




of variables on estimates of CS/v across studies, it is important to recognize




that the features of each recreation site  (Xs) and the recreational activities




undertaken (XA)  should influence  the true value for consumer surplus per unit.




Moreover, differences in the assumptions made for the variables in L( . ) across




studies will contribute  to variations in estimates of CS/v. Assuming differences




in these specifications for other economic and  demographic variables  are not




important,  the true surplus  might be hypothesized to be a function of variations




in Xj and X,  as  in equation (5) below.   The  only specification for the demand




function which satisfies this condition  is the semi-log form.  If b designates




the absolute value of the price coefficient, then 1/b is often used as a measure




of the Marshallian consumer surplus per  unit  of use  (i.e. depending on how the




quantity variable for the model is defined).7




      To the extent economic and demographic  assumptions  are  greatly different




for the same type of site across studies,  then we  would expect a0 to vary with




them.   Equation (5) assumes that each type of site and primary activity can be




classified into the categories identified by  the sets  of  variables included in




Xj and XAI with  the  subscript i used to designate each  estimate.









                         (CS/v)T1 - o, + O.X,, + aA XAI                (5)
(CS/v)T is measured per unit of use to reflect differences in the  conditions  of




access across studies.  This formulation implicitly assumes the average consumer




surplus per unit of use should be  comparable  (for the same types  of resources,




uses, and  individuals) when the conditions  of access are comparable.



      Estimates of  (CS/v)T will be functions  of  demand parameter  estimates,  as

-------
                                                                          2.17
well as  the variables  determining individual demand.  Because these estimated

parameters can be  shown  to  be  functions of the true values of the parameters,

it is reasonable to hypothesize that the estimated consumer surplus per unit  of

use, (CS/v)e is some function of (CS/v)T.  Our proposal for summarizing empirical

work implicitly" maintains  that there are more factors involved--the variables

describing each study's maintained behavioral assumptions  (Xa) , as well as  each

analyst's judgments  (Xo and Xe).   Equation (6) hypothesizes that  these effects

are additive  influences  to  the true value and therefore would be reflected  in

the bias in any estimator for  (CS/v)T.  Linearity is a simplification.

      Equation (6) has  no intercept because we hypothesize that there is no fixed

bias, independent  of the modeling  assumptions,  in the estimates for  consumer

surplus  per unit.   The fixed bias  will  depend on the model used.  Of course,

variables may well be omitted,  but these are more likely reflected in the error

term, e,, because they can be expected to vary with each study.



      (CS/v),, - /3(CS/v)T1 -I- 7Z, + .e,                      (6)
      where Z, - a vector of variables describing modeling decisions
            (i.e. Z, - (X,,  XB,  Xt,) with 7 a conformably dimensioned vector of
            parameters

            «, - stochastic error
Substituting  (5)  into (6) we  have the basic  form of our estimating  model in

equation  (7).

                       (CS/v)(l  - 0a0 + fia.Xu + 0aAXA, + tZt  + «,        (7)
Under  ideal  conditions  ft would be unity.

-------
                                                                          2.18
      An  important  byproduct of an  attempt  to model  the  results of  applied



economic  research  is  the  development of hypotheses for  the  components of Z,.



This  process requires  reconsidering  the  logical  structure that  we  assume




describes the development  of  economic models.   While some progress has been made




in  macro-economic,  time-series  applications  (see  Hendry),  few findings  are




available  to use  for  applications  to  environmental  resources.   Thus,  our-




discussion will be an informal  first-step toward the more comprehensive efforts




required  if  we  are to  use meta-analysis in evaluating and  improving applied



economic methodologies.




      Following Hendry, if we regard any economic model  as a strategy determined




by the problem at hand and the  information available, then we can be reasonably




confident that  some  elements of modeling decisions (such  as  the treatment of




substitutes or  the specification of  the  opportunity cost of  time)  should play




a role  in the  "true"  demand function for a  recreation  site.   But  we cannot




specify in advance which of  the available assumptions is correct.  Moreover, we




may expect this judgment to  change depending on the application.  Thus, we can




distinguish studies that fail to  recognize these factors from  those that do, but




we cannot specify a best strategy for each case.




      In applications of meta-analysis in other disciplines, these judgments are




used to develop quality weights. These weights are  applied to the results from




each study as part of the development of the statistical aggregate.  We  have not




used this approach in our econometric analysis for two reasons.  First,  and most




importantly, the correct treatment of these modeling judgments in a statistical




summary depends on whether we  believe  they  affect the bias or variance in the




estimates.  Weighting  implicitly assumes that the  estimates based on incorrect




modeling  judgments remain unbiased but simply have less informational content

-------
                                                                          2.19
(i.e. have higher variance).  While this may well be -appropriate for summaries




of studies involving primarily controlled experiments,  it does not seem as clear-




cut for economic applications.




      Second,  because  several decisions can  be  identified as  reflections  of




specific maintained hypotheses in each study, weights for each (even if the first




issue favored weighting)  require a set of subweights for each of these decisions.




We do not feel this is possible given our current level of understanding of how




people make recreation decisions.  Indeed,  our empirical analysis provides the




first evidence on how influential these judgments are for the existing estimates.




      With this background, we can distinguish variables that are largely data-




based decisions  where there  is  little  guidance available  in  economic theory




(those in Xo and XE) from those that are based on theory (X,).   By testing the




influence of the former on our statistical summaries, we can provide some direct




evidence on the role of these types of decisions on the existing estimates.  From




the perspective of transferring model results, we would prefer that these types




of decisions had a small  role in explaining the (CS/v)  estimates for comparable




recreation sites.




      Table 1  defines the  specific variables used in our analysis.  (CS/v),, is




measured by the real (constant dollar) consumer  surplus per unit of use. As one




would likely expect, most of our variables  are qualitative.  Because (CS/v),, is




derived from empirical models based on quite  different data sets and precision




in estimating the parameters relevant  to  the estimation of the consumer surplus,




it is reasonable to expect heteroskedasticity.  Indeed,  as  Bockstael and Strand




observed, it should be possible to estimate the variances in these estimates for




the consumer surplus.  There are two potential problems with implementing this




approach.   First,  the  information routinely reported in  travel  cost demand

-------
                                                                          2.20
studies is generally not  sufficient to construct  approximate  estimates  of  the




variances  for  the  (CS/v)  estimates.   Second,  and equally important,  recent




sampling experiments and bootstrap calculations indicate  the approximations used




in constructing these estimates can themselves  be subject  to  important  errors



(see Smith [1989]  and Kling and Sexton).




      The panel nature of our data set introduces another  source of non-spherical




errors.    If,  for  example,  we assume  a  simple  random  effects model,  then




autocorrelation will be present.   In  this case,  it arises because  there is a




common error shared by results  from different models  reported within the same




study.  In principle, we might also want to distinguish (in the formulation used




for the error process) whether the different estimates reported for each study




reflected different modeling assumptions for the same site,  the same basic model



applied to different recreation sites,  or  some combination of  these effects, as




might be present in the regional travel cost models.




      An  estimator  that  accounts  for  the composite  effects  of all  of these




factors would  require  imposing  considerable  prior information to estimate  the




relevant variances and covariances  for  the estimates of  (CS/v)e, across studies.




To  avoid  imposition  of  largely  arbitrary  assumptions,  we  have  adopted an




alternative strategy--estimate equation (7) with  ordinary least squares  (OLS),




but report the Newey-West version of the White consistent covariance estimator




for  OLS  in  the  presence  of  heteroskedasticity  and  a  generalized form of




autocorrelation.*    As  the results  in  Table 2  indicate, our basic conclusions




are largely unaffected by the  standard errors used in tests  of the effects of




individual variables.



      Table 2  reports  our estimates for several alternative  models describing




the factors influencing the  real consumer  surplus.  The numbers  in parentheses




below  the estimated  coefficients  are  the t-ratios  calculated with the  OLS

-------
                                                                          2.21
standard errors, while  those  in brackets are the  t-ratios using  the  standard




errors from  the  adapted White consistent  covariance  matrix.   Eight models are



reported to illustrate different aspects  of our summary. The first three ignore




the role of  recreation activities and  focus  exclusively on either assumptions



variables (column (1)) or  the variables describing the type  of site (column (2))




or both (column (3)).  Column  (4) expands  the analysis in column (2) to include




the primary recreational activities  supported by  the  site.  Columns (3) and (5)




treat  the  definition of  site  type and primary  recreational activities  as




alternative proxies for the same effects, and include one of the two sets with




the other variables  describing  the  modeling strategies.   Columns  (6)  and (7)




report  our    most  detailed model  (6)  and  the  same  model omitting  only the



variables describing assumptions derived largely from data-based judgments.  The




last  column  offers an  alternative  to  our most  detailed  model,  deleting the




variable for  the year of the data used in the study.




      The variable "Year" was considered to evaluate an interesting suggestion




made by an anonymous reviewer of an earlier version  of this paper.  This reviewer




suggested that we  might be able to  investigate whether recreational resources




were  growing more  or less scarce by including this type  of  variable.   Under




ideal conditions,  this is an intriguing possibility.  However, we believe this




variable serves primarily  as a  proxy variable  for the composite of changes in




the types of data,  estimators, and methodological advances that have taken place




over  the  tine  period spanned by  our  review.    These factors cannot  be




distinguished from the relative  value (comparable to a relative price) one would




like to evaluate for the scarcity issue.   We  report as "final" models equations




that include all types of effects with  and without  the year variable.  However,




we believe  that  column (8) is  probably  a better overall description (despite

-------
                                                                          2.22
the statistical significance of year) because of the consistency in the parameter




estimates with other less complete models and  the  quite  consistent  pattern  of




change in  the  variables  describing each study's characteristics when year  is



included.




      Our results have implications for three types of questions.  First, because



the studies we reviewed span a period during which the conceptual models,  data



sets, and  estimators for recreation demand analysis  improved, we can evaluate




the implications of a wide range of modeling judgments for  consumer surplus (i.e.




CS/v) estimates.   Second,  the studies  considered also  include an array  of




different types of recreation sites. This permits an evaluation for the relative




importance  of  the  type of  site  for these  estimates.    Finally,   they  have




implications for the feasibility of using econometric reviews of the empirical




benefits literature  in  the  task associated with benefits  transfer  for  policy



evaluations.




      It is important to recognize at the outset that the feasibility of using




econometric methods in literature reviews would be greatly enhanced with a change




in  reporting  conventions for empirical  results.    These  conventions  are  so




variable across studies that the set of available estimates with a detailed set




of explanatory variables  is almost half the size of our full sample of estimates.




Because missing values for particular classes  of  variables changed our sample




composition  dramatically,  we  investigated  their   effects  by  considering




alternative subsets of the potential explanatory variables specified to influence




(CS/v).  This  process explains the rationale for the first five columns in Table




2.   The estimated effects of the variables describing the modeling strategies




are  quite  stable  across  models  in  terms  of  their  signs and  statistical




significance.  Virtually  all  the decisions on  the  assumptions  associated with

-------
                                                                          2.23
modeling   strategies   that   we   describe  with  qualitative  variables   were




statistically significant factors in determining the real consumer surplus (CS/v)




estimates.




      When  the  variables are interpreted  in terms  of the classification  we




proposed in developing equation (7), the key economic assumptions (such as the




inclusion of a substitute price or measure of the  implicit costs of travel time)




are generally significant determinants of the estimate for the (CS/v) and conform




with a priori expectations.  The  adjustment  for  the  measure of  use indicates,




as we would expect,  smaller benefits per unit in  terms of days  versus trips.




The parameter restrictions implicit in the use of a regional travel cost model




appear to increase estimates of (CS/v).  This finding is more difficult to relate




to economic theory.  The restrictions imposed by  the regional travel cost model




have implications  for  the  implicit  extent  of a  recreation market;  for whether




sites are  considered equivalent  (by recreationists)  in terms  of the estimated




demand responses  to own price  and  income;  and for  the   definition  for what




constitutes substitute sites.




      Some modeling judgments are based  on each  application's data and do not




have a rationale in economic theory.   We have classified the  variables describing




the functional form and estimator in this category.  While  one might argue that




the estimator follows from prior information on the sampling process, we believe




the  potential  sensitivity  of  estimates  Co  parameterization  for the  error




structure  or  its  distribution often leads analysts  to an implicit pretesting




process.   In these cases, results from  different  estimates  are compared as part




of the development of the "final" reported results.  Because we  have  adopted this




view of the process,  we have included the estimator in the data-based variables.




      One  way of  evaluating the  sensitivity of  estimates to  data  specific

-------
                                                                          2.24
judgments is  to test the null  hypothesis  that the variables  associated  with



these decisions do not exert a  significant  influence on (CS/v).  The results in




columns (6)  and (7) indicate that  this hypothesis  is decisively rejected at the



one percent significance level.




      The use "of  a maximum  likelihood estimator  and  selection of a log-linear




demand  specification  seem  especially   important individual  choices.    The




sensitivity of results in each  case may reflect biases arising in other studies




that do not make these assumptions.   This may be  especially true for the use of




procedures  adjusting for the  on-site,  intercept  nature  of most  micro level




recreation surveys.  Nonetheless,  the importance of both decisions for estimates




could  be reduced with  improved  information  on the nature of  households'




recreation site choices, including the amount of use and the time and resource




constraints underlying these decisions.




      We are able  to distinguish  separate effects  for our measures of the type




of recreation site and for the  primary activities supported by a site.  Because



the site definitions  are not mutually exclusive categories, we need  to interpret




the results carefully.   For example, a trip  to a  lake in a national park would




be worth $19.94 more than one of comparable length to  a coastal area (i.e., the




sum of  the  coefficient for National Park,  41.13,  and that  of Lake,  -21.19 in




column  (8) of Table  2).  The results indicate that sites supporting wilderness




activities do not  appear different than those for developed camping, comparing



their consumer surplus estimates.   This seems implausible, given  the activities




involved, and is likely to result  from the small number of travel  cost estimates




for wilderness areas  (i.e.  about  10 of the 399 used in the models).




      Finally, this type of model offers the potential for "checking" the benefit




transfer  estimates developed  in  policy  analysis.   Because  we do not  have  a

-------
                                                                          2.25
theoretical basis for specifying how (CS/v) should behave across different types

of recreation sites and modeling strategies,  it would not be prudent to recommend

this  type of  model  for  predictions  of  consumer  surplus  per  unit of  use.

Intangible dimensions of a  research  study exist  which are difficult to  encode

in the quantitative  terms required for an econometric summary.   These factors

may well be  important to how  policy  analysts  should use  a particular study in

a benefits transfer.  At this  stage, we can say  that these types of empirical

summaries  can  serve as a  consistency  check on  the processes used  in  policy

analyses to gauge the implications of selecting a different set of assumptions.

They  also  offer a first step  in a more  general question--how do we want to

summarize  the results of applied  demand analysis?  Should the focus be  on the

consumer surplus per unit of use  or the own-price elasticity of demand?  Either

could be used (with supplementary assumptions) as a basis  for evaluating policy

uses of benefit studies on  the research shelf.



IV.   Implications

      As the literature reporting benefit  estimates  for environmental resources

expands, the task of summarizing  what  we  know and how to use it in evaluating

new  policies  that  affect  environmental and  other  resources  becomes  more

difficult.  Our findings here  indicate  that econometric methods can be used to

summarize  the results from  diverse empirical  studies. Indeed, in our specific

application (travel cost recreation demand models), this approach provided clear

support  for  the  issues identified  in the  theoretical   and  recent empirical

literature as central to implementing  the model.  They include:

      (a)   the implications of the treatment of an individual's time constraints
            for his  (or her) opportunity  costs of time (see Bockstael, Strand,
            and Hanemann);

-------
                                                                          2.26
      (b.)   the  identification  and  treatment of substitute sites  in modeling
            recreation demand (see Rosenthal);

            and

      (c)   the  adjustment of  estimates  from on-site micro data  sets  for the
            specification effects of these sampling procedures (see Shaw).

These factors are important from a conceptual perspective, and they could help

to resolve the rather wide variation in real consumer surplus across studies.

      More generally,  these  results offer  some confirmation  that systematic

factors  influence  the  disparity in results  across  studies.   However,  applied

econometric  analyses  of recreation  demand  require  substantial  discretionary

judgments  to overcome  limitations  imposed  by  data and  by our  knowledge  of

economic agents' behavior.   Some of these  factors arise  from differences in the

resources  involved and  others  from the  assumptions  used in these studies.

Because it appears possible to  separate the  influence of these factors, reviews

of empirical research using econometric  methods  to estimate these  types  of

response surfaces based on  the empirical findings can also have important policy

applications.  They offer a method for  bounding  (or for  checking)  the estimates

derived for new  or improved resources.   They can serve  to  identify the factors

leading  to the  greatest disparity in benefit estimates.   And,  finally,  these

cross -study  empirical  summaries may also help  to  isolate the areas requiring

further research.

-------
                       Table 1.  DESCRIPTION OF VARIABLES FOR ANALYSIS
Name
Mean
Definition of Variables
(CS/v)
SURTYPE
Type of
Recreation
Activities
Type of
Recreation Site
Substitute Price
Opportunity Cost
type #1
Opportunity  Cost
type #2
Fraction of wage


Specific Site
Demand
Specifications
Year

Estimators  Used*
25.24   Marshallian consumer surplus  estimated per unit of use,  as
        measured by each study (i.e. ,  per day or per trip) deflated
        by consumer price index (base - 1967)

   .86   Qualitative variable for measure of site use -  1 for per
        trip measure, 0 for per day measure

   -•-   Water-based  recreation   (swimming,  boating,   fishing),
        hunting, wilderness hiking, and developed camping were
        identified as the primary activities.  The first three are
        introduced as qualitative variables with developed camping
        as the omitted category.

   —   Lake, river, coastal area and wetlands, forest or mountain
        area, developed or state park, national park with  or without
        wilderness significance are the designations. Coastal area
        and wetlands was  the omitted category.  Variables are unity
        if satisfying designation, zero otherwise.

   .29   Qualitative  Variable -  1 if  substitute  price  term was
        included in the demand specification,  0 otherwise

   .24   Qualitative Variable for  the measure used to estimate
        opportunity cost of travel time - 1 if an average wage rate
        was used.

   .32   Qualitative Variable for  the second type of opportunity
        costs of  travel  time measure,  - 1 for  use of income per
        hour; the omitted category was the  use of a projection for
        an individual specific wage  rates.

   . 37   Fraction of wage rate used to estimate opportunity cost of
        travel  time

   .24   Qualitative Variable for use of a state or regional travel
        cost  model describing demand  for  a set of sites  - 1,  0
        otherwise.

   —   Linear, log-linear  and semi-log  (dep) are qualitative
        variables  describing the  specification of functional form
        for demand (semi-log in  logs of independent variables was
        the omitted category).

   —   The year of the data used in each  study.

   ---   OLS,  GLS,  and  ML-TRUNC  are  qualitative  variables  for
        estimators  used,   omitted    categories   correspond  to
        estimators with  limited  representation  in studies--the
        simultaneous equation estimators.
      "ML-TRUNC  refers  to maximum likelihood estimators adjusting for truncation  and
 tobit estimators.  GLS  includes both  single equation generalized least squares  and
 seemingly unrelated regressions.

-------
                                                                  2.28
Table 2. DETERMINANTS OF REAL CONSUMER SURPLUS PER UNIT OF USE*
Independent
Variables
1
Intercept 20.30
(6.19)
[3.92]
SURTYPE -9.97
(-2.72)
[-1.36]
rXA) Tvr»e of
Recreation
Water-Based
Activities
Hunting
Wilderness
(Xs) Tvpe of Site
Lake
River
Forest
State Park
National Park

2
27.03
(3.68)
[3.64]
15.38
(2.97)
[2.34]


-18.69
(-3.24)
[-2.36]
-14.29
(-2.99)
[-1.95]
-18.45
(-2.36)
[-1.93]
24.95
(3.47)
[3.27]
.56
(0.04)
[0.08]

3
18.75
(0.58)
[1.04]
19.88
(3.74)
[3.55]


-20.32
(-3.52)
[-2.48]
-19.03
(-2.19)
[-1.75]
-25.99
(-3.01)
[-2.49]
22.37
(3.44)
[3.19]
-3.77
(-0.23)
[-0.13]
Models
4
23.48
(1.57)
[3.71]

14.50
(0.83)
[1.08]
17.35
(1.33)
[4.23]
-12.10
(-0.66)
[-2.49]
-17.47
(-3.12)
[-2.28]
-12.19
(-2.57)
[-1.86]
-15.37
(-1-31)
[-2.53]
14.10
(2.40)
[1.64]
30.71
(2.16)
[2.51]

5
-.30
(-.01)
[-0.01]
1.03
(0.23)
[0.12]
24.50
(1.97)
[2.72]
20.02
(1.53)
[1.63]
10.92
(0.76)
[0.62]





6
5174.24
(3.95)
[3.39]
28.75
(4.84)
[4.71]
24.43
(0.78)
[1.95]
-2.33
(-0.18)
[-0.26]
-26.57
(-1.47)
[-1.95]
-22.16
(-3.88)
[-2.57]
-16.44
(-1.91)
[-1.60]
-1.36
(-0.05)
[-0.16]
28.39
(4.28)
[3.30]
49.37
(1.33)
[1.58]

7
4904.00
(3.75)
[3.52]
16.94
(2.78)
[2.05]
-9.07
(-0.26)
[-0.96]
-1.10
(-0.08)
[-0.14]
-17.52
(-0.91)
[-1.47]
-13.21
(-2.42)
[-1.60]
3.23
(0.44)
[0.32]
-20.74
(-0.64)
[-2.25]
24.46
(3.44)
[3.07]
-5.43
(-0.14)
[-0.25]

8
-25.20
(-0.57)
[-1.74]
19.18
(3.46)
[3.10]
45.39
(1.44)
[4.01]
13.78
(1.07)
[1.46]
.60
(0.04)
[0.07]
-21.19
(-3.65)
[-2.55]
-19.80
(-2.27)
[-1.80]
6.84
(0.23)
[0.82]
22.18
(3.37)
[3.20]
41.13
(1.09)
[1.24]

-------
                                                                                         2. 29
T_able  2
Independent
Variables
fXB) Model
Assumption
Substitute Price
Opportunity Cost
Type #1
Opportunity Cost
Type #2
Fraction of Wage
Specific Site/
Regional TC Model
^TXD) Model
^Specification
Linear
Log -Linear
Semi -Log (Dep)
(XE) Estimator
OLS
GLS

1 2

-18.73
(-3.27)
[-4.58]
-14.97
(-2.10)
[-2.09]
3.95
(1.02)
[0.45]
37.24
(8.56)
[3.83]
22.23 .
(4.10)
[3.35]





Models
3 4

-13.71
(-2.12)
[-1.80]
-16.49
(-2.11)
[-2.48]
-15.86
(-3.30)
[-2.87]
48.59
(9.76)
[6.94]
24.21
(3.85)
[2.77]
-2.87
(-0.27)
[-0.31]
23.37
(2.37)
[2.88]
16.89
(1.86)
[2.97]
-14.45
(-0.48)
[-0.84]
-8.58
(-0.28)
[-0.54]

5

-23.80
(-3.76)
[-3.18]
-21.68
(-2.94)
[-2.72]
-13.59
(-2.75)
[-1.93]
55.88
(11.41)
[7.33]
21.75
(3.54)
[2.08]
12.99
(1.19)
[1.10]
28.57
(2.67)
[2.05]
15.97
(1.62)
[2.07]
-24.20
(-0.76)
[-1.39]
-24.77
(-0.78)
[-1.53]

6

-11.42
(-1.82)
[-1.43]
-6.03
(-0.73)
[-0.71]
-10.97
(-2.22)
[-1.90]
45.10
(9.09)
[6.70]
16.49
(2.55)
[1.62]
-15.33
(-1.37)
[-1.41]
15.61
(1.37)
[1.59]
9.29
(0.97)
[1.74]
-28.96
(-0.96)
[-1.39]
-21.88
(-0.73)
[-1.13]

7 8

-18.58 -14.39
(-3.00) (-2.26)
[-4.10] [-1.80]
8.03 -14.28
(0.97) (-1.75)
[0.95] [-1.98]
5.84 -15.89
(1.39) (-3.26)
[0.71] [-2.80]
27.02 48.59
(6.01) (9.76)
[2.54] [6.91]
23.54
(3.71)
[2.64]
-2.94
(-0.27)
[-0.29]
24.65
(2.36)
[2.68]
18.61
(1.96)
[2.86]
-16.21
(-0.53)
[-0.92]
-8.58
(-0.28)
[-0.53]

-------
                                                                                 2. 30
Table 2   (continued)
Independent
Variables

ML-Trunc


Year


R2
n


123
-67.38
(-2.15)
[-3.43]



.25 .15 .42
399 399 399
Models

4 5
-77.35
(-2.38)
[-3.65]



.15 .36
405 405


6
-85.06
(-2.74)
[-3.63]
-2.61
(-3.98)
[-3.63]
.45
399


7 8
-68.98
(-2.20)
[-3.46]
-2.47
(-3.74)
[-3.52]
.30 .43
399 399
     'The  numbers  in  parentheses below  the  estimated  parameters  are  the ratios  of  the
coefficients to their estimated standard errors.  The  numbers  in brackets use the  Newey-West
variant of the White  consistent  covariance estimates for the  standard  errors in calculating
these ratios.

-------
                                  CHAPTER 2                               2- 31

                                  FOOTNOTES


  University  Distinguished Professor,  North Carolina  State University,  -and
Resources for the Future University Fellow; Assistant Social Scientist,  Marine
Policy  Center,  Woods Hole Oceanographic  Institution,  respectively.   A large
number  of individuals  contributed to  this  effort  by  providing the  source
materials for both published and unpublished  papers.  Since we wrote to all the
individuals whom we could identify as active researchers in recreation economics,
and all  Chairs'.of Departments of Agricultural Economics and of Economics, we
cannot identify them individually. Thanks are due Michael Hanemann for calling
our attention  to the meta-analysis  literature outside economics  and  to Peter
Caulkins, Jerry Carlson,  Bill Desvousges, Ted McConnell,  and three  anonymous
referees for exceptionally careful and constructive comments on earlier drafts
of  this paper.   This  research  was  partially  supported  through  the  U.  S.
Environmental Protection Agency Cooperative Agreement No. CR812564.


1.    Hedges and Olkin credit  Glass  with the  first use of meta-analysis in
      educational and psychological research.  There are important differences
      in the use of  these methods for applications in these disciplines, as well
      as for medical research, in comparison with economics.  All  of the former
      have involved controlled experiments, where the statistical analysis can
      be treated as aggregating independent observations from each study's sample
      of experimental findings  (see Cordray).

2.    Bockstael and McConnell [1983] is a notable  exception to this  work, because
      they  use the  formal  structure  of  a household production  framework to
      describe  the measure of the demand  for non-marketed commodities and the
      role of the assumption of weak complementarily.

3.    This problem is analogous to the issues raised in modeling the demand for
      electricity in the presence of  declining block rates  (see Taylor [1975]
      for an early discussion) or in the more recent analyses of  hedonic models'
      ability to recover estimates of  the  willingness-to-pay functions for  non-
      marketed  resources.  See fiartik and Smith  [1987].

4.    These types of questions can be found on the recent Public  Area Recreation
      Visitors  Surveys  conducted by  the U.S. Forest Service, as  well as on  a
      wide variety of other micro-level site-specific surveys.  This framework
      presumably arises because of the difficulty of encoding (with an on-site
      survey) a consistent set of substitute sites.    See Smith and Desvousges
      [1986]  for discussion of  a procedure used  in  surveying water-based
      recreation participation patterns as part of a contingent valuation survey.

5.    Hof and  King  [1982]  give the impression that  the  issues  are clear-cut.
      In practice,  data inadequacies  and  on-site surveys  make  the process of
      inferring the feasible set of substitutes and of treating them consistently
      exceptionally difficult.

-------
                                                                          2.32
6.    Collinearity in the cross price measures makes it difficult to precisely
      estimate their effects  on demand.  It does not affect the magnitude of the
      estimated coefficients. However,  to  the  extent these are not estimated at
      conventional standards  for  statistical significance, practitioners  can
      easily be faced with a  dilemma  in judging how  to interpret and respond to
      test results in these cases.

7.    While this estimator for (CS/v) has  been commonly used in the literature,
      without a consistency check to screen for negative values,  it will not have
      finite moments (see Smith [1989]).

8.    A lag of  eleven periods  was  used in implementing the  Newey-West version
      of White's estimator.

-------
                                    CHAPTER 2

                                    APPENDIX

                    Real Consumer Surplus per Unic of Use and
                         Own  Price  Elasticity of  Demand*
                                                     2.33
                                                          Ranse (Estimate)
     Author
 Identification  Number  of
	Number	Estimates
                                                     (CS/v)
                  Own Price
                  Elasticity
Karl C. Samples and           1
Richard Bishop

Marc 0. Ribaudo    -           2
and Donald J.  Epp

Donald H. Rosenthal [1985]    4

Christine Sellar              5

Cindy F. Sorg,               13
John B. Loomis, D. Donnelly,
G. Peterson, L. Nelson

Abraham E. Haspel and        17
F. Reed Johnson

Fredric C. Menz and          22
Donald P. Wilton

John K. Mullen and           23
Fredric C. Menz

William G. Brown,            25
Colin Sorbus,
Bih-lian Chou-Yang, and
Jack Richards

J. A. Sinden                 34

R. E. Capel and R. K. Pandey 37

Ronald J. Sutherland (1982a) 45

V. Kerry Smith and           51
William H. Desvousges (1985)

V. Kerry Smith, William H.   52
Desvousges, and Ann Fisher

John B. Loomis (1986a)       62

John B. Loomis (1986b)       63

W. David Klemperer,          67
Gregory J. Buhyoff,
P. Verbyla, and L. Joyner
                     11
                     22

                     11

                     51
                      1

                      1

                     40

                     44
                      1

                      3

                      8
  .11 -   6.24


     3.66


  .46 -   5.85

 2.89 -  15.17

 9.19 -  20.81



20.60 -  36.84


 9.02 -  20.60


 7.41 -  13.12


    15.93
      .29

     9.26

 1.36 - 40.32

 1.97 - 219.78


      7.21


     12.53

11.53 - 22.36

  .92 -  3.90
    -.49


 -1.79  to  -4.58

•0.003  to  -0.02
                                                 -1.49
     -.54

    -1.05



   .04 to -2.99
 •1.63 to -1.71

-------
                                                                            2.34
Appendix  (continued)
     Author
Identification  Number of
    Number	Estimates
                                                          Range  (Estimate)
                                                     (CS/v)
                  Own Price
                  Elasticity
Cindy F. Sorg and
Louis J. Nelson
      71
Cindy F. Sorg,               72
John B. Loomis, D. Donnelly,
G, Peterson, L. Nelson

Dennis M. Donnelly,          73
John B. Loomis,
Cindy F. Sorg and Louis Nelson

Stephen Farber               79

Werner J. Sublette and       82
William E. Martin

William J. Vaughan and       84
Clifford S. Russell
                     1

                     4
21.20 -  33.50


13.23 -  14.41



 6.67 -   9.35



    17.45

 7.88 -  37.54


 3.23 -   7.88
Thomas Gifford Sawyer
C. Tim Osborn
Trellis G. Green
Daniel Wayne McCollum
Colin Norman Sorhus
Faisal Moftah Shalloof
Bahram Adrangi
Steven Eric Daniels
Chung -Huang Huang
Margaret Tambunan
V. Kerry Smith and
Raymond Kopp
V. Kerry Smith (1975)
96
97
98
99
103
105
109
112
113
114
115
116
1
1
2
28
4
8
2
5
43
73
2
2


59
8
9
49
7
4
4
4
4

48
88
.96
.06
.23
.14
.68
.99
.93
.77
.22
5
.11
.07
- 146
- 146
- 28
- 91
- 10
6
- 60
- 327
- 11
.03


.95
.95
.47
.59
.02
.23
.20
.22
.84

-.17
-.27
-.0016 to -.005
—
-.73 to -.81
—
...
-5.97 to - 6.96
-.05 to -.84
-.0003 to -.93
-1.59 to -1.71
• » •
     These results are for only those studies included in our most detailed model
based on  399  estimates from 35 studies.

-------
                                                                          2.35


                                   CHAPTER  2

                                  REFERENCES

Adrangi, Bahram.   1982.  Change in Economic  Efficiency Resulting from Allocation
      of Oregon  National  Forest Land  to  Skiing,  dissertation, University  of
      Oregon, June.

Bartik, Timothy J.  and V. Kerry Smith. 1987. "Urban Amenities and Public Policy,"
      in Handbook of Regional and Urban Economics. Vol. II, edited by E. S. Mills
      (Amsterdam:  North Holland).

Becker, Gary S. 1965.  "A Theory of the Allocation of Time," Economic Journal 75
      (September): 493-517.

Bockstael, Nancy E. and Kenneth E. McConnell.  1981.  "Theory  and Estimation of
      the Household Production Function for Wildlife Recreation,"  Journal  of
      Environmental Economics and Management 8 (September):  806-814.

	  	    .   1983. "Welfare Measurement  in the  Household  Production
      Framework," American Economic Review 73 (December): 806-814.

Bockstael, Nancy E. and Ivar E.  Strand, Jr. 1987. "Regression Error and Benefit
      Estimates," Land Economics 63 (February): 11-20.

Bockstael,  Nancy  E. ,  W.  Michael  Hanemann,  and  Ivar  E.  Strand,  Jr.  1987.
      Measuring  the  Benefits of Water  Quality Improvements Using  Recreation
      Demand  Models.  Department  of  Agricultural  and  Resource  Economics,
      University of Maryland, Vol. II of report to U.S. Environmental Protection
      Agency.

Bockstael, Nancy E., Ivar E. Strand, Jr., and W. Michael Hanemann. 1987. "Time
      and  the  Recreation  Demand  Model,"  American  Journal  of  Agricultural
      Economics 69 (May): 293-302.

Brown, William G., Colin Sorhus, Bih-lian Chou-Yang, and Jack Richards.  1983.
      "Using  Individual  Observations  to  Estimate  Recreation Demand Functions:
      A  Caution,"  American Journal  of  Agricultural  Economics  65 (February):
      154-157.

Capel, R. E.  and R. K. Pandey.  1973.  "Evaluating Demand for Deer Hunting:  A
      Comparison of  Methods," Canadian  Journal of Agricultural  Economics 21
      (November).

Cesario, Frank J.  and  Jack L.  Knetsch.  1970.  "Time Bias in Recreation Benefit
      Estimates," Water Resources Research 6 (June): 700-704.

Clawson, Marion. 1959.  "Methods  of Measuring the Demand for and Value of Outdoor
      Recreation," Resources for the Future, Reprint No. 10, February.

Clawson,  Marion and Jack  L.  Knetsch. 1966.  Economics  of  Outdoor  Recreation
      (Baltimore:  John Hopkins University).

-------
                                                                          2.36
Cordray,  David  S.    1987.    "Strengthening  Causal  Interpretations  of  Non-
      Experimental Data:   The Role of Meta-Analysis," unpublished manuscript to
      appear in  L. Sechrest, J. Bunker, and E.  Perrin,  ed. ,  Improvine Methods
      in Non-Experimental Research (Menlo  Park,  Co.:   Sage Publications).

Daniels, Steven Eric.  1986.  Efficient  Provision of Recreational Opportunities:
      The  Case  of  U.   S.  Forest  Service  Campgrounds,  dissertation,   Duke
      University, January.

Deyak, Timothy"A,,  and V. Kerry Smith.  1978.  "Congestion  and  Participation in
      Outdoor  Recreation:   A Household  Production  Approach,"  Journal  of.
      Environmental Economics and Management 5 (March):  63-80.


Donnelly, Dennis M. ,  John B. Loorais,  Cindy F. Sorg, and  Louis  Nelson.   1985.
      "Net Economic Value of Recreational Steelhead Fishing in Idaho," Bulletin
      RM-9, Rocky Mt. Forest and Range Experiment Station.

Farber,  Stephen.   1985.   "The Value of Coastal Wetlands for Recreation:   an
      Application  of Travel  Cost  and Contingent  Valuation  Methodologies,"
      unpublished paper  (October).

Freeman, A.  Myrick  III.  1979. The Benefits of Environmental Improvement:  Theory
      and Practice (Baltimore:   Johns Hopkins University).

Glass,  G.  V.    1976.  "Primary,   Secondary,  and  Meta-Analysis  of  Research,"
      Educational Researcher. Vol. 5, No.  1, 3-8.

Green, Trellis G.  1984.  Compensating and Equivalent Variation of the Florida
      Saltwater Tourist Fishery, dissertation,  Florida State University, January.

Haspel, Abraham E.  and F. Reed  Johnson.  1982.   "Multiple  Destination Trip Bias
      in Recreation Benefit Estimation," Land Economics 58 (No. 3, August).

Hedges, Larry V. and Ingram Olkin.   1985. Statistical Methods for Meta-Analysis
      (Orlando,  Florida:  Academic Press).

Hendry, David  F. 1983.   "Econometric Modeling:   The 'Consumption Function' in
      Retrospect," Scottish Journal  of Political Economy 30 (November): 193-220.

Hof, J. G. and D. A. King.  1982. "On the Necessity of Simultaneous Recreation
      Demand Equation Estimation," Land Economics 58 (November): 547-552.

Hotelling, Harold. 1947.  Letter  to the National Park Service in Economics of
      Outdoor  Recreation--The Prewitt Report.

Huang, Chung-Huang.  1986. The Recreation Benefits of Water Quality improvement
      in Selected  Lakes  in Minnesota, dissertation,  University of Minnesota,
      January.

-------
Kealy, Mary Jo and Richard Bishop. 1986. "Theoretical and Empirical Specification
      Issues in Travel Cost Demand  Studies,"  American Journal  of Agricultural
      Economics 68 (August):   660-667.

Klemperer,  W.  David,  Gregory J. Buhyoff,  P.  Verbyla, and  L.  Joyner.   1982.
      "Valuing  White-Water  River   Recreation by   the  Travel  Cost  Method,"
      unpublished paper.

Kling, Catherine  L.  and Richard J..  Sexton.   1989.  "Bootstrapping  in Welfare
      Analysis," unpublished paper,  University of California,  Davis, March.

Light, Richard  J.  and David B. Pillemer.  1984.   Summing Up:   The  Science  of
      Reviewing Research (Cambridge:  Harvard University Press).

Loomis, John B.  1986.  "Economic Losses to Recreational Fisheries due to Small-
      head Hydro -power Development:  A Case Study of the Henry's Fork in Idaho,"
      Journal of Environmental Management 22 (January) .

Loomis, John B.,  Cindy F. Sorg, and  Dennis  M.  Donnelly.   1986.   "Evaluating
      Regional Demand Models for Estimating Recreation Use and Economic Benefits:
      A Case Study," Water Resources Research 22 (No. 4,  April):   431-438.

McCollum, Daniel Wayne.   1986.  The Travel Cost  Method:   Time. Specification.
      and Validity, dissertation,  University of Wisconsin-Madison, May.

McConnell,  K. E.  and  I.  E.  Strand,  Jr.  1981. "Measuring the  Cost  of Time  in
      Recreational Demand Analyses:   An Application to Sport - f ishing, " American
      Journal of Agricultural Economics 63: 153-156.

Menz, Fredric  C.  and Donald P. Wilton.   1983.   "Alternative  Ways  to Measure
      Recreation  Values   by  the  Travel  Cost Method,"  American Journal  of
      Agricultural Economics 65 (May):  332-336.

Mitchell, Robert Cameron and Richard  T.  Carson.  1989.   Using  Surveys to Value
      Public  Goods:   The  Contingent Valuation   Method   (Washington ,  D . C . :
      Resources for the Future) .

Mullen, John K.  and Fredric C. Menz.   1985.  "The Effect of Acidification Damages
      on  the Economic Value of the Adirondack Fishery  to  New York Anglers,"
      American Journal of Agricultural Economics 67  (February):  112-119.

Naughton, Michael  C. , George  R.  Parsons, and William  H.   Desvousges.   1988.
      "Benefits Transfer:    Conceptual Problems in Estimating  Water Quality
      Benefits Using Existing Studies," unpublished paper,  February.

Newey, Whitney K.  and Kenneth D. West.   1987.  "A Simple, Positive Semi -Definite,
      Heteroskedasticity  and Autocorrelation  Consistent Covariance  Matrix,"
      Econometrica 55 (May):  703-708.
Office of Policy Analysis.  1987.  EPA's Use of Benefit-Cost Analysis:
      1986 (Washington, D. C. :   U.  S.  Environmental  Protection Agency, August),
      EPA-230-05-87-028.

-------
                                                                          2.38
Osborn, C. Tim.  1981.  "The Value of Recreational Benefits due to Controlling
      Erosion in the North Lake Chicot Watershed," M.  S.  thesis,  University of
      Arkansas, May.

Ribaudo,  Marc  0.  and  Donald J.  Epp.    1984.    "The  Importance  of  Sample
      Discrimination in Using the Travel Cost Method  to  Estimate the  Benefits
      of Improved Water Quality," Land Economics 60 (November):  397-403.

;;osenthal, Donald  H.   1985.  Representing Substitution  Effects  in Models  of
      Recreation Demand, dissertation, Colorado State University.

	.   1987.  "The  Necessity of Substitute Prices  in Recreation
      Demand Analysis , " American Journal of Agricultural Economics  69 (November):
      828-837.

Samples,  Karl C.  and  Richard C.  Bishop.    1985.    "Estimating  the Value  of
      Variations in Anglers'  Success Rates:  An Application  of the Multiple-Site
      Travel Cost Method," Marine Resource Economics 2 (No. 1).

Sawyer, Thomas  Gifford.   1976.   "An  Economic Study of the  Demand for Publicly
      Provided Outdoor Recreation at Beaver Reservoir," M. S. thesis, University
      of Arkansas, January.

Sellar, Christine.  1982.  The Value  of Recreational  Boating  at  Lakes in East
      Texas, dissertation, Texas A & M University.

Shalloof, Faisal Moftah.   1985.   Impact of Various Factors Upon Benefits from
      Big Game Hunting Estimated by the Travel Cost Method,  dissertation, Oregon
      State University, January.

Shaw, Daigee.   1988.  "On-Site Samples' Regression:   Problems  of Non-Negative
      Integers,  Truncation,   and  Endogenous  Stratification,"   Journal  of
      Econometrics 37 (February): 211-224.

Sinden, J. A.   1974.  "A Utility Approach to the Valuation of Recreational and
      Aesthetic  Experiences,"  American Journal  of Agricultural  Economics 56
      (February):   61-72.

Smith, V. Kerry. 1975.   "Travel  Cost Demand Models for Wilderness Recreation:
      A Problem of Non-Nested Hypotheses," Land Economics  51 (May):  103-111.

	.  (ed.). 1984.  Environmental Policy Under Reagan's Executive
      Order:  The Role of Benefit-Cost Analysis  (Chapel Hill, N.  C.: University
      of North  Carolina Press).

	.  1989.  "Nearly All Consumer Surplus  Estimates Are Biased,"
      unpublished paper, North Carolina State University, April.

Smith, V. Kerry and William  H.  Desvousges.  1985.  "The Generalized Travel Cost
      Model and Water Quality Benefits:  A Reconsideration," Southern Economic
      Journal   50  (October): 371-381.

-------
                                                                         2.39
                    1986.   Measuring Water  Quality Benefits (Boscon:   Kluwer
      Nijhoff).
Smith, V. Kerry and Raymond J. Kopp. .1980.  "The Spatial Limits of the Travel
      Cost Recreational Demand Model,"  Land Economics 56 (February):   64-72.

Smith, V. Kerry, William H. Desvousges, and Ann  Fisher.   1986.   "A Comparison
      of Direct  and Indirect  Methods for  Estimating Environmental Benefits,"
      American- Journal of Agricultural  Economics  68 (No. 2,  May).

Smith, V. Kerry, William H. Desvousges, and Matthew  P.  McGivney.    1983.  "The
      Opportunity  Costs  of Travel  Time  in Recreation  Demand Models,"  Land
      Economics 59  (August): 259-277.

Sorhus, Colin Norman.  1981.   Estimated Expenditures by Sport  Anglers and Net
      Economic Values  of  Salmon and Steelhead for Specified Fisheries  in the
      Pacific Northwest,  dissertation,  Oregon State University, June.

Sorg, Cindy  F.  and John B. Loomis.  1982.  "A  Critical  Summary  of Empirical
      Estimates of  the Value  of Wildlife, Wilderness,  and  General Recreation
      Related to  National  Forest Regions,"   (Fort Collins, Colorado:  U.  S.
      Forest Service, Rocky Mountain Experiment Station).

Sorg, Cindy F. and Louis J. Nelson.   1986.  "Net Economic Value of Elk Hunting
      in Idaho," Resource Bulletin RM-12,  Rocky Mt. Forest and Range Experiment
      Station.

Sorg, Cindy F.,  John B. Loomis, D. Donnelly, G. Peterson, and L. Nelson.  1985.
      "Net Economic Value  of  Cold and Warm Water  Fishing in  Idaho,"  Resource
      Bulletin RM-11, USDA Forest Service, November.

Sublette, Werner J.  and William E.  Martin.  1975.   "Outdoor Recreation in the
      Salt-Verde Basin of Central Arizona:   Demand and Value," Technical Bulletin
      218, Agricultural Experiment Station, University of Arizona, June.

Sutherland, Ronald  J.   1982a.   "The Sensitivity of  Travel Cost  Estimates of
      Recreation Demand to the Functional  Form and  Definition of Origin Zones,"
      Western Journal of Agricultural Ecqnomics  7  (No. 1, July).

	.   1982b.   "A Regional Approach  to  Estimating Recreation
      Benefits of Improved Water Quality," Journal of Environmental  Economics
      and Management 9 (September):   229-247.

Tambunan, Margaret.   1986.   Targeting  Public Investment:  An Application to
      Recreational.  Planning in Minnesota.  Ph.D.  dissertation, University of
      Minnesota (November).

Taylor, Lester D.  1975.  "The Demand for Electricity:  A Survey," Bell Journal
      of Economics 6 (Spring): 74-110.

-------
                                                                         2.40
Vaughan, William J.  and Clifford  S. Russell.  1982. "Valuing a Fishing Day:   An
      Application of a Systematic Varying  Parameter  Model,"  Land  Economics  58
      (November): 450-463.

Ward, Frank A.  and  John  B.  Loomis.  1986. "The Travel Cost Demand  Model  as  an
      Environmental Policy Assessment Tool:  A Review  of Literature,"  Western
      Journal of Agricultural Economics 11 (December):  164-178.

White,  Halbert.  1980.  "A  Heteroskedasticity-Consistent  Covariance   Matrix
      Estimator and a Direct Test  for Heteroskedasticity," Econometrica 48:  817-
      838.

-------
                  CHAPTER 3
WHAT HAVE WE LEARNED SINCE HOTELLING'S LETTER?

               A META ANALYSIS
                V. Kerry Smith
                      and
                Yoshiaki Kaoru

-------
                                                                        3.1
                What Have We Learned Since Hotelling's Letter?

                               A Meta Analysis

                       V.  Kerry Smith and  Yoshiaki  Kaoru

I.  Introduction

      In 1947, Harold Hotelling proposed the first indirect method for measuring

the demand for,a non-marketed commodity.   His  letter,  responding  to a request

by the director of the National Park Service for methods that might be used to

measure  recreation benefits,  introduced  the  travel  cost  recreation  demand

method1.   About twelve  years  later,  Trice and Wood  [1958]  and Clawson [1959]

independently implemented the methodology.  Because there have been hundreds of

applications in the intervening thirty years, a comprehensive literature review

could easily  fill  several lengthy papers2.  Moreover, given  the  diversity of

recreation sites and types of data,  the task of developing a consistent synthesis

is exceptionally difficult.

      This paper proposes the use  of econometric methods for quantitative reviews

of empirical  literature.  Our strategy  builds on the  concept of statistical

review or meta-analyses introduced into the education and psychology literature

by Glass [1976]  (see also Hedges and Olkin [1985] and Cordray [1987] for detailed

discussion).   Because  empirical  studies  in  economics are  rarely controlled

experiments, the data aggregation methods proposed for most met a analyses must

be  replaced  by  the  multiyariate methods routinely  applied  in  econometric

analysis.   This paper  uses  the  travel  cost recreation demand literature to

illustrate what can be learned from a meta-analytic review.



II.  Data. Model and Results

      The data for  this  meta-analysis of travel cost recreation demand models

were derived from a larger study investigating the feasibility of transferring

-------
                                                                        3.2
recreation benefit estimates from the situations where  they were  estimated  to




new applications of policy interest (see Smith and Kaoru  [1989]).  As  part  of




that effort, we  reviewed  200  published and unpublished studies of the demand




for  recreation resources prepared  from 1970  to 1986.   The  set of  studies




considered was developed  by:   (1)   reviewing all journals  (both  economic  and




noneconomic)  that consistently  publish  recreation   demand  studies;  several




computer literature searches and dissertation abstracts; and by contacting active




researchers  in this  area,  chairpersons  for  all  economics and  agricultural




economics  departments  with graduate  programs  in the  U.S.,  and  the research




experiment stations of the U.S. Forest Service.




      Seventy-seven of these studies reported sufficient information to permit




estimation of the benefits provided by the site(s) involved in each study.  They




represent  the  initial data base  for this  study.   Forty-seven  were unpublished




(Master's and Ph.D. theses and papers or reports) and 30 published3.   Seventy-




nine percent of the studies were  prepared  in 1980 or later.  Thirty-one studies




reported sufficient information to estimate the own price elasticity of demand




implied by each demand model.  Our  analysis was  confined to the studies whose




models yielded theoretically plausible  elasticity estimates  (i.e., negative




values).   Overall, these studies lead to 211 own-price elasticity estimates4; 88




percent of these cases also had sufficient detail to permit a meta analysis of




the determinants of the estimated price elasticities.




      Our analysis is based on a simplified view of model development adapted




from Hendry's  [1983]  work in the context  of  macro models.   The  arguments we




hypothesize to be the important determinants of the quantity demanded of a normal




good or service are reasonably well-defined from theory (i.e., prices, income,

-------
                                                                           3.3




and  perhaps   variables  reflecting   individual   tastes).     The   empirical




implementations of  these  models  depend on the problem(s) being  addressed and




the data available.  For the most part, inadequacies in data introduce a large




number  of  compromises.    Our  specific  application  is  important  to  these




compromises.




      The essential  element  in Hotelling's proposal was  the  recognition that




people pay an implicit price  for the use of a recreation site.  This cost is the




total of the travel-related costs to visit the site,  including both the vehicle-




related  and time  costs. The pricing  of the  time costs has been an important




research  focus of  the  literature.     Equally  important,  the   definition  of




substitutes for a particular recreation site and the measurement of how a site




is used  are also  important distinguishing features  of past  studies.  The type




of data  available affects the estimator used and  has been  important  to the




diversity of estimates  in recent applications.  Theory does not offer guidance




on the functional  form or definition of what constitutes homogeneous services




from one (or more) recreation site(s).  In addition  to  these practical modeling




decisions,  we would expect that  demands for  different types of sites would b«




different3.




      On  the  basis  of these  types  of arguments,  we might  hypothesize that




estimates of the demand parameter of interest,  y,  would be a function of:  what




is demanded (i.e., the type  of site)  Xx; how  the economic  arguments  are defined




and measured X^ and potentially some  of the details of implementation X3,  as in




(1)




            Xi - <*o + °lXli +  a2^2i + <*3X3i + 
-------
                                                                           3. 4



error,  reflecting  omissions,  modeling  mistakes and the  approximate  nature of




equation (1).




      The composition of Xlf X2,  and X3 will  depend on what we designate as y.




Because  the  own-price elasticity of demand is  usually  a key  motivation for




developing demand estimates, it seems a natural choice for y.  However, it need




not be  the  only one.   One  important by-product  of this  process of developing




these types of empirical summaries is the identification of this issue.  It is




quite possible that different model features would be statistically summarized




for different uses of  the literature, i.e. ,  one for policy analysis, another for




classifying recreation sites, etc.




      Table 1  reports the  estimates for two  specifications  for equation  (1).




The first column  includes variables  describing the type of site; the economic




assumptions made in developing the models and  the  data-based features (such as




the parameter restrictions used, functional form and  estimator).




      The results  indicate  that our empirical summary has been exceptionally




successful.  The  type of site and economic  assumptions  made do matter, as we




would expect.   In  interpreting  the signs of the coefficients, note the own




elasticity has been entered as a negative value.   Somewhat more troublesome is




the fact that assumptions without clear connection to economic theory, arising




from the data (and therefore the estimator)  or the functional  specification used




for the models also matter.  The second column reports the results without  these




variables.  While  the effects of the remaining variables are quite stable, we




reject this exclusion restriction based on an F-test restricting a subset of the




coefficients to zero.




      Two types of test results are reported for each estimated parameter--one




based  on the  ordinary  least squares  covariance  matrix and  a  second  that

-------
                                                                          3.5



recognizes the prospect for nonspherical errors.   This arises because our sample




resembles  a  panel  in  that  many  studies  report  multiple  estimates--either




different results for different sites or comparisons of the effects of modeling




assumptions.   In  both cases  we  might  expect   some  correlation between  the




estimates.  Consequently, we used the Newey-West [1987]  variant of White's [1980]




consistent  covariance  matrix to   allow  for  generalized  forms   of  both




heteroskedasticity and  autocorrelation.  These are  reported  in  brackets below




the conventional t-ratios and do not change our basic conclusions.








III.  Implications




      Hotelling's letter offered enormously valuable  advice.  The travel cost




recreation demand model is now widely accepted among resource economists, as well




as in federal  guidelines for benefit analysis (see U. S. Water Resources Council




[1983] and U.S. Department of Interior [1986]).   It is generally regarded as a




robust methodology.  '    Our  findings suggest   that  this perception must  be




interpreted carefully.  While the model has been  successful for a wide range of




applications  in estimating  plausible  demand  relationships for recreational




sources,  a systematic analysis  of the record indicates  that modeling assumptions




do matter.  Estimates  of the own price elasticity  of demand depend on how the




issues identified  in  the  current  recreation  demand literature  as  important




theoretical  questions--the  measurement   of  the  opportunity  cost  of  time,




definition of substitution alternatives and measurement of use (i.e. using trips




or days  as  the dependent variable)--are  resolved.  They are also affected by




decisions that are often data-based with little  theoretical justification.

-------
                                                                          3. 6
       Table  1.  Estimated Price Elasticity of Demand from Travel Cost Models:
                               A MetaTAnalysisa
Price Elasticity of Demand
Variable
Intercept

Qualitative Variable for
Measure of Use
1 = trip 0 - per day
Qualitative Variables for
Type of Site
(Overlapping Categories)1*
Lake

River

Forest

State Park

Presence of Substitute Price
(-1)
Use Average Wage Rate to
Measure Opportunity Cost of
Travel Time (-l)e
Use Family Incoae per Hour to
Measure Opportunity Cost of
Travel Time (-l)e
Fraction of Wage Used for
Opportunity Cost of Time
Full Excluding
Specification Judgemental Variables
1033.99
(7.28)
2.11
(3.33)




-.02
(0.05)
-1.77
(-2.54)
-3.77
(-4.74)
2.28
4.28
-1.83
(-6.72)
4.25
(4.20)

1.63
(4.18)

-1.72
(-4.39)

[4.62]

[3.40]





[0.05]

[-2.38]

[-3.40]

2.97

[-5.62]

[2.98]


[3.12]


[-3.26]
829.97
(5.92)
2.45
(4.70)




-.57
(-1.36)
-1.80
(-2.27)
-4.03
(-4.61)
2.04
(3.81)
-.78
(-3.15)
3.25
(3.28)

1.12
(3.93)

-1.73
(-6.41)

[4.12]

[3.73]





[-1.35]

[-2.23]

[-4.66]

[3.56]

[-1.23]

[2.55]


[3.15]


[-4.56]
Regional Travel Cost Model
  (pooled across set of site -1)
  -.68
(-1.49)  [-1.50]

-------
                                                                          3. 7
Table 1 (continued)
      Variable
	Price, Elasticity of Demand	

     Full               Excluding
 Specification    Judgemental  Variables
Linear Demandd
  (=1)

Log-Linear Demand4
  (-1)

Semi-Log Demand
  Dependent Variable11 (-1)

  OLSa
  (-1)

  GLS9
  (-D

  ML-Truncation"
  (-D

Year of the Data Used
  in Each Study
  n
   2.39
 (2.15)   [2.76]

  -.22
(-0.20)  [-0.28]

   .67
 (0.55)   [0.53]

   .22
 (0.19)  [-0.30]

   .35
 (0.31)   [0.52]

 -1.44
(-1.24)  [-1.77]

  -.52
(-7.30)  [-4.63]
                                           .65
                                       185
  -.42
(-5.93)
[-4.13]
                         .45
                      185
      "The numbers in parentheses below the estimated coefficients are the ratios
of coefficients to their OLS estimates of the standard errors.  Those in brackets
use  the Newey-West  [1987]  estimates  of the  standard errors allowing  for a
generalized form  of  heteroscedasticity and autocorrelation.

      bThe omitted category is coastal area and wetlands.

      cThe omitted category is  using a  wage  model to  predict  an  individual
specific usage.

      dThe omitted category is a semi-log using independent variables.

      "The omitted category is a simultaneous equation estimator.

-------
                                  CHAPTER  3                                3i 8

                                  Footnotes


1.    Hotelling's [1947] letter originally described the method as follows:

            Let concentric zones be defined around each park so that
            the cost of  travel to the  park from all points  in one
            of these zones is approximately constant.  The persons
            entering the park in a year, or a suitable chosen sample
            of'tihem, are to  be listed according  to the zone from
            which they came.   The fact that they come means that the
            service of the park is at least worth the cost, and this
            cost can probably be estimated with fair accuracy....A
            comparison of  the  cost of  coming  from a zone with the
            number of people  who do  come  from it, together  with a
            count of the population of  the  zone,  enables us to plot
            one  point  for each  zone  on  a demand curve for the
            service of the park.  By a judicious process  of fitting,
            it should be possible to get a good enough approximation
            to this demand curve to  provide,  through integration,
            a measure of consumers'  surplus...


2.    Recent  reviews  of  this  literature  include  Ward and Loomis  [1986],
Bockstael, McConnell, and Strand [1989] and Smith  [1989].

3.    The classification of  published  and  unpublished is somewhat misleading.
We used the most complete source for developing our estimates and did not include
a separate summary for a thesis that  subsequently led to a published paper.  In
some cases, unpublished  Ph.D.  theses have  yielded papers after our review was
completed.

4.    There have been several approaches to this problem in the literature.  All
impose restrictions on the modeling of recreation decisions, based on a priori
judgments.  For  example,  regional  travel  cost  models  restrict the  demand
parameters for collections of sites  in the same  general  area to  be  equal (or
change in systematic ways with specified characteristics).  The varying parameter
framework is similar but uses sites  drawn  from anywhere in the U.S., provided
they supported comparable recreation.  The  random utility models identify a set
of characteristics and  the group of  sites  assumed to comprise the choice set.
There are other formulations  as well.   None follows directly  from  theory.  Each
requires  a different  set of assumptions  about how people make recreation
decisions.  Our data do not include random utility models.  As of  1986, too few
studies  used this  framework  to  distinguish  it from  results  based on  more
conventional demand models.

5.    The travel  cost  model  is  usually described  as a derived  demand  for a
recreation site's services because each visitor produces  recreational activities
(e.g., fishing, hiking, swimming, etc.).  If we assume the household production
functions for these activities are different,  then we would expect differences
in the site demands depending on the activities undertaken.

-------
                                  CHAPTER 3                               3>9
                                  References

Bockstael,  Nancy  E. ,   Kenneth  E.  McConnell  and  Ivar  E.  Strand.     1989.
      "Recreation," in J.B. Braden and C.D. Kolstad, editors,  Measuring Demand
      for  Environmental Commodities,  unpublished  manuscript,  University  of
      Illinois.

Cordray,  David  S.    1987.   "Strengthening  Causal  Interpretations  of  Non-
      Experimental Data: The Role of Meta-Analysis," unpublished manuscript  to
      appear irt.-L. Sechrest, J.  Bunker, and  E. Perrin,  ed. ,  Improving Methods
      in Non-Experimental Research (Menlo Park,  Co.:  Sage Publications).

Clawson, Marion.  1959.  Methods of Measuring  the Demand for and Value of Outdoor
      Recreation. Reprint No. 10  (Washington, D.  C. :  Resources for the Future),
      February.

Glass,  G.  V.    1976.  "Primary,  Secondary,  and  Meta-Analysis  of  Research,"
      Educational Researcher.  Vol. 5,  No.  1,  3-8.

Hedges, Larry V. and Ingram Olkln.  1985. Statistical Methods for Meta-Analysis
      (Orlando, Florida:  Academic Press).

Hendry, David F.  1983.  "Econometric Modeling:   The 'Consumption Function'  in
      Retrospect," Scottish, Journal of Political Economy 30 (November): 193-220.

Retelling, Harold.  Dated 1947.   Letter to  National  Park Service in An Economic
      Study of the Monetary Evaluation of Recreation in the National Parks
      (U. S. Department of the  Interior, National Park Service and Recreational
      Planning Division, 1949).

Newey, Whitney K. and Kenneth D.  West.  1987.   "A  Simple  Positive Semi-Definite,
      Heteroskedasticity  and  Autocorrelation  Consistent  Covariance  Matrix,"
      Econometrica 55  (May): 703-708.

Smith, V. Kerry and Yoshiaki Kaoru.   1989. "Signals versus Noise:  Explaining
      the Variations in Recreational Benefit Estimates," unpublished paper, North
      Carolina State University, revised June.

Smith, V.  Kerry.  1989. "Travel Cost Recreation Demand  Methods:   Theory and
      Implementation,"  unpublished paper,  North  Carolina  State University,
      revised May.

Trice,  Andrew H.  and  Samuel  E. Wood.    1958.     "Measurement  of  Recreation
      Benefits," Land  Economics  34 (February):   195-207.

Ward, Frank A.  and John B. Loomis. 1986.   "The  Travel Cost Demand Model as an
      Environmental Policy Assessment  Tool:   A  Review of Literature," Western
      Journal of Agricultural Economics 11:  164-78.

White,  Halbert.  1980.    "A Heteroskedasticity  Consistent  Covariance Matrix
      Estimator and a  Direct Text for Heteroskedasticity" Econometrica 48:
      817-38.

-------
                                                                          3. 10


U.S. Water Resources Council. 1983. "Economic and Environmental Principles and
      Guidelines for Water and Related Land Resources Implementation Studies,"
      (Washington, D.C.:  U.S. Government Printing Office).

U.S. Department of the Interior, Office of the Secretary. 43CRF Part 11, "Natural
      Resource Damage Assessments:  Final Rule,"  Federal Register 51  (No. 1148,
      August 1): 27673 - 27753.

-------
                   CHAPTER 4
NEARLY ALL CONSUMER SURPLUS ESTIMATES ARE BIASED
                 V. Kerry Smith

-------
                                                                          4.1
                                                                 April 4,  1989



                Nearly All Consumer Surplus  Estimates Are  Biased


                                V. Kerry Smith*


I.  Introduction

     After eight years of a national mandate for benefit-cost analysis in the

evaluation of new major regulations,  today benefit measurement is a

significant preoccupation of many resource economists.   A series of recent

papers  (beginning with Bockstael and Strand [1987]) have raised important

questions about how we evaluate demand models intended for benefit measurement.

While the primary focus of this work has been travel cost recreation demand

models, the issues they raise are general and equally relevant to benefit

measures derived from single equation demand models for any commodity.  By

recognizing that the consumer surplus estimates are random variables, these

authors have argued for greater attention to the construction of interval

rather than point estimates, especially when these can reflect the variation in

benefit measures arising from estimation uncertainty.

     Some authors have maintained that these effects should influence the

selection of a functional specification for the demand model.  For example,

Adramowicz et al. [1989]  concluded their simulation analysis approximating the

sampling distributions for consumer surplus estimates by suggesting that:

          "...for the linear and semi-log forms price parameter estimates
          close to zero create instabilities, a feature not exhibited by
          the double log and linear log forms.  The analyst should be
          aware of this in examining his or her results.  Hence, i,f two
          forms are relatively similar regarding overall fit (judged via t
          and F statistics), but one has a smaller variance of the
          associated welfare measure, that form should be selected"
          (p. 12, emphasis added).

-------
                                                                            4. 2



Bockstael and Strand do not consider this issue.   They focus instead on what




the analyst assumes is the source of the model's  error because this source



motivates different ways of constructing consumer surplus estimates.




     This papier argues that these discussions have overlooked an important



aspect of the estimation strategies used in most  applied recreation demand



modeling.  Estimators are selected to provide the "best" estimates of the



specified demand function without necessarily considering how these parameter



estimates would be used.  Indeed, most of the consumer surplus estimates used



for policy purposes (see Smith and Kaoru [1988] and Walsh et al. [1988]) are




derived from studies that were not specifically intended to derive benefit



estimates for the recreation resources they studied.  They sought to illustrate



new estimators, test hypotheses  (e.g., alternative treatments of the




opportunity cost of time), or evaluate the effects of functional form.  The



Adramowicz et al. conclusion suggesting that properties of the consumer surplus




estimates should be considered in selecting a final specification for estimated




recreation demand models raises  a more general issue.  If benefit measurement




is the objective, shouldn't we use estimators defined to enhance the




performance of our welfare estimates rather than modifying the criteria for




selecting a functional form to "adjust" for the performance of conventional,




"general purpose" estimators with some specifications for demand models?



     To motivate further consideration of this question, I develop  three




points.  First, most conventionally estimated demand functions will yield




biased consumer surplus measures.  Because of these results, the selection of a




demand specification solely on the basis of the variability in consumer surplus




estimates can be misleading.  There is no assurance that tightly clustered

-------
                                                                             4.3



estimates about the wrong central tendency are better than more dispersed



estimates about the true value.



     Second, I derive an alternative estimator for consumer surplus per unit of



use.  This new..method accepts bias in estimated consumer surplus and seeks to



minimize the mean squared error in the consumer surplus per unit of use.  The



semi-log form is used to illustrate the method because it was found to cause



problems in the Adramowicz et al. study and it is the simplest to implement.



     Finally, I conclude by discussing issues associated with implementing the



estimator and by presenting some evidence from an illustrative Monte Carlo



study.







II.  Properties of Marshallian Consumer Surplus Estimates



     Because the consumer surplus (CS) estimates derived from most popular



demand specifications are nonlinear functions of the estimated parameters, they



will be biased even if the demand specification is correct!  Table 1



illustrates this point using three common specifications for travel cost demand



models.  The estimated (Marshallian) consumer surplus per unit demanded is


                                           2
reported for each form in the first column.   The next two columns report the



approximate variance and bias associated with the ordinary least squares (OLS)



estimates of these demand models.  A second order Taylor Series approximation



was used to develop these relationships.



     Several aspects of the derivations should be noted.  As the first row



indicates, the semi-log form requires the least additional assumptions for



measures of "average" consumer surplus per unit.  The other two forms require



further explanation.  In the case of the linear form, the sample mean was

                                                                   4

assumed to be the level of use and was treated as a random variable  with the

-------
                                                                             4.4
            TABLE  1:  Approximate Properties of Consumer Surplus per Unit
                             Across Demand Specifications
   True
   Model
                          CS/q
      Var(CS/q)
                                           Bias (CS/q)
    - a - 0P + u
                            1
                           T-

                            0
      no finite
      moments
                                          no finite
                                          moments
q - a - 0P + u
                                          -2
                                          q
               Var(q)   Var(0) (1+40)

               ~?	T~
                              q Var(0)  times
                                " 2ft +  1 "
                                                                                20-
    - a -
                ub   P
                                - 1
                                                                             K2Var(0)
are:
     aThe expressions for the approximate variance and bias of CS/q in the semilog case

        Var (0)       Var (0)
        	T	 and 	r	 respectively.
          0             0

However, we know from earlier research (e.g., Bergstrom [1962], Zellner [1978]) that the
maximum likelihood estimator of CS/q will not have finite moments.  Closed expressions
for the variance and bias would therefore be incorrect.  This outcome reflects one of
the hazards of using approximations to characterize the properties of nonlinear
functions of random variables.  This finding does not imply that mesures of the location
and scale parameters of the distributions for alternative estimtaes of CS/q could not be
derived, and thus provides motivation for the sampling results reported later in the
paper.

      The definitions for the constants involved in these expressions are given as
follows:
           (1-0)'
                  (k
                    1-0
                        - 1)
  P

(1-0)
log k)
1

2
             2P
           (1-0)'
                        - 1) -
                                 2P
                               (1-0)'
                                      (k'
         P    1 B        2   P
log k) + — k1''' (log k)z - 	
        1-0
                                         1-0

-------
                                                                            4.5

log-linear form, the choke price was assumed to be a multiple (k)  of the  price

selected for the evaluation, and the quantity is assumed to be the predicted

quantity that would correspond to that price.  As Adamowicz et al. suggest,

there are numerous possible ways of treating the upper price limit used for

this case.         -  .. -.

     These selections imply that the variance and bias for each estimated CS/q

measure are not exactly comparable across specifications.  This is not crucial

to the argument because the objective of the table is to illustrate that even

when the true specification is selected (an assumption underlying the

derivations in each row of the table), the resulting consumer surplus estimates

will be biased.  The magnitude of the bias will depend on how each estimate is

computed, what is assumed about other potential sources of error, the

performance of the estimated demand models in each case, and the  true values

for the underlying parameters.

     The reason why the semi-log form leads to CS/q estimates with substantial

estimated variance for small values of ft is clear.  They do not have finite

moments. However, to evaluate whether there .would be improvements using another

form, one must consider the bias arising when semi-log is the true demand

specification and either the linear or log-linear is adopted because of

perceived instability in the benefit estimates.  This is not reported in the

table; it it the information needed to judge the merits of the Adamowicz et al.

proposal.

     The table does illustrate that the strategy they propose in  their

concluding remarks (cited above) is inappropriate.  The perceived variability
                                                       A
in approximate expressions for the scale parameters of CS/q, such as the

approximation used for the variance, can arise simply as a result of the

-------
                                                                            4. 6

magnitude of the true value for ft in this relationship for the approximation
                    A
for the variance of 0.  Instead, we should consider how estimators of the

demand function's parameters might be designed to improve the properties of the

consumer surplus, estimates they yield.



III.  An Alternative Strategy

     A simple example based on a variation of an estimator originally proposed

by Theil [1971] can be used to illustrate a different strategy for benefit

estimation.  Consider the minimum mean squared error (MMSE),  linear estimator

of the consumer surplus.  Taking the semi-log specification for the demand

function (which is both most "popular" and regarded as among the most

unstable), a straightforward derivation of this estimator is possible.  In this

case, the estimated CS/q, designated now as s, is given by 1/0.  The general

form for this estimator is given in equation  (1).  The tilda (-) is used to

distinguish this estimator for ft from the ordinary least squares estimator used

in the derivations in Table 1.

                              1     T
                         8	CTq                           (1)
                              P

                         where q is a T x 1 vector of observations for the
                         log of the quantity demanded of the service of a
                         recreation site for each of T individuals measured as
                         a deviation from the mean of log q.

If we consider only the case of models with quantity as a function of price, as

in (2), then (3) and (4) describe expected value and variance for s with the

assumption of classically well-behaved errors,


              q - - ft P + u                                           (2)

                  with P a T x 1 vector of travel costs

-------
                                                                            4.7


              E (I) - -ft [CTP]                                         (3)


              Var (s) - (72CTC                                         (4)

                           2         T
                    where a I - E(u u )


The mean squared, error for s is defined in equation (5).   After solving the


conditions for a minimum of (5), we have equation (6)  as  one expression for the


corresponding estimator.




          MSE (I) - (- 0[CTP] - - )2 + a*CTC                         (5)

                                ft

                           -T-    2   -1 -T-
               s  - - — r (P1? + a /ft) L PXq                          (6)
     As with the solution to Theil's original problem, the estimator is a

                                   2
function of true values for ft and a , which are not observable.  Nonetheless,


operational counterparts can be defined.  For example, Farebrother [1975]


proposed that Theil's estimator could be implemented using consistent estimates

          2
of ft and a  in place of the true values.  By an analogous argument, a


consistent estimator for s can be defined.  In what follows, the OLS estimates

           2
for ft and a  will be used and the estimator designated as the approximate


minimum mean squared error method (AMMSE) .    •••


     Implementing this estimator when the focus is on a single parameter is


straightforward and requires no new estimates.  Indeed, it can be calculated


for virtually all existing studies.  To describe the estimator in cases


involving multiple independent variables, I use the expression for the OLS


estimator derived by partitioning the full set of independent variables into


two components with the own price in one and all other specified determinants,


including the unit vector for the intercept, in the other.  Using the


expressions for a partitioned inverse of the moment matrix for the independent

-------
                                                                             4.8
variables and the cross moment with the dependent variable,  the OLS estimates


                                                                    2
of ft are given by equation (7) and the AMMSE using OLS to estimate a  and ft in
(8).
          ft - (P^P)"1 PTMzq                                         (7)


          where:  Z - matrix of other determinants of demand (including a

                      column of ones for an intercept).


                  q - vector of the log of quantity (not in deviation form).

                              T  -1 T
                 MZ - i - z (z z)  z
s
                  1   -T  -   A 2 A -1 -T
                      -   -         -  -

                     (P MZP +  " //?)   P MZq
               A2                    2
          with a  - OLS estimate of a .
With some substitutions, this can be reduced to equation (9):



                  1                                                   (9)
          s — -
                                                A        A-

          where VA - the estimated variance for ft (i.e., a  (I
                 P




An argument similar to that of Farebrother can be used to demonstrate that the



AMMSE is consistent.  However, what is likely to be more relevant for



applications is the performance of this type of estimator in small samples.







IV.  An Illustrative Sampling Study



     To fully describe the comparative performance of OLS versus AMMSE in small



samples would require extensive research along the lines of Kling's recent



experimental comparisons ([1988a], [1988b]) of random utility versus



conventional travel cost demand models.  This is beyond the scope of this

-------
                                                                             4.9



paper, so Table 2 offers instead a limited set of experiments that may suggest




some of the issues that need to be considered in improving estimates of



consumer surplus from travel cost demand models.




     Four paramejierizations of each model were considered.  Two were




hypothetical and imply values for s at either end of the range from most




applications.  Two correspond to actual estimates for water-based recreation



sites taken from Smith and Desvousges [1986].  The key demand parameter for



each is reported in the column headings for Table 2 (the intercept was held




fixed at 2.33).  Each experiment involves 500 independent replications where




OLS and AMMSE  (with the OLS estimates as the starting values) are applied to



the task of estimating s using samples of 100 observations.  The true models




include only the travel cost.  A fixed set of 100 values for travel cost was



drawn using the absolute values of random variates drawn from a normal




distribution with a mean of 20 and standard deviation of 28.  These were




invariant across replications and experiments.  The error was assumed to be an




independent normal centered at zero with a standard deviation of 5.




     Table 2 summarizes the results of these experiments.  As Adramowicz et al.




suggested and  results described as part of the discussion of Table 1 imply,




under controlled conditions the OLS estimates of the semi-log demand model



(even when it  la the correct form) lead to quite variable consumer surplus



estimates.  This pattern becomes more pronounced as the absolute magnitude of




the price coefficient declines and the corresponding consumer surplus per unit




increases.  However, two potentially important qualifications to this pattern




seem to warrant further study.



     First, the overall pattern (across the 500 replications) for the estimator




designed based on the MSE of the consumer surplus per unit is superior to OLS

-------
                                                                               4.  10
    TABLE 2:  Small Sample Properties of OLS and AMMSE:   Some Illustrative Experiments'
               s - $2.00
            OLS      AMMSE
 0 - .0473
 s - $21.14
OLS     AMMSE
 ft - .0125
 s - $80.00
OLS    AMMSE
 ft - .005
 s - $200.00
OLS    AAMSE
All Replications
Mean 2.01 _ 2.01. , 20.34 . 22.65 .
MSE 1.2x10"^ 1.2x10 6.8x10 4.7x10
n 500 500
Postive Values of
s for OLS
Mean 2.01 . 2.01 , 31.83 - 32.25 -
MSE 1.2x10 1.2x10 2.3x10 2.3x10
n 500 474
Positive Values of
s for AMMSE
Mean 2.01 . 2.01 , 30.49 . 32.32 >
MSE 1.2x10 1.2x10 1.5x10 2.3x10
n 500 473
-82.98, 8.17 -37.77 0.34
1.3x10 5.2x10 2.7x10 8.2x10
500 500
85.80 92.82 . 105.78. 103.24,
1.4x10 2.0x10 3.3x10 3.2x10
336 273
79.64 . 93.66 88.75 , 105.56.
9.4x10 2.0x10 2.3x10 3.2x10
333 267
          aThe demand intercept used in these experiments was 2.33.  n designates the
number of replications used in the summary statistics for each experiment.

-------
                                                                           4. 11



for small values of j8, considering both the estimated MSE and the bias.




Indeed, the average OLS estimate for CS/q is negative (because of large



negative outlying estimates for s).   AMMSE exhibits comparable performance to



OLS for the lowest values of s considered.  It dominates OLS in terms of




estimated MSE for s .=? $21 and based on HSE and bias for larger values of s.



     Second, these results are sensitive to the assumed procedure (i.e.



pretesting/estimation strategy) that any summary of the sampling results



assumes analysts would use in evaluating the models involved.  It is unlikely



that positive estimates of the own price effect would be accepted in most




applied work.  Because these estimates are what give rise to the outlying



negative CS/q estimates for OLS (and AMMSE), a different performance pattern



emerges if we screen estimates and assume these would be rejected.  The second




and third sets of results report the summary statistics for OLS and AMMSE when



only positive estimates of CS/q are retained to approximate the sampling




distributions.  As the number of replications (n) indicates, negative estimates




for CS/q become more important as the size of ft declines.




     Now the OLS performance pattern is less negative (and concern over the use




of the semi-log less dramatic).  OLS remains superior to AMMSE (regardless of




which is used to screen the samples) until the experiment with the largest true




value for s.  Here the record is approximately comparable.








V.   Implications




     Frequently the applied researcher is warned that data mining, pretesting,




or equivalently the active use of judgment in the evaluation of empirical




models is to be avoided because with this practice, one violates the




assumptions of classical inference and cannot claim conventional properties for

-------
                                                                            4.12



the resulting estimates.  While the analytical results underlying this




admonition are certainly correct,  they imply the sampling properties of the




resulting estimators will be different than those attributed to the




conventional '("pure") formulations.  At a general level, this analysis has



suggested that "different" may not mean "worse" in all cases.  This may be




especially true for nonlinear transformations of the estimates where judgment




can eliminate cases that are obviously inconsistent with the theory underlying



the model.



     A conclusion more specific to my objective is that there seems to be a



role for developing estimators based on the economic parameter of interest.




This strategy contrasts with one that considers the overall fit of general



behavioral models or the properties of all parameters in these models.  There



are conditions (and in the case of travel cost recreation demands they



correspond to a wide range of applications) in which the approximate  (linear)



minimum mean squared error estimator would have superior properties to the OLS




estimate of consumer surplus per unit of use.  While these results should be




carefully qualified, they do motivate consideration of  different strategies for




evaluating the stochastic properties of consumer surplus estimates.   These




alternative approaches should recognize how a model's estimates are to be used




and characterize the Judgments that are made before these estimates would be




accepted for policy applications.

-------
                                  CHAPTER 4                                 „   „
                                                                            4.13

                                     NOTES
*University Distinguished Professor,  Department of Economics and Business,
North Carolina State University.   Partial support for this research was
provided by Cooperative Agreement No. CR812564 from the U. S. Environmental
Protection Agency to Vanderbilt University.


1.   In February, 1981, Executive Order 12291 was issued.  It required a
benefit cost analysis for all major regulations where statutes did not prohibit
use of such analyses.  While there were quite disparate views as to whether
this order alone would increase the role of economic analysis in the evaluation
of environmental regulations (see essays in Smith [1984] for early
discussions), it seems now, after eight years, to have changed the way
regulations are discussed and evaluated.  Benefit-cost results are a part,  and
certainly not the exclusive part, of evaluations of new and existing
regulations.  See U. S. Environmental Protection Agency [1987] for an interval
review of the effects of the benefit-cost mandate.

2.   This is frequently the focus of benefit transfer exercises used in policy
analyses and in summary studies designed to provide valuation estimates in
anticipation of policy evaluations.  The reviews by Sorg and Loomis [1982]  and
Walsh et al. [1988] for the U. S Forest Service as part of implementing the
multiple use and sustained yield legislation are examples of this approach.

3.  See Mood, Graybill, and Boes [1974], pp. 180-82, for examples and
discussion.

4.   Bockstael and Strand  [[1987] used the total consumer surplus.  By adopting
this formulation, I avoid consideration of the source of the errors.

5. The derivation underlying these results does not drop the covariance terms
as Adramowicz et al. [1989] does.

6.    Another approach first identified as relevant to this problem by
Bockstael et al. [1984] was developed by Zellner  [1978].  It is the minimum
expected loss estimator.  Zellner's loss function expresses the mean squared
error relative to the true value of the parameter.  For the semilog model,  it
offers a direct estimate of the cs/q as s*

                 I
                 ft J
                               f
It was also evaluated in the sampling experiments reported below but was found
to be inferior in all cases to the AMMSE estimator.

7.   Clearly this practice must be used cautiously.  It would prevent rejecting
existing theory based on contradictions.  This is not what I intend to imply
here.  Rather my focus is on maintained theory underlying a model that is not
subject to test in any specific application.

-------
                                   CHAPTER 4                                4.14

                                   REFERENCES
Adamowicz, Viktor L., Jerald L. Fletcher and Theodore Graham-Tomasi.   1989.
     "Functional Form and the Statistical Properties of Welfare Measures,"
     American Journal of Agricultural Economics (May), in press.

Bergstrom, A.R.* 1962.  "The Exact Sampling Distribution of Least Squares and
     Maximum Likelihood Estimators of the Marginal Propensity to Consume,"
     Econometrica 30 (July):  480-490.

Bockstael, Nancy E. and Ivar E. Strand, Jr.  1987.  "The Effect of Common
     Sources of Regression Error on Benefit Estimates," Land Economics 63
     (February):  11-20.

Bockstael, Nancy E., W. Michael Hanemann, and Ivar E. Strand, Jr.  1984.
     "Measuring the Benefits of WAter Quality Improvements Using Recreation
     Demand Models," Draft Report EPA No. CR1-811043-01-0, Department of
     Agricultural and Resource Economics, University of Maryland, College Park,
     Maryland.

Farebrother, R. W.  1975.  "The Minimum Mean Square Error Linear Estimator and
     Ridge Regression," Technometries 17 (February):  127-128.

Mood, A., F. Graybill, and D. Boes.  1967.  Introduction to the Theory of
     Statistics. Second Edition (New York:  McGraw-Hill Book Co.).

Kling, Catherine L.  1988a.  "Comparing Welfare Estimates of Environmental
     Quality Changes from Recreation Demand Models," Journal of Environmental
     Economics and Management 15 (September):  331-340.

                  _.  1988b.  "The Reliability of Estimates of Environmental
     Benefits From Recreation Demand Models," American Journal of Agricultural
     Economics 70 (November):  892-901.

Smith, V. Kerry, ed.  1984.   Environmental Policy Under Reagan's Executive
     Order:  The Role of Benefit Cost Analysis (Chapel Hill, N. C.:   University
     of North Carolina Press.

Smith, V. Kerry.  1988.  "Selection and Recreation Demand," American Journal of
     Agricultural Economies 70 No. 1 (February): 29-36.

Smith, V. Kerry and William H. Desvousges.  1986.  Measuring Water Quality
     Benefits (Boston:  Kluwer Nijhoff).

Smith, V. Kerry and Yoshiaki Kaoru.  1988.  "Signals or Noise?:  Explaining the
     Variation in Recreation Benefit Estimates," unpublished paper,  North
     Carolina State University,  November.

Theil, Henri. 1971.   Principles  of Econometrics (New York:  John Wiley & Sons).

-------
                                                                            4.15

U. S. Environmental Protection Agency.   1987.  EPA's Use of Benefit-Cost
     Analysis:  1981-1986 (Washington,  D.C.:  Office of Policy Planning and
     Evaluation).

Walsh, Richard G., Donn M. Johnson, and John R. McKean.  1988.  "Review of
     Outdoor Recreation Economic Demand Studies with Nonmarket Benefit
     Estimates, 1968-1988," Colorado State University, Fort Collins, Colorado,
     Unpublished'Paper (June).

Zellner, Arnold.  1978.  "Estimation of Functions of Population Means and
     Regression Coefficients Including Structural Coefficients:  A Minimum
     Expected Loss (MELO) Approach," Journal of Econometrics 8 (October): 127-158,

-------
               CHAPTER 5
DEMANDS FOR DATA AND ANALYSIS INDUCED BY

          ENVIRONMENTAL POLICY
          Clifford S. Russell
                  and
             V. Kerry Smith

-------
                                                                           5.1
                    Demands  for  Data and Analysis  Induced by
                              Environmental  Policy

                                      by
                    Clifford S.  Russell and V. Kerry Smith*
I.  Introduction

    Economic analysis of environmental policies is, if not uniquely,  at least

unusually difficult.  Resolution of these difficulties requires substantial

investment in data collection and model construction, only some of which is

directly economic. Some of the reasons for the difficulties of environmental

benefit-cost analysis are well known, appearing in intermediate and even

elementary microeconomic and policy analysis texts.  Thus, even the average

economics undergraduate major can be expected to appreciate that there is a

problem finding demand functions for many services of the natural environment

because they are public goods.  At more advanced levels, they will learn about

such thorny technical issues in implementing proposed solutions to this

problem.  Those interested in policy learn about the conflicting maze of

environmental legislation, including problems of overlapping jurisdiction,
     ^Respectively, Director, Vanderbilt Institute for Public Policy Studies,

Vanderbilt University; and University Distinguished Professor of Economics,

North Carolina State University.  Partial support for this research was

provided through U.S. Environmental Protection Agency Co-operative Agreement

No. CR812564.  Thanks are due Ernie Berndt, Tom Tietenberg, Peter Caulkins,

Bill Desvousges, and Paul Portney for their suggestions on research related to

this paper.

-------
                                                                           5.2
differences in burdens of proof and,  most significantly, disagreements between



laws over what role, if any, economic analysis should play.




     But neither the technical economic matters nor the special policy




problems, challenging though they may be, provide the principal explanation for




our assertion that the benefit-cost analysis of environmental policy may well




be uniquely difficult.  Rather, that assertion is based on the central place in



such analyses of the complex relationship between policy implementation choices



on the one hand and the relevant natural systems (especially atmosphere, water




bodies, soil and resident plant communities and  ground water) on the other.




     To set the stage for a more careful examination of this assertion, let us



consider the nature of the system of environmental regulation and the origin of



the complications in which we are especially interested.  Figure 1 combines an



overview of the linkage between policy choice and resulting benefits, with




indications of the complications arising at each stage in the linkage.  In the




next three sections we shall examine in turn each of the links in the figure:




          In section II, we shall describe some of the problems implied by the




          way standard setting is constrained and practiced and by the



          necessity of choosing an accompanying implementation plan;




          In section III, we concentrate on the central role of knowledge of




          natural system* interacting with choice of implementation system;




          In section IV we come to some of the more obviously and traditionally




          economic issues connected with valuing environmental services.




     The final section of the paper brings together the analysis of sections




II-IV with a brief assessment of key emerging policy issues to produce our




version of a catalog of data-gathering strategies likely to be most relevant

-------
                                                                           5.3
and valuable for analyzing future decisions on the allocation and management of


environmental resources.


     While our approach to identifying data needs builds from specific  examples


of current policy issues, the questions raised are general ones.   Thus,  we do


not attempt to catalog.what we consider to be the most important  environmental


policy issues in 1988 and then base an evaluation of data requirements  on them.


Instead we argue that the interactions between the statutes defining the


character of environmental policy and the role of natural systems for


economic agents' behavior affect the problems that would appear on any  list


that might be composed.  Thus, regardless of whether one believes global


warning or indoor air pollution is among the most pressing environmental


questions, economic analysts will need to incorporate what is known about a


form of these interactions in developing their analyses




II.  Choosing Standards and Implementation Plans


     Table 1 (adapted from EPA 1987a) summarizes the major criteria to  be


considered by the Administrator of EPA in deciding on standards under a variety


of legislative mandates.  Two features of this table are especially striking.


First, the criteria used to choose standards frequently focus on a subset of


the information that would be part of a full benefit-cost analysis.  For


example,  under the Clean Air Act the primary standards for criteria air


pollutants are to be based on human health effects but cannot include

                                                         2
compliance costs in the process of defining the standard.   In contrast, under


the Clean Water Act, the definition of one type of technology-based standard,


Best Conventional Treatment (BCT), can be based on costs (in comparison with


the marginal costs of secondary treatment at municipal waste treatment

-------
                                                                           5.4
facilities) but not on the specific benefits to be realized at individual water



bodies.




     Second, the mandates involve significant areas of overlap where different



regulatory analyses are intended to influence the same types of pollutants in




the ambient environment, e.g.: primary standards for Criteria air pollutants



and New Source Performance Standards defined on the basis of the effects new




discharge sources would have on the concentrations of these pollutants.




     One important implication is that economic analysis (in this case,



benefit-cost analysis) usually involves an evaluation of the net




effect of standards chosen on some basis other than economic efficiency.



Another is that standards set under one provision of one law may well overlap



in their effects with those set under another provision or law. This raises



difficulties for the definition of benefits --at least whenever marginal




benefits are non-linear -- because of the interdependence of baselines.




     Other serious problems introduced by the standard setting operation can be



considered in a few specific examples.  Setting an environmental standard of



either the discharge or ambient sort requires that  the regulator must:




(Richmond 1983)



          identify the pollutant to be regulated.



          select the form of the standard (i.e., a technology to achieve an




            emissions rate or an ambient concentration).



          choose the concentration or discharge amount  that will be the average




          target.



          pick the averaging time over which the target is to be met (an hour,




          a day, a week, a year...).

-------
                                                                           5.5
          define the exceedance rate(s) of interest (e.g.:  a weekly average




          standard might be paired with a daily upper limit).




          define what constitutes a violation, taking account of the



          statistical error structure displayed by the monitoring equipment and




          other relevant sources of uncertainty (such as measurements made




          across a sample of different applications of a technology where the



          standard is technology based).




     Thus,  evaluating the net benefits of an environmental  standard is a



complicated business. Not only do we lack information on the effects of average




(or peak, as applicable) exposures to particular pollutants (or ecological



effects of average concentrations), we also should be able to evaluate




alternative patterns of allowed exceedance.  In practice we are fortunate if we




have the data from which to estimate dose-response functions over any range and



averaging time.




     The case of the particulate matter (PM) ambient air quality standards




permits us to see some of the troubles that can arise even within this limited




context.  The benefit-cost analysis done for PM was the most expensive of those




discussed in the EPA report cited above (EPA 1987, FN 1). It seems reasonable




to assume that the quality of the analysis reflects these expenditures.




     The first and largest problem in analyzing PM benefits was that the




available laboratory evidence on the health effects of airborne particulates




did not match up with the available ambient measurements.  Laboratory




toxicology suggested that particles smaller than 10 microns across were




responsible for whatever health damage was observed.  Since preventing health




damage was the mandated basis of the standard, the ambient standard had  to be




written in terms of these small particles.  Ambient measurements, with a few

-------
                                                                          5.6
isolated exceptions, had for years been done in terms of total suspended




particulates (TSP).   As a consequence,  epidemiological studies aimed at finding




health effects associated with airborne particulates inevitably labored under



an imposed errors-in-variables problem.




     More fundamentally, however, analyzing the total net benefits of the 10-




micron PM standard required that the relation  between TSP and the distribution




of particles by size, both before and after a standard, be understood.  In




addition, the analysis does not end with health because other benefits  could be




identified that depended on other sizes of particulate matter.  In fact, the PM




study conducted by EPA (and subcontractors) involved separate assessments of




the health benefits (including mortality and morbidity effects), the household




benefits from  reduced soiling and materials damages, and the benefits  to the



manufacturing  sector from reduced soiling and materials damage.  The first two



relied on judgmental appraisals (see MathTech [1983]) of the "best" estimates



of dose response relationships and the last two involved the development of new




models linking consumer expenditures (on commodities related to household




cleaning) and sectoral cost functions to measures of particulate




concentrations.



     The importance of the institutional setting in combination with the



technical and natural systems also can be seen in the cost estimates for the




PM standard.  Developing these estimates required a specification of how




states would formulate their State Implementation Plans (SIPs), the degree of




compliance with the plans, and the resulting estimated levels of particulate




emissions.  Emissions then had to be translated into estimates of the ambient




concentrations of particulates.  Of course, uniform ambient air quality




standards do not imply uniform levels of actual air quality, a point we return

-------
                                                                           5.7
to in Section III.  The changes in air  quality from a specified baseline




defined spatially will depend on how  the assumed SIP describes the process




(the set of discharge reductions) used to  meet the standard in each air



quality control region.




     To stress the analytically arbitrary nature of the institutional context,



we report an example drawn from Liroff [1986].  When states decide how to




achieve the National Ambient Air Quality Standard (NAAQS) for a pollutant,  they



may have a choice among different average levels of emission reduction



depending on which mathematical model of the local atmospheric system they



choose to use to predict ambient concentrations.  Table 2, based on Liroff's




Table 2.2, shows the choice facing Ohio in designing its implementation plan




for ground-level ozone.  The two alternative models lead to alternative



patterns of predicted ambient concentrations, though both would show no



violation at any monitoring site.   Thus, the predicted net benefits of the




ozone NAAQS in Ohio (and generally in any  state) will depend on the choice of




modeling technique, not just on the average level of the standard.  Of course,




it is possible that either or both models may be wrong.  Neither pattern of




reductions might in fact result in meeting the NAAQS.



     Let us now turn to natural system information and modeling, and the




implications of how we handle such matters for our estimates of the benefits of




environmental standards








III. Bringing in the Natural World



     The evaluation of environmental policies inevitably involves economists




with the systems that make up the ambient environment.  If a policy mandates a




reduction in polluted waste water discharges from industrial and publicly owned

-------
                                                                          5.8
sources, the streams, rivers, lakes, and ponds that constitute the receiving




waters translate the discharge reduction into ambient quality improvements that




are valued by individuals.  Turning this notion around, if public policy




involves mandated upper limits for concentrations of pollutants in the ambient




atmosphere, the transportation, dilution, and transformation processes at work




in that atmosphere must have a key role in determining how much discharge



reduction has to be accomplished to meet the standard.




     While this seems intuitively clear, the importance of knowledge of those




processes is greater than these observations suggest.  There are two reasons



for this.  One is ubiquitous; the other is found to be central to some




situations and not to others.  The ubiquitous influence is space, the




differential location of pollution sources and pollution receptors in the two




dimensional plane.   Additional complication is introduced by the non-linearity




of most environmental processes.



     Consider the role of location.  In the simple situation, a policy ia




represented graphically or mathematically by a single marginal benefit (or




damage) and a single marginal cost function.  These may have as arguments




either ambient pollutant concentrations or pollution discharged.  The optimum




policy is defined by the usual MB-MC condition.  The addition of spatial detail



merely replicates this condition at each location.  That is, efficient policies



equate the marginal benefits to the marginal costs of realizing a given level




of ambient quality at each location.  In conventional Pareto efficiency terms




this corresponds to equality of the relevant sum (for that location and its




residents) of the marginal rates of substitution for environmental quality (in




relation to a numeraire) to the corresponding shadow price describing the real




costs of attaining it.  The natural system is implicit in the definition of the

-------
                                                                           5.9
real marginal costs.   When perfect mixing of all pollution discharges produces


uniform concentrations of pollutant everywhere in the ambient environment --as


is roughly true for some air pollutants under certain physical and


meteorological conditions -- the simple model offers a reasonably good


approximation.


     But in the largest number of cases, it does not.  For a mandated policy of


emission reductions, even if that policy involves equal percentage reductions


at all sources, the amount of ambient quality improvement will, in general, be


different at every point in the relevant environmental medium.  If the policy


to be evaluated involves mandated ambient quality standards, the situation is


even more at odds with the simple model.  Not only will the concentration in


the standard characterize only a few points in the environment after the policy


is implemented, but which points those are and by how much the quality at every


other point is better than the standard will, in general, depend on exactly how


the standard is implemented.


     Both environmental quality levels and, more important, improvements in


quality attributable to a policy are different at every point in the


environment.  Moreover, every point is usually characterized by different


levels of human "use.*  Thus, for example, some points in the atmosphere


coincide with dense residential population, some with sparse; some coincide
                                                          \
with industrial plants, some with office buildings,  some with vacant space.


Similarly, along a river some segments will have heavy recreational use  (or


prospectively have such use) because of conditions of access, bank type,


current, and temperature.  Other segments may be unattractive to recreationists


for reasons having nothing to do with the level of pollution at that location.

-------
                                                                           5.10
     Therefore, the estimates of benefits of proposed (or actual) environmental
management policies are intrinsically dependent on the accuracy of our
knowledge of the natural world processes, upon the detail required for the
spatial resolution and on the implementation plan assumed in the analysis.
The net benefits of a given policy, P, can be written in fairly general terms
as follows :
(1) NB (P) - BL 
-------
                                                                           5.11
discharge at every source, that defines the vector •JD1 (P)	D  (P)k

                                                   I 1          n    J


These discharges are transformed by the functions f.  (D.) into ambient



pollution (or "quality levels; and the resulting quality at each receptor



location is valued using the functions B, (  ).  If, on the other hand,  the
                                        K


policy P requires an upper limit on ambient pollution at any receptor location,



call it S^, analytical implementation implies funding a vector of discharges



satisfying the requirement.  This will depend on the functions f.  (D.), for we



are solving a problem of the following form:




     find T>i (P) such V  Z f   (Dt (P)) * S^.







 This is different than the description in textbooks because the policy is not



defined to meet an efficiency criterion.  We simply use  (1) to evaluate how its



implications relate to the net benefits realized with some baseline or status



quo position.  There may be no such vector. More often, since n > m, there will



be an infinite number.  The benefits flowing from the choice S. will depend on



which vector D (S, ) is evaluated. This is because every such vector will, in



general, produce a different pattern of ambient quality across receptor



locations. Further, in this general formulation, there is no presumption that



quality better than the standard is valued at zero.



     To illustrate what happens if we ignore the natural system, we offer one



very simple and two not-so-simple examples.  First, consider a hypothetical



region with two sources of air pollution and three receptors or agreed-on



monitoring locations.  The sources are, in fact, linked  to the receptors by an



atmospheric system that can be characterized by a matrix of transfer



coefficients, T, as follows:

-------
                                                                           5.12
                                   Receptor



                                   I     II     III



               Source:  A          2      1     0.5



                        B          122
Ambient quality, Q, is determined on the basis of source discharges as:
     (2) Q - DT where D - (DA> Dfi)
And the benefits of discharge reductions are assumed obtainable, as damages



avoided, from a quadratic damage function.
                    2                    8
      (3)  G.(Q) - Q.  for each receptor i
If initial discharges are D _ - 4, D__ - 2, the base or  initial quality levels
                           AU   •    BO


are:9
      (A)   {Qio> -  (10,8,6)
with resulting damages:






           III



     (5)   S    G± - 200



           i-I

-------
                                                                           5.13
     The effect of what we might call environmental ignorance is illustrated by



illustrated by considering  three different ways of evaluating the benefits of



setting increasingly stiff ambient quality standards,  S.:



     (1)  We know nothing about the environment (in particular, we do not know



          T),  so we simply work from the regional average concentration before



          the standard is set and assume that the standard is the average



          concentration after it is set.  Let us denote this approach to



          estimating benefits as method 1. designatsd B ,   Then:
                           B/-3
                                            Q
                                             .
- G(S.)

     J
          where j indexes the severity of the standard.



     (2)  We still know nothing about the atmospheric system (T) but



          disaggregate benefits.  In this formulation, benefits are calculated



          only for receptors where the initial quality level is worse than the



          standard.  Moreover, it assumes that at every such point, after the



          imposition of the standard, quality just equals the standard.  This



          is method 2 (B2), given by (7):
          (7)                 B2 - 2 G(Qlo)  - G(S.j)
          for all i such that Q.  > S..
                               io    j


      (3)  We know and use T.  Implementation policy  is  a  "roll-back"  rule  from



          base period discharges.  That  is, with  particular  standard,  S.,  the

-------
                                                                           5.14
          roll-back rule specifies that each discharge is reduced by the

          proportion R., given by:



         (8)                        Max (Qio) - S
                            R. ™         .- - -1.    J
                                    Max
          so that benefits are
         (9)               Bj - Z  |G(Qlo) - G Q(R  ))
          where
         (10)             Q
-------
                                                                          5.15
any particular nonbinding receptor is reduced,  and with it the  sources  for the


                     12      3
differences between B '  B  and B  diminish.



     The marginal benefits calculated ignoring the natural system are an



especially unreliable guide to optimal policy choice.   These results are not



simply artifacts of our example.  Two more realistic cases illustrate the peril



of ignorance of the natural world's systems.  The first is based on the data



developed for the Baltimore, Maryland, region in the paper by McGartland, et



al. [1988], using their air quality results (for total suspended particulates)



and translating them into versions of our surrogate benefit measures. The



primary difference is that Method 3 reflects a least-cost rather than  a roll-



back scheme for implementation.  So we refer to it as Method 3' (see Appendix A



for data and methods).  Table 4 contains a summary of the results for total and



marginal (surrogate) •benefits for each estimation approach or level of



knowledge.  The marginal benefits calculated by McGartland, et al. are  shown at



the bottom of the table.



     Thus, in in a much more realistic example, the methods that ignore the



natural environment produce problematic estimates of marginal and total



benefits.  Method 1 shows no benefits because the base case average TSP



concentration is already below all the standards considered.  Method 2  produces



substantial underestimation of both marginal and total benefits.  It ignores



improvements at receptors that have quality better than the standard before it



is imposed.



     Modified Method 1 depends on simple reduction of the average TSP



concentration for the region for each standard level by the same percentage as



that standard represents a reduction of the baseline standard that McGartland,



et al. use in their benefit calculations, 120 micrograms per cubic meter.   It

-------
                                                                               5.16
    produces  total  benefit  numbers roughly similar to those obtained in Method 3',

    the method reflecting best  available knowledge of the environment.  But this

    apparent  improvement does not extend to marginal benefits.  The actual pattern

    obtained  via Method 3'  shows an early peak at 110 ug/m  ,  followed by a dip and,

    then,  subsequent  increases.  Indeed, marginal benefits  are still increasing at

    the strictest standard  shown.

         Of course, one might criticize this example as well, nothing that we are

    not working with  a 'real' damage function.  Our last example does just: that;

    using  data on water quality changes, as measured by dissolved oxygen, generated

    by a complex and  quite  realistic model of the Delaware  River Estuary; a mapping

    of dissolved oxygen (DO) into sustainable recreation types from a second

    source; and an  annual per capita willingness to pay for the availability of

    water-based recreation  by type from a third source.   (The details of the data

    and calculations  are set out in Appendix A.)  The results for total and

    marginal  benefit  estimates  are given in Table 5.

         Thj  patterns of marginal benefits once again display the largest effects

    from ignorance  of the natural world.  Method 1  implies  there would be no

    benefits  of going from  the  baseline situation to a standard of at least 3.5

    parts  per million (ppm) of  DO for  every reach of the  river.  But  the marginal

    benefit of tightening the standard from 3.5 ppm to 5.0  ppm of DO  is 420.2.
L 5
    Under  Method 2  -- reach-by-reach disaggregation, but  assuming benefits only for

    reaches that are  initially  worse than the  standard  --  the marginal benefit of

    the 3.5 ppm standard is 184.5 and  that of  the 5.0 ppm standard is 326.1.  This

    pattern is almost exactly the reverse of that observed when complete knowledge

    is used in Method 3.  In this case,  the marginal benefits associated with the

    lower  standard  are 372.7, while those associated with the next improvement to

-------
                                                                           5.17
5.0 ppm are 208.7.  Thus, even though the total benefit estimated to be



associated with the tougher standard are roughly similar for Methods 2 and 3,



the marginal benefit patterns are very different.



     The results of these examples may be so obvious that their applicability



seems  doubtful.  Who would ever use methods such as (1) or (2)?  The answer --



and this is the key to our later recommendations --is just about everyone.  An



examination of the invaluable compilation of benefit estimates published by



Freeman [1982] reveals that every one of the reported air pollution benefit


                           1     2
studies uses a version of B  or B ' with most relying on a method very like



Method 1.  The water pollution benefit studies he summarizes all use a version


    2
of B  in which full attainment of the most ambitious standards (or ambient



quality goals) of the Clean Water Act (CWA) is assumed.



     As important as pointing out the prevalence of benefit estimates based on



ignorance of natural systems is an attempt to understand why the cause also



merits consideration.  In the case of water pollution control benefits, the



answer is5 generally that insufficient resources have been invested Jr. the



research needed to -reduce our ignorance.  Translating the technology-based



discharge standard definitions of the CWA into actual discharges from tens of



thousands of point sources of water pollution is hard enough.  But then



translating such changes in discharges, were they available, into changes in



water quality indicators that in turn can be valued by  individuals involves



data gathering, modeling, and basic conceptual research efforts beyond what the


                                                   12
sponsors of such research have been willing to pay.     Finally, the data on



valuation that is available generally is in the form of step functions unsuited



to the valuation of benefits of small improvements in quality, especially at



reasonably clean receptor locations.

-------
                                                                           5.18
     For air pollution benefits, the state of the art of emission inventories



and air quality modeling has for some time been capable of supporting the  sort



of disaggregated, location-specific benefit estimates obtained by McGartland et




al. [1988] for Baltimore.  When national total benefit estimates have been the




object of the exercise, however, it apparently has been too daunting a task to




manage the necessarily massive data banks and atmospheric models.



     Finally, before we turn to the next concern of this paper, the valuation



of environmental quality changes, we should consider the effects of




implementation plans on benefits.  This is the primary concern of McGartland et




al. [1988].  While their paper actually is addressed to the relevance of




benefit estimates for the choice between regulatory approaches ("command and



control" versus use of economic incentives), their results provide a fine




illustration of the point that for any given level of environmental knowledge,




estimates of benefits will depend on the method of implementation -- the




pattern of discharges -- assumed.




     Thus  in Table 6 we reproduce their marginal benefit estimates for the



command and control and  "least cost" implementation approaches.  In this case,




neither set of estimates can be characterized as "wrong."  Both  reflect the




best environmental information available.  Nonetheless, they are very




different. Thus, the statement that a particular standard yields particular




benefits has meaning only when an implementation method is explicitly assumed.




     The same standard,  treated as an upper bound on a pollutant's allowable




concentrations, can imply an infinite number of aggregate marginal benefit




patterns because these benefits will depend on how the standard  is implemented




and on what the natural  system implies this implementation plan  will yield as




the ambient concentrations for each receptor location.  In most  theoretical

-------
                                                                           5.19
treatments of these issues,  this problem is avoided by simplifying assumption.




The benefits are taken to be measured at a single, representative point in the




environment.  The costs of improving quality at that point are assumed to




reflect the environmental transformations implicitly.








IV.  Evaluation Benefits:  Learning from Past Research and Identifying




     New Initiatives13




     The statutory guidelines creating the demand for valuation measures for




environmental resources and the time horizons written into the statutes make it




impossible to develop new benefit-cost studies for each decision.  This has led




to growing interest in the methods used to transfer valuation (or demand)




estimates derived in one situation to a new one. Both the McGartland et al.




[1988] study of air quality in Baltimore and our own analysis of water quality




in the Delaware River used valuation estimates derived from one or more studies




in the literature (see Appendix A).  For the most part, these were derived from




judgmental reviews of the literature and propose a best estimate (or a range of




values).




     Because the services of environmental resources exchange outside markets,




the methods used to estimate consumers' values for them have developed along




two lines.  The first focuses on observable behavior that can be linked by




assumptions to the resource of interest.  Methods relying on this strategy have




usually been labeled the indirect approaches.  They  include: the travel cost




recreation demand models, hedonic price functions (property value and wage




rate), hedonic travel cost functions, damage-averting cost models, and factor




productivity (or reverse value added) methods.  In each case, an individual's




(or a firm's) actions are assumed to be partially motivated by a desire to

-------
                                                                           5.20
obtain the service of an environmental resource (or to avoid the detrimental



effects of pollution to that resource).  Using models based on these actions,



researchers attempt to estimate the marginal value of changes in the quantity



or the quality of the nonmarketed resource.



     The second group of methods relies on survey techniques that ask



respondents how they would value (contingent valuation) or change their



behavior (contingent behavior) in response to a postulated, hypothetical change


                                             14
in the services of an environmental resource.    This method assumes that an



individual's response to a hypothetical situation provides an authentic



description of how he (or she) would respond to an actual change.



     The purpose of this section is to suggest that efforts to summarize and



evaluate benefit estimates offer another kind of opportunity --to evaluate



what we have learned about the values of environmental resources; to examine



the sensitivity of these estimates to the modeling decisions required to



develop them; and, based on these two appraisals, to  identify new data and



analyses required to resolve the uncertainties leading to  the disparities in



valuation estimates.  The required analyses treat the results from past studies



as data to "test" whether differences in the estimates (across studies) reflect



systematic variations in the resources being valued or in  the assumptions and



the methods underlying them.



     While this approach appears to be a new one for  evaluating empirical



research in economics, it is not new to other social  and health sciences.15



"Meta analysis" describes a research method that seeks to  provide systematic



summaries of the findings from empirical evaluations  of educational or social



programs.  Du Mouchel and Harris [1983], for example, proposed a similar

-------
                                                                          5.21
strategy for the transfer of risk assessment models  from animal  to human




populations.




     Our objective is broadly similar.   However,  we  seek to  evaluate  whether




there are systematic influences on the  values estimated for  specific  types of




environmental resources and whether these influences can be  distinguished from




the assumptions and features of the methods.  Ideally,  such  an analysis would




be undertaken within a single empirical study where  consistency in data




sources, reporting conventions, and statistical modeling criteria could be




maintained across the resources and models studied.   Unfortunately,  this  was




not possible.  Consequently, we summarize the results of a pilot study




conducted by Smith and Kaoru [1988] that uses the existing literature as  the




basis for an examination of the determinants of valuation estimates for




recreation resources.  The focus on value estimates  is deliberate because,




regardless of the original objective of the research, benefit estimates have




been the single most important policy use of the outputs this type of research.




     Equation (1) defines the basic model.  To use it, we maintain that the




valuation estimate relevant for our example, the real consumer surplus, RCS,




per unit of use of a site is a function of four types of variables: the type of




recreation site, X.; the assumptions inherent in the model specification, X^;




the form of the deaand model, X_; and the estimator vised, Xg.









              (11)






where X   and a., j - S, A, D, E are conformably dimensioned vectors and e^  is




the stochastic error for the ith estimate.




     Smith and Kaoru [1987, 1988) have reviewed over 200 published and




unpublished travel cost demand models prepared over the period 1970  to 1986  and

-------
                                                                           5.22
developed a data set summarizing the valuation estimates,  features of the




resources involved in these demand studies, and characteristics of the models




involved.  The results reported here are based on 77 studies.   They yield 734




observations for the consumer surplus per unit of use.   The individual




observations vary by recreation sites, demand specification,  modeling



assumptions, and estimator used.




  There was enormous variability in the information reported across studies.




Often the objective of the research was something other than estimating the



values for a recreational facility.  It may have been testing a specific




hypothesis, with the results reported confined to the specifics of the



hypothesis test.  Smith and Kaoru did not attempt to contact individual authors



to supplement (or check) what was reported in the individual papers.  Rather




their data set relies exclusively on the information reported within these




limitations.   Table 7 defines some of the variables that could be consistently




defined across the studies in each class of variable.



     To interpret the results obtained from statistical analyses of valuation




estimates across different studies, we must formulate specific hypotheses




concerning how and in what dimensions these estimates might be sensitive to




modeling judgments.  A beginning step in this process can be found in past




literature reviews (i.e., Ward and Loomis  [1986], Smith and Desvousges  [1986])




as well as in what seem to be established conventions in developing travel-cost




demand models.  A few such protocols would include:




     (1)  Use trips as the quantity measure where possible, and attempt to




          segment the sample when it is known that  the length of stay per trip




          is different.

-------
                                                                          5.23
     (2)  Take account through sample segmentation of differences that might




          arise from use during different seasons, or during different time




          periods when there may be different time or resource constraints.



     (3)  Treat travel time as an element affecting the cost of a trip.




     (4)  Include vehicle-related costs and the costs attributed to travel time




          as well as any entrance fees or site usage costs (i.e. parking costs,



          lift fees for skiing, etc.) in the unit cost estimated for a trip.




     (5)  Use substitute prices to'measure effects of substitute sites rather




          than an index of substitution; complete systems of demand functions




          are unnecessary if the objective is to measure demand for one of the



          sites.




     (6)  Reflect quality features of the site in the demand models.



     (7)  Recognize that heteroscedasticity is likely to be an issue with zonal



          data and that selection effects can be important with individual



          data.




     (8)  Avoid the problems posed by cost allocation issues that can aris"




          with multiple destination trips by segmenting the sample according to




          the distance traveled to the site.



     (9)  Substitute sensitivity analysis for strict adherence to one




          particular functional form for the demand function.



     Equally important, areas exist for which there are either  insufficient




data  or the absence of a clear consensus.  These are:






     (1)  the measurement of the opportunity cost of travel time; simple




          scaling of the wage rate was not found  to be consistent with several




          of the demand studies based on individual data, yet explicit




          recognition of multiple prices for recreation time is generally

-------
                                                                          5.24
          beyond the'information set available in most current studies;  no




          compromise has been proposed to deal with this problem;




     (2)  the treatment of the attributes of a site's services;




          and




     (3)  definition of a recreation site for modeling demand, especially where




          there are many comparable sites within a small geographic area or




          where there is one large "site" that extends over a wide area.




What has been missing in past assessments is some gauge of how important the




decisions might be in influencing the valuation estimates that result.




     From the perspective of being able to transfer valuation estimates, we




would prefer that the empirical estimates of equation (11) be consistent with a




maintained hypothesis that o  - a_ - a_ - 0.  That is, judgmental modeling




assumptions contribute to the variability in benefit estimates but do not




impose systematic influences on the size of the benefits estimated.  Of course,




to the extent this is not our conclusion, then we believe the process has




identified areas where further research, modeling, and data collection may be




warranted.




     Table 8 provides some descriptive statistics from the Smith-Kaoru data on




the features of the studies, classified by the type of site involved.  It




reports the number of estimates for each type of resource, the mean and range




in real consumer surplus (per unit of use) estimates, the proportion of the




studies based on individual (as compared with origin zone) data, and the range




of years represented in the studies.  It is clear that there  are exceptionally




wide variations in the consumer surplus per unit of use  -- from under $1 to




over $100 in five of the seven cases.  Two of these have estimates over $200.




These differences could represent dramatic differences in the character of the

-------
                                                                           5.25
resources in each group, in the models used, or in the characteristics of the



recreationists in each sample.




     Table 9 reports the ordinary least squares (OLS) estimates for five models



which consider whether the variations in real consumer surplus across studies




can be "explained" by the classes of variables hypothesized in equation (11).




Models 1 and 2 in the table contain the least variables, with 1 considering




only qualitative variables describing the type of recreation site and 2




variables describing the model specification.  The remaining three models




introduce groups of variables to illustrate the sensitivity of the estimates to




the model specification, as well as to the reductions in sample size implied by



these more detailed formulations.  These reductions arise from the incomplete




information available in the papers used to construct the Smith-Kaoru data




base.  Model 5 is their preferred specification.




     The numbers in parentheses below the estimated coefficients are the t-




ratios calculated with the OLS standard errors.  Those in brackets below models




3 through 5 are the t ratios using the standard errors ».stim*,«:ed from the




Newey-West [1987] proposed adaptation of the White [1980] consistent covariance




matrix.  They are reported to gauge whether the panel nature of these data



might have influenced any judgments on the importance of variables describing




the sites or Che modeling decisions.



     The Smith-Kaoru data set is a panel because there are a number of cases of




multiple consumer surplus estimates reported from a single study.  These can




reflect different models estimated with data for a. common recreation site,




different sites and associated data, or both.  Given this diversity in the




source of multiple observations per study, the model does not readily conform




to either a simple fixed or a random effects model.  Newey and West's

-------
                                                                           5.26
covariance estimator allows for a generalized form of autocorrelation and




heteroscedasticity.  As such, it provides a convenient gauge of the potential




effects of the stochastic assumptions maintained in estimating the determinants



of the real consumer surplus.




     Several conclusions emerge from this statistical summary of the




literature.  The results clearly support the basic approach to reviewing




empirical literature.  The models' estimates indicate that the type of



resource, the modeling assumptions, specification of the demand function, and




estimator can influence the resulting real consumer surplus estimates.




     For the most part, individual variables had effects consistent with a




priori expectations. Nonetheless, there is at least one important aspect of the




variable definitions that should be recognized.  Our site classification



variable is not a class of mutually exclusive categories.  Some sites fall in




multiple categories.  For example, a state park with a lake would imply unitary




values for both of these variables.  The estimated coefficients must also be




interpreted relative to an omitted category (coastal sites and wetlands),




because all sites fell within at least one of these definitions.  Thus the




differential a state park with a lake would imply in per unit consumer surplus




over coastal areas is about $2.00.  Nearly all the variables describing




modeling decisions were found to be statistically significant factors in




describing the variation in real consumer surplus.



      Examples of these results, that are on the one hand consistent with




intuition yet also disturbing from the perspective of developing benefit




estimates that are readily transferred, include the effects of the treatment




of:  substitute price measures; the value of the opportunity costs of time; the




specifications used to capture the effects of multiple sites  (e.g., the

-------
                                                                           5.27
regional travel cost model); the demand specification (notably the  double-log



form); and estimator used to account for the truncation effects present with



site-intercept surveys.




     Overall these findings emphasize the sources of ambiguity in demand



modeling described earlier.   While the Smith-Kaoru findings are just a start




and should be interpreted cautiously, some specific areas can be targeted




despite this qualification. More careful consideration is warranted of why the




treatment of time costs and the selection of an estimator are so important to




these valuation estimates.  In the first case, the sensitivity reflects the



fact that we do not know how the constraints to an individual's time affect his




recreation decisions or how an individual's implicit values on time vary with




the nature of his choices.  Data can be sought on both issues.




     Similarly, the importance of the choice of estimator probably reflects the




difficult subsidiary issues involved in deciding how to deal with the sampling




(Shaw [1988]) and selection (Smith [1988]) effects associated with intercept




surveys.  An effort to improve the situation through data collection would




involve returning to the early population surveys (i.e. samples designed to be




representative of all households, not Just users) that elicited information on




households' recreation choices.  These surveys originally were sponsored by the




Bureau of Outdoor Recreation (see Cicchetti et «1. [1969]).  However, any new




surveys would require information on the sites individuals use and their




patterns of use to overcome the problems that arise in the on-site surveys. (The




early BOR surveys did not collect this type of information.)  Understanding the




"market" for a recreation site lies at the heart of evaluating why substitute




prices and the qualitative variable for regional travel cost models were




important.

-------
                                                                           5.28
     We know very little of how individuals learn about and subsequently define




(for choice purposes) the recreation opportunities available to them.




Decisions on the use of "local" recreation sites versus more distant "national"




sites will most, certainly be made with different time horizons and constraints.



How are these decisions to be distinguished and can they be modeled separately?




Progress in modeling recration decisions requires answers.




     The empirical models also identify an important role for the functional



form selected to describe demand.  The recreational demand literature has seen




increasing criticism of the use of arbitrary specifications selected largely




for convenience or based on some fitting criteria.  Several recent studies have



argued that behavioral derivations of demand models would be preferrable.  That




is, they suggest models should begin with specific utility functions and derive




estimating equations by assuming optimizing behavior and by specifying the



budget and time constraints assumed to face individuals. Of course, analytical




tractability constrains how these efforts can proceed.




     We believe thai, there is not an obvious answer t? the question of imposing




prior theory versus using approximations. In a genuine sense, all applications




are approximations.  What is important is whether they way they are undertaken




affects the findings in important ways.  The Smith-Kaoru results indicate that




greater efforts are needed in developing more robust specifications.  Both




enhanced data and theory will be required to meet this need.








V.   Recommendations for Data and Analytical Development




     When compared with the effort and experience devoted to  the conventional




topics considered under the auspices of the Conference on Income and Wealth,




the record of empirical analyses of public policies for the management of

-------
                                                                           5.29
environmental resources is quite limited.  While there has been rapid progress




in the last two decades, our ability to deliver estimates of individuals'




values for a wide array of environmental resources and, & fortiori,  for changes



in specific aspects of resource quality lags significantly behind the




expectations of current environmental statutes and the projected needs for




coming to grips with emerging policy issues.  We have tried to describe the




sources of these demands and the clear interaction between the needs for



economic and non- economic information,




     In what follows 'we propose to use three themes to organize our proposals




for new data developments in support of empirical research in environmental



economics:  learning about natural systems, learning what we know, and




responding to emerging policies.  As we noted at the outset, our objective is



to consider first generic problems extending over multiple problems that




require data and second, broad classes of environmental problems that seem




likely to be important policy issues in the near future.  The policy



orient:* c? on is del 'vr addressing data and modeling needs




are scare, and we need to consider their net returns here just as in other




allocation decisions.
A.   Learning About Natural




     As we have stressed at several points, analysis of the benefits  (or




damages) of proposed or actual changes in the use of natural resources




inevitably depends on our abilities to trace the effect of the changed use




through to a change in the valuation by consumers of a resource service.  This




implies that we must be able to:

-------
                                                                           5.30
     -characterize the current state of the relevant system(s);




     -identify a mechanism by which the change in use affects the system;




     -model how. the change has affected (for ex post damage assessment)  or



     will affect (for.ex ante regulatory analysis) the ambient quality of




     the system in terms relevant to consumer valuation.




     In many cases, our knowledge is deficient in every one of these




categories.  For example,  we have a lot of data on water quality but are




generally short of information that systematically covers all the water bodies




that our activities affect and that our regulations are designed to protect or




enhance.  Further, the available information usually covers items relevant to




scientists' search for understanding of aquatic biological or chemical




processes rather than those that can be related to consumer valuation.  Even




so, to a large extent, our abilities to model aquatic processes are inadequate.




The models often do not accept as inputs discharges or give as outputs




indicators of use or of resulting ambient quality relevant to policy evaluation




needs.



     The great need here is for data gathering and model building efforts to




reflect the demands of policy analysis.  Identifying the need is a great deal



easier than meeting it, for the required interaction has all the difficulties




of interdisciplinary research plus those of interstate and interagency




jurisdictional disputes.  Leadership from U.S. EPA and the Council of




Environmental Quality clearly is called for.

-------
                                                                           5.31
B.   Learning What We Know




     Nearly a decade ago, in closing his overview of the state of the art in




benefit estimation, Freeman [1979] observed that economists  could advise the




EPA administrator how to measure benefits from a particular  pollution control




policy.  All that was needed were the data and learning that accompany



implementation.  The intervening decade has seen some positive investments in




both data collection and in empirical modeling.  However, we cannot be overly



sanguine about what has been accomplished.  For the most part, the efforts have




been very specialized -- relying on existing data on consumer behavior or




developing special purpose contingent valuation surveys to estimate how



individuals would value (or respond to) changes in very specific resources.



This process has made it clear that under currently shrinking budgets (or even




with modestly expanding resources), we cannot possibly estimate the values for



all the resources of current interest.




     The notion of evaluating the conditions for transferring estimates from




one resource to another is a relatively new one.  Tt has been an important part




of the practice of developing the information benefit-cost evaluations



involving non-marketed resources.  Freeman [1984] distinguished top-down and




bottom-up transfers, where the former attempts to allocate an aggregate benefit




for a change in *11 of one type of resource (e.g.. the share of the national




benefits from a water quality improvement attributed to  one site), and the




latter refers to using micro-estimates for the household and a specific




resource in other contexts and aggregating.  Naughton, Parsons and Desvousges




[1988] recently have considered the generic issues in performing benefit




transfers at the micro-level using the pulp and paper industry.  Their results




suggest that a tranfer-based strategy for policy analyses is desirable but may

-------
                                                                           5.32
require restructuring the design of future benefit estimation studies for



environmental resources.




     Another possibility proposed by Mitchell and Carson [1986]  involves using



survey methods to obtain estimates for national improvements in an




environmental resource from individual households.  These estimates would then




be attributed to individual areas based on the amount of the resources present




in the area.  The example these authors used involved water quality




improvements, and comparison of their approach with the results from a separate




contingent valuation indicated a fairly close correspondence between the




estimates derived from a specific survey and those from their national survey



adjusted with their proposed proportioning method.  At this stage, however, the




literature is very preliminary.  There has been no attempt to develop how the



tasks invovled in deriving transferrable models are related to the factors



(i.e., household and resource characteristics) affecting the variation in




benefit estimates across resources and user groups.




     First, we a.i'2t learn wr;at we know from experiences to date, and then we




must proceed to identify what we need to learn.  There is a long tradition in




resource economics involving attempts to develop consensus practices in



benefit-cost analysis and even specifying benchmark valuation estimates for




resource services most closely aligned with water resource projects.  These




attempts were traditionally associated with the Water Resources Council.  Our




suggestion here is that we should extend these efforts to the .valuation




estimates for all environmental resources and thereby move beyond a judgment-




based, single value for each type of resource service.



     By treating the existing set of estimates for changes in the quantities or




qualities of environmental resources, it is possible to develop a systematic

-------
                                                                           5.33
appraisal of whether the scate of the art has advanced to the point where we




can associate variations in estimates with differences in the procedures used




or with features of the resources (or consumers) involved.   This process should



identify the areas with greatest uncertainty.




     The experience with the Smith-Kaoru pilot study of travel cost demand




studies suggests that a more systematic approach, contacting authors to fill in



missing details, is essential if a reasonably adequate database is to be




developed in areas in which there has been less research activity.  Such




efforts would also promote the development of statistical methods for dealing



with the unique features of "panel" data sets composed from existing empirical



studies.








C.   Emerging Policy Needs




     We have classified our views of the emerging policy needs into four



categories and consider each in turn.








1.   Environmental Risk




     This is one of the most difficult areas for current uses of economic




analysis, especially because it appears that individuals' responses to a wide




range of enviromental risks do not conform to our conventional




characterization of rational behavior.  A recent EPA publication  (see U.S.




Environmental Protection Agency [1987a]) has highlighted just how dramatically




inconsistent are public concerns and the rankings of environmental risks based




on expert opinion (U.S. EPA [1987b]).  A comprehensive program of data




acquisition and research is needed to determine how and why households value

-------
                                                                           5.34
reductions In these types of risks more highly than other sources that often



have greater likelihoods of serious effects.




     This type of analysis will be important to the design of information



programs associated with pollutants EPA does not currently regulate,  such as




radon, and to the development of labeling standards for products for which they




do have responsibility.  It is also likely to play a central role in defining




"clean" for Superfund sites, in establishing priorities for policy initiatives




involving monitoring the underground storage tanks, and in devising new




policies associated with more stringent drinking water standards.








2.   Air Quality




     Acid deposition is hardly "emerging" as an issue; rather the reverse.  But



that is not because the scientific questions have been answered and the



problems solved.  Indeed, there is still debate in the scientific literature




over the relative contribution of different compounds and source locations to



observed low pH f»reeipitati«p  fog. and dry e.r?dic deposition.  Under these



circumstances, benefit estimation linked to a discharge-reduction policy cannot




proceed to meaningful results.  So a clear need is for further research into




long-run atmospheric transport and chemical transformation processes, with the




ultimate aim of allowing predictions of the form:  If we reduce sulfur dioxide




(SO.) discharges in this region by this much, average pH of precipitation in




this other region will increase by this much.




     Even then we shall still be several steps from successful benefit




estimation for a policy of SO. reduction.  It must be possible to extend




predictive natural system models to such issues as the link between average




annual (or season-specific) precipitation, pH, and soil quality to vegetation

-------
                                                                           5.35
health and growth, and to aquatic ecological system functions.   For example,  if



we reduce SC>2 discharge in the Middle West,  will New England and New York lakes




and ponds have better fish populations (more and larger fish of more highly



valued species)?




     Only with those tools in hand will it be possible for economists to




produce meaningful benefit estimates for the sorts of policies  that are




regularly debated in the Congress.  To prepare for that day, the problems of




benefit (or damage) function transfer must be addressed in this problem




setting.  In particular, it is necessary to consider how best to use the



results of national studies on the one hand (e.g., Vaughan and Russell [1982]




and local studies on the other (e.g., Smith and Desvousges  [1986]) to value



regional effects.




     A second air quality issue with even larger potential economic




implications is ground-level ozone and particularly the value of trying to



attain the currently mandated National Ambient Air Quality Standards for that




secondary pollutant.    Here it is necessary to iuurove our knowledge of:




     -the sources and actual levels of the precursor pollutants (especially




     volatile organic compounds (VOCS)), of ground level ozone in urban and




     rural areas;




     -the morbidity effects of different levels of ozone;



     -the effects of-ozone on vegetation and a variety of materials such as




     paints, plastics, and synthetic rubbers.



Our estimates of the damages attributable to days of sickness of various types




and severities must be refined.  Moreover,  theoretically consistent but




practically implementable ways of measuring the value of damage to materials




providing services to households, businesses and governments must be developed.

-------
                                                                           5.36
3.   Water Quality




     One of the key policy initiatives in water quality will be associated with



the national estuarine program.  For point sources of waterbome pollution, the



first round of effluent guidelines will be in place with over thirty




regulations promulgated.  All should be in place by the early 1990s.  The




future here is best characterized as one requiring extensions in the ability of




economic valuation to realize greater degrees of resolution in valuing small



changes in pollutants.




     Present methods and data would not permit such evaluations.  Clearly an




improved understanding of the linkage between the technical dimensions of water




quality and individuals' perceptions of and corresponding valuations for that




quality will be necessary.




     Non-point sources, especially agricultural runoff of pesticides and




fertilizers to surface waters, represent the largest unregulated source of




water pollutants.  Presently, EPA  lo«ss not l»*ve. •withoi.J.ty  :o regulate these




sources.  However, recent opportunities to coordinate the selection of areas




for the Department of Agriculture's Conservation Reserve Program, based on the



effects of pollutants on water resources, expose a new area for economic




valuation.  Can we we set priorities for the selection of lands for inclusion




in this system based on their contributions to non-point source pollution?  To




answer this question we need both economic and non-economic data.  Agriculture




has been willing to pay premia over normal reserve payments for withholding




lands that might otherwise contribute to impairing significant environmental




resources.

-------
                                                                           5.37
4.   Stock Pollutants and Global Climate Change




     This last area is fundamentally different than the first three emerging




issues we discussed in that the policy time horizon is long-term and extends




over several decades.   While not a new issue (Revelle [1985] has suggested it




was identified over 100 years ago), it has achieved a more prominent role on



the policy agenda with the Global Climate Protection Act of 1987.  This




legislation assigns EPA the responsibility of summarizing the scientific



understanding of the greenhouse effect (i.e. the role of the accumulation of



carbon dioxide, chlorofluorocarbons, methane, and other trace gases in the




upper atmosphere in increasing average surface temperatures on earth) and in




enumerating the policies available for stabilizing these concentrations.



     As  in our other examples, a key need in this area is for greater




understanding of the natural system.  In this case it is the link between these




atmospheric gases and the extent and timing of any global warming, as well as




of the implications of that global warming for regional weather patterns.  This




issue raises SOBS distinct methodological needs because of the extent of




scientific uncertainty over these questions, the tine horizon for the potential




climatic changes, and the irreversibility of the process.



     The requirements for economic information depend, in part, on the progress




made in  improving our understanding of the natural system.  As this proceeds,




there is a clear need to understand the processes by which economic activities




adapt and the institutions that facilitate such adjustment.   Historical and




cross-cultural analyses may well offer the only means for developing such




insights.  Equally important, there is a fundamental need to describe the




inherent uncertainties in a way that is genuinely informative for policy.




While not unique to this problem, this issue of communicating the inherent

-------
                                                                           5.38
uncertainties remains one of the most significant problems facing economists




involved in environmental policy.




     Finally, in evaluating these data and modeling needs as compared with




other data priorities, it is important to recognize that in contrast to




positive uses of economic analysis where a lack of data may prevent decisions




from being made, this is not the case in normative applications.  Decisions are




made regardless of whether the economic information is available.   In some




cases they are very bad ones.   Consequently, here new data developments




represent opportunities to improve the quality of decisions and the resource




allocations affected by them.

-------
                                                                           5.39
                                                       Russell and Smith



                                   CHAPTER 5




                                   FOOTNOTES




1.  To say that the analysis is difficult (and expensive)  is not to say that it




    is of dubious value.  The U.S.  Environmental Protection Agency's (EPA)




    review (1987a) of its use of benefit-cost analysis concludes that for three




    regulatory decisions, B-C analysis identified improvements with potential




    benefits of over $10 billion (lead in automotive fuels, $6.7  bil.; Used




    lubricating oil, $3.6 bil.; and pre-manufacturing review of toxic



    substances, $.04 bil.).   Further,  EPA estimates the costs of all regulatory



    impact analyses (RLAs) done under  the terms of President Reagan's Executive



    Order 12291 as less than $10 million.  Therefore, the return to analytical




    investment appears to be over 1000 to 1 in the aggregate.




          Several cautions are in order in interpreting this conclusion.  Most



    fundamentally, our argument in this paper, if one accepts it, must



    inevitably throw some doubt on these  benefit estimates.  Second, we canno-




    necessarily project such a return  ratio in the future because it is likely




    that the biggest and easiest targets  have already been attacked.  And




    finally, we should include a grain of salt because the self-interest of




    those preparing the report was consistent with finding large returns.



2.  This statutory requirement has not prevented benefit cost information from




    being included in the Regulatory Impact Analyses prepared for cases




    involving the primary standards.  The proposed standard subjected to




    analysis is health based.  It is too  early to know whether the final




    standard that emerges after OHB review can be argued to have been affected




    by the benefit-cost findings.

-------
                                                                           5.40
3.  Location is, of course, three dimensional,  and altitude can make a big




    difference in some situations; but the points we make are only reinforced




    by considering a third dimension,  while exposition is much simpler for two.



4.  This last point is stressed by McGartland et al. [1988].   We shall return



    to it below.




5.  Our discussion assumes that the regulations in question will,  in fact, be




    complied with. Making sure this is even roughly the case requires



    investment in monitoring and enforcement.  These costs should be counted as



    costs of the policy, and their amount and how they are used will help




    determine the realized level of benefits. It is also true that choices open




    in the design of implementation systems can affect monitoring and




    enforcement costs and thus also indirectly affect benefits by that route.



    We ignore these added complications, though they open up an entirely new



    and largely unexplored source of demand for data and analysis.



6.  Reasonably straightforward theoretical expositions are available that




    Include differential location.  For example, see F^rs-oid  [19/2] and




    Tietenberg [1978] and Siebert  [1985].



7.  The matrix T may be thought of as representing  the steady-state solution to




    a set of differential equations that reflects the transportation of




    pollution by average winds characterized by velocities and directions, and




    the diffusion of the pollution particles due to random motion in the plume.




    If the units of discharge are, say  (average) tons per day, the units of the




    elements of T could be (average) micrograms per cubic meter.




8.  For simplicity, it is assumed that  the same damage relation applies at each




    receptor location, though as just stressed, we  would expect the damages, for

-------
                                                                            5.41
     a given pollutant concentration to differ across the various points in the




     regional space.




 9.  Here we calculate {Q} on the basis of DQ and T, but for the argument that




     follows, it. is important to note that baseline ambient quality is actually




     realized and therefore can be measured.  Thus, there is no inconsistency in




     assuming knowledge of (Q) and ignorance of T.  As a practical matter,




     however, we may very well be ignorant of (Q) in any but loosest, one might



     say anecdotal, sense.  See, for example, Russell, et al. [1983],  To be




     useful, our knowledge of ambient quality conditions must be reflected in




     measurements that are:




     -  meaningful in terms of their links with or effects on human valuation of



      environmental services, and



        connectable to pollution discharges that will have to be changed to change




        ambient quality.




     We return to this matter of baseline quality in the final section.




10.  The actual pr;. terns of- orabient qunlit;.' produced by the roll-back



     implementation method under the baseline and the alternative standards are:

I
II
III
Base
10
8
6
S8
8.0
6.4*
4.8*
S4
6.0
4.8
3.6a
S
u
4.0
3.2
1.6
S2
2.0
1.6
1.2
(a) indicates a quality level not reflected at all  in benefit calculation (2).

-------
                                                                            5.42
11.  It should be emphasized that there Is no reason to expect a mathematically



     desirable -- or even smooth -- pattern for marginal benefits.   The complex




     relation among standard, discharge reduction amounts and location required




     under a. given implementation method, and resulting pattern of ambient quality




     changes, can produce virtually any pattern of marginal benefits.




12.  For a description of efforts to use natural world models in water quality




     benefit estimation, although some of the threshold aspects of the B(2) method



     are still used, sea Vaughan and Russell [1982],




13.  This section is based on research undertaken by Yoshiaki Kaoru and the second




     author and is reported in more detail in Smith and Kaoru [1988].




14.  See Mitchell and Carson [forthcoming] for an overview of the issues involved



     in using these methods.




15.  It is not completely new to economics.  Berndt's [1976] early attempt to



     reconcile the diverse estimates of elasticities of substitution between




     capital and labor is similar to our objectives.  However, in his case, the



     focus *-«-•? on thfc assumption inherent in the estimation models and Iht-.ir




     likely implications for the estimates.  Somewhat more closely aligned is the




     Hazilla-Kopp [1986] summary of their findings on the sensitivity of the



     characterization of substitution possibilities across different modeling




     decisions made with the 36 different manufacturing sectors they analyzed.  In




     this case, the analysis parallels what we propose, but their objective was to




     summarize their own findings, rather than detect sources of differences




     across studies conducted by different individuals.




16.  Thanks are due to Tom Tietenberg for suggesting that we make this point more




     explicit based on his review of an earlier draft of this paper.

-------
                                                                            5.43
17.   Ozone is "secondary" because it is formed in the atmosphere from chemical




     reactions involving sunlight and certain "primary" or discharged pollutants,



     especially volatile organic compounds such as gasoline and solvents and




     oxides of nitrogen.

-------
                                                                         5.44
                                                       Russell and Smith
      Figure 1.  SCHEMATIC DESCRIPTION OF ISSUES IN USING ECONOMIC ANALYSIS
                           FOR ENVIRONMENTAL POLICIES
   Logical Structure for Analysis
 Definition of technology, discharge,
 or environmental quality standard by
     type of pollutant and media

  Development of implementation plan
          to meet standard

     Change in effluent loadings
     POLLUTION DISCHARGES TO THE
             ENVIRONMENT
Changes in one or more dimensions of
 resources depend on the spatial and
       environmental resources

  Change in ecological habitat and
          nonhuman species

       Change in human species
   AMBIENT ENVIRONMENTAL QUALITY
    CONDITIONS (SERVICE LEVELS)
     Economic agents change their
  patterns of consumption of other
    related goods or their use or
valuation of environmental resources
  Rationale for the Logical Structure

The institutional structure governing
the definition and implementation of
environmental policy is complex.   As a
result of multiple, overlapping statutes
defined by both environmental media
(e.g.:  Clean Air and Clean Water Acts)
and the types of residuals generated
(e.g.:  Resource Conservation and
Recovery Act, Comprehensive
Environmental Response, Compensation and
Liability Act, etc.), policies must be
responsive to multiple objectives.
Moreover, they can involve the
definition of standards in a format
inconsistent with available
environmental data or in generic terms
that require considerable judgment to
implement, enforce, or evaluate.

The services of environmental resources
are produced within a complex physical
system.  The effects of different
patterns and types of uses of these
temporal aspects of those uses.  In
particular, the pattern of environmental
quality corresponding to a chosen
standard depends on the implementation
program to be used to attain the
standard.

The services of environmental resources
exchange outside markets, and therefore,
the information normally present from
market exchanges is not available.
Indeed, as part of their ordinary
consumption choices individuals may not
have been required to consider changes
comparable to those envisioned in any
specific policy analysis.  Information
on the quality and character of these
services can be quite technical, involve
subtle measurement problems, and is
unlikely to be generated through the
informal processes individuals use to
learn about other commodities they
consume.

-------
                                                                          5.45
                                                            Russell  and  Smith
TABLE 1.          COMPONENTS OF ECONOMIC ANALYSIS IDENTIFIED
                        IN EFA'S ENABLING LEGISLATION*
Economic
Legislation/Regulation   	Benefits	Costs   	Impacts

                         Human   Other           .                Cost
                         Health  Effects  Welfare   Compliance  Effective

Clean Air Act

Primary NAAQS             x        x
Secondary NAAQS                             x
Hazardous Air Pollutants  x
New Source Performance    *•        x        *          **          **       **
Motor Vehicle Emissions   x        x        x          x           x        x
Fuel Standards            x        x        x          x
Aircraft Emissions        x        x        x          x                    x

Clean Water Act

Private Treatment                  f       ***         x           x        x
Public Treatment                   f

Safe Drinking Water Act

Max. Contaminant Level
  Goals                   x
Max. Contaminant Levels                                x                    x

Toxic Substances Act      x        x        x       .   x           x        x

Resource Conservation
  and Recovery Act        x                 x

CERCLA (SARAi

Reportable Quantities     x                 x
National Contingency               x                               x

Federal Insecticide. Fungi-
cide and Rodenticide Act

Data Requirements                           x
Minor Uses                x        x        x          x           x        x

-------
                                                                           5.46
Table 1. Continued

Economic
Legislation/Regulation
Atomic Energy Act
Radioactive Waste
Uranium Mill Tailings
      Benefits
                       Costs
                            Impacts
                         Human   Other
                                        Cost
                         Health  Effects  Welfare   Compliance  Effective
x
x
X
X
X
x
a.   Source:  Adapted from EPA's Use of Benefit Cost Analysis: 1981-1986. Table

      3-1, p. 3-2.

b.   NAAQS designates the National Ambient Air Quality Standards and relates to

      the criteria air pollutants.

c.   "Other effects" refer to non-health effects on humans and firms.

d.   "Welfare effects* refer to visibility and aesthetics, effects on nonhuman

         species,  crops, sodding, materials damage.

e.   The typt .f  analy.^s here ilr^rads 
-------
                                                                           5.47
TABLE 2.           AVERAGE EMISSION REDUCTIONS OF VOLATILE
                  ORGANIC COMPOUNDS PREDICTED TO BE REQUIRED
                 TO MEET OZONE NAAQS IN SELECTED OHIO CITIES
..City '
Cleveland
Akron
Toledo
Columbus
Canton
Youngs town
Dayton
Cincinnati
Technique 1
87%
35%
47%
43%
22%
64%
61%
40%
Technique 2
50%
18%
25%
25%
10%
44%
40%
50%
Technique
Selected
1
2
2
2
2
2
2
1
a.  Known as "EKMA."
b.  Known as "Rollback.1
     Source:  Adapted from Pacific Environmental Services, Study of the 1979

     State Implementation Plan Submittals  (Elmhurst, IL:  Report prepared for

     U.S. National Commissioner on Air Quality, December 1980), pp. 7-12. and

     published  in Richard Liroff, 1986 Reforming Air Pollution Regulation

     (Washington, DC:  The Conservation Foundation).

-------
                                                                      5.48
             TABLE 3.  AGGREGATE AND MARGINAL BENEFITS:
                THE TWO SOURCE-THREE RECEPTOR EXAMPLE
                               Aggregate Total    Aggregate Marginal
                               Benefits by Std.     Benefits by Std.

                              8     6428642
B1
B2
B3
Average initial 'regional
concept of quality
relative to standard
Actual initial quality
relative to standard
Actual initial quality
0
36
72
84
92
128
144
152
168
180
188
192
0
18
36
42
28
28
30
30
20
18
18
12
relative to actual
quality as determined
for roll-back implemen-
tation

-------
                                                                           5.49
     TABLE  4.   SURROGATE BENEFITS OF REDUCTIONS IN TOTAL SUSPENDED PARTICULATES
         FOR BALTIMORE BY LEVEL OF IGNORANCE AND STANDARD (MILLION PER YEAR)
Level of

Method 1
Total
Method 1
(modified) a
Total
Marginal
Method 2
Total
Marginal
Method 3
Total
Marginal
McGartland, et al.
Marginal Benefits
(millions «f 1980
dollars)
m

0


12.3
12.3

2.6
2.6
.
7.7
7.7

7.2


no

0


23.6
11.3

6.0
3.4

19.7
12.0

12.9


105

0


34.5
10.9

9.7
3.7

27.7
8.0

9.1


Standard
100

0


45.2
10.7

15.4
5.7

34.9
7.2

8.5


(ug/m J)
21

0


55.1
9.9

21.2
5.8

46.2
11.3

13.2


1
90

0


64.5
9.4

28.2
7.0

59.1
12.9

15.1


85

0


73.6
9.1

35.2
7.0

73.7
14.6

16.4


Source:  See Appendix A for a description of the data and method.



a  The modification consists of comparing initial average concentration to projected


average concentrations for each standard, where the projection depended on the


percentage change in the standard.

-------
                                                                          5.50
TABLE 5.  SURROGATE BENEFITS OF IMPROVEMENTS IN WATER QUALITY IN THE
                  DELAWARE ESTUARY BY LEVEL OF IGNORANCE OF STANDARD
                                       Water Quality Standard
              " „                     (ppo of dissolved oxygen)

                                       3.5                 5.0

Method 1
     Total                              0                  420.2
     Marginal                           0                  420.2

Method 2
     local                            184.5                510.6
     Marginal                         184.5                326.1

Method 3
     Total                            372.7                581.4
     Marginal                         372.7                208.7

-------
                                                                   5.51
  TABLE 6.  MARGINAL BENEFITS OF REDUCTIONS IN TOTAL SUSPENDED
PARTICULATES FOR BALTIMORE BY IMPLEMENTATION METHOD AND STANDARD
                   (MILLIONS OF 1980 DOLLARS)
                                             3
                    Level of Standard  (ug/m  )
Implementation
Method
Command and Control
Least Cost
115 110 105 100 95 90 85
2.2 10.5 9.7 11.5 7.5 10.0 6.5
7.2 12.9 9.0 8.5 13.2 15.1 16.4
           Source:  McGartland et al.  [1988],  Table 1.

-------
                                                                           5.52
                   TABLE 7:  DESCRIPTION OF VARIABLES FOR ANALYSIS
Name
          Definition  of Variables
RCS
Surtype
Recreation Site
Variable
Substitute Price
Opportunity Cost
type #1
Opportunity Cost
type #2
Harshallian consumer surplus estimated per unit of use ,  as
measured by each study (i.e., per day or per trip) deflated by
consumer price index (base - 1967)

Qualitative variable for measure of site use - 1 for per trip
measure, 0 for per day measure

Lake, River, Coastal area of Vetlands, Forest or Mountain
area, Developed or state park, National park with or without
wilderness significance are the designations, Variables are unity
if satisfying designation, 0 otherwise

Qualitative Variable - 1 if substitute price term was included in
the demand specification, 0 otherwise

Qualitative Variable for Measure used to estimate
opportunity cost of travel time - 1 if an average wage rate was
used

Qualitative Variable for the second type of opportunity
costs of travel time measure, - 1 income per hour used (omitted
category was predicted individual specific wage)
Fraction of wage     fraction of wage rate used  to estimate opportunity cost of travel
Specific  Site
Demand
Specifications
Estimators  Used
Qualitative Variable for use of a state or regional Travel Cost
model describing demand for a set of sites - 1, 0 otherwise

linear, log-linear and semi-log (dep) are qualitative
variables describing the specification of functional form for
demand  (semi-log in logs of independent variables was the omitted
category) .

OLS, GLS, and ML-TRUNC are qualitative variables for estimators
used, omitted categories correspond to estimators with limited
representation in studies including the simultaneous equation
estimators.

-------
                                                                           5.53
                TABLE 8:  A COMPARISON OF TRAVEL COST DEMAND RESULTS
                                 BY TYPE OF RESOURCE
Type
of
Resource
RIVER
LAKE
FORESTS
NATIONAL
PARKS
WETLANDS
STATE
PARKS
COASTAL
AREAS
Real
Consumer Surplus
Number of
Estimates Mean
257 $17.05
483 $16.85
114 $31.36
12 $44.01

9 $45.86
107 $42.49

28 $35.49

Range
$.29
$.09
$.80
$23.48

$17.45
$.67

$.67

- $120.70
- $219.80
- $129.90
- $120.70

- $120.70
- $327.20

• $160.80

PIb YEARS0
.61 1966 - 1983
.55 1968 - 1983
.59 1968 - 1984
.50 1980 - 1983

.78 1980 - 1983
.07 1972 - 1983

.61 1972 - 1984

a.  Real consumer surplus deflates the nominal estimates by the consumer price index

    (base 1967)

b.  This variable designates the proportion of the studies based on samples of

    individual recreationists' trip-taking decisions compared with origin zone

    aggregate rates of use.

c.  The range of years in which the data used in these studies were collected.  Thus,

    this variable designates the range of years across the studies in each category in

    which behavior was observed.

Source:  Smith and Kaoru [1988]

-------
                                                                           5.54
TABLE 9:  THE DETERMINANTS OF REAL CONSUMER SURPLUS PER UNIT OF USE
Independent
 Variables
                                              Models
Intercept


23.72
(5.62)

16.07
(2.08)

20.30
(6.19)
[3.92]
27.03
(3.68)
[3.64]
18.75
(0.58)
[1.04]
Surtype
Tvoe of Site (Xg)

Lake
River
Forest
State Park
National Park
 7.99     -4.13
(2.76)    (-1.45)
-11.70
(-3.18)
 -5.57
(-1.93)
  -.45
(-0.93)
 19.93
 (4.44)
  2.54
 (0.20)
Model  AfrffVttptiftll  (X
Substitute  Price
Opportunity Cost  of
Type  #1
 Opportunity Cost of
 Type  #2
 -9.97    15.38      19.88
(-2.72)   (2.97)     (3.74)
[-1.36]   [2.34]     [3.55]
         -18.69     -20.32
         (-3.24)    (-3.52)
         [-2.36]    [-2.48]

         -14,29     -19.03
         (-2.99)    (-2.19)
         [-1.95]    [-1.75]

         -18.45     -25.99
         (-2.36)    (-3.01)
         [-1.93]    [-2.49]

          24.95      22.37
          (3.47)     (3.44)
          [3.27]     [3.19]

            .56      -3.77
          (0.04)    (-0.23)
          [0.08]    [-0.13]
                    -18.73              -13.71
                    (-3.27)             (-2.12)
                    [-4.58]             [-1.80]

                    -14.97              -16.49
                    (-2.10)             (-2.11)
                    [-2.09]             [-2.48]

                      3.95              -15.86
                     (1.02)             (-3.30)
                     [0.45]             [-2.87]

-------
                                                                           5.55
TABLE 9 (continued)
Independent
Variables 1 2
Fraction of Wage
Specific Site/Regional
TC Model
Model Specification (JO
Linear 2.35
(0.31)
Log-Linear 14.63
(1.89)
Semi-Log (Dep) 11.26
(1.52)
Estimator (X_)
OLS
GLS
ML-Trunc
R2 .11 .03
n 722 722
Models
345
37.24 48.59
(8.56) (9.76)
[3.83] [6.94]
22.23 24.21
(4.10) (3.85)
[3.35] [2.77]

-2.87
(-0.27)
[-0.31]
23.37
(2.37)
[2.88]
16.89
(1.86)
[2.97]

-14.45
(-0.48)
[-0.84]
-8.58
(-0.28)
[-0.54]
-67.38
(-2.15)
[-3.43]
.25 .15 .42
399 399 399
 a.    The  numbers  in parentheses  below the estimated parameters are the  ratios




      of the coefficients  to their estimated standard errors.   The numbers in

-------
                                                                      5.56
brackets are the Newey-West [1987] variant of the White  [1980] consistent




covariance estimates for the standard errors in calculating these rat-ios.




Source:  Smith and Kaoru [1988]

-------
                                                                           5.57
                                                            Russell and Smith





                                  APPENDIX A



                  CALCULATING SURROGATE BENEFITS BASED ON THE

         BALTIMORE AND DELAWARE RIVER ENVIRONMENTAL QUALITY PROJECTIONS





Air Quality Surrogate Benefits





     McGartland et al. [1988] reproduce their atmospheric model's projected



patterns of total suspended particulate concentrations for 23 receptor



locations in Baltimore for two alternative implementation approaches.  We used



and reproduce their Table 2 here as Table 1-A.  (Ve ignore their results for 83



micrograms/m , ug/m .)  We follow them in taking the pattern associated with



the 120 ug


  3
/m  standard as our base situation.



     While McGartland et al. describe the basis for their damage, and hence



benefit estimates, they did not provide the functions they used.  However, it



turns out th.'t -.  surrogate function that reproduces the pattern of their



marginal benefits is  easy to find.  We used a simple quadratic damage



surrogate.  That  is:
                                                    2   3
     (A-l)     G. - Damage at receptor 1 -  [TSP ppa] xlO



                           (in millions)
     Benefits of increasingly strict standards are then simply:
     (A-2)     Bi - Gi (120) - Ct(J)



               for  j <120 ug/m

-------
                                                                           5.58
     We reproduce here, as Table 2-A,  a sample calculation of the damages and




benefits for six receptor locations, one standard, and three methods.




Inspection of Table 1-A reveals immediately that Method 1 yields an estimate of




zero benefits for all standards, since the initial average quality is already




better than the strictest standard to be examined.

-------
                                                                           5.59
                     TABLE 1-A.  TSP CONCENTRATIONS BY RECEPTOR

                                  LEAST-COST CASE
Receptor
Location
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
-120-
67.8
64.6
56.2
85.4
94.3
107.2
116.3
93.3
119.7
52.4
80.2
102.8
61.6
53.3
120.0
56.4
72.4
84.9
51.6
'7 -
64.0
64.6
105.3
-115-
67.4
63.7
56.0
83.9
92.5
102.6
113. 8b
88.7
115. 3b
51.6
78.4
101.1
60.8
52.8
114. 9b
56.4
69.9
84.0
51.4
f.5.3
63.6
64.3
102.8
-110-
66.2
62.2
55.5
81.2
89.0
99.7
107 . 8b
86.1
110. 4b
49.1
77.4
91.9
58.9
51.8
110. 4b
55.3
66.5
74.9
50.8
54.4
61.2
62.0
98.9
-105-
66.0
61.8
55.5
78.7
86.2
97. 9b
104. 3b
84.4
105. 5b
47.5
72.0
88.6
57.5
51.2
101. Ob
55.1
65.1
74.2
50.5
6.5,3
60.8
61.8
97. 7b
-100-
65.3
60.9
55.3
76.8
83.8
95. Ob
100. Ob
81.6
100. Ob
46.0
70.1
84. 3b
56.0
50.6
99. 6b
54.3
63.5
73.0
50.1
62.1
60.0
59.7
95. Ib
-95-
63.7
58.7
54.6
73.7
80.5
90. 7b
95. 5b
75.6
95. 2b
43.4
68.8
79. 7b
53.9
49.4
93. Ob
52.9
59.4
66.4
49.3
60.0
57.1
56.5
90. 4b
-90-
61.6
55.5
53.7
70.9
76. 9b
85. 7b
90. Ob
69. 9b
89. 5b
40.9
65.7
74. 5b
51.4
48.1
79. 5b
52.2
53.1
62.5
48.3
57.5
55.0
55.4
83. 8b
-85-
59.3
51.7
52.5
68. Ib
73. 5b
80. 8b
85. Ob
63. 5b
84. 7b
38.2
63.5
69. 2b
49.2
46.4
53. 3b
50.9
43.3
55.9
47.3
54.4
52.0
53.1
74. Ib
             UNWEIGHTED AVERAGES OF RECEPTOR TSP LEVELS

             80.1     78.3     75.3     73.3     71.4    68.2    64.4    59.6

            POPULATION-WEIGHTED AVERAGES OF RECEPTOR TSP LEVELS

             77.4     75.7     72.9     70.9     69.0    66.2    62.9    59.3
b    Indicates a concentration reflected in the calculation of benefits using Method
     2.

                           Source:  McGartland,  et al.  1988

-------
                                                                                5.60
TABLE 2-A,  EXAMPLES OF SURROGATE DAMAGE AND BENEFIT CALCULATIONS BY METHOD
Damages at
Receptor Base Level •"
2 4.2
7 13.5
10 2.7
12 10.6
15 U.4
18 7.2
For average level
6.4

Total
Marginal
Method 1 (modified) Method 2 Method 3
Damages Damages Damages
at 110 Benefits at 110 Benefits at 110 Benefits
4.2 0 3.9
12.1 1.4 11.6
2.7 0 2.4
10.6 0 8.4
12.1 2.3 12.1
7.2 0 5.6

5.4 1.03
x 23
23.6 6.0
11.4 3.4
0.3
1.9
0.3
2.2
2.3
1.6



19.7
12.0
   For Modified Method 1,  base  average  surrogate damages - damages at t;;
-------
                                                                           5.61
Water Quality Benefits






     Water quality benefits are based on predicted water quality improvements in




the Delaware estuary published in Spofford et al. [1976],  The quality indicator




used is dissolved oxygen (DO) and the base levels are interpolated from their




Figure 2 reproduced here as Figure 1-A.  Improvements associated with alternative



standards are taken from Table C-3 in the source.  Their run using a 3.0 ppm



standard is used here as a surrogate for a 3.5 ppm standard because in all but




one reach better than 3.5 ppm is attained under it.  The predicted levels of DO




for that standard and for a fun with a 5.0 ppm standard are set out in Table 3-A.




The implementation plan implicit in these runs is the least cost arrangement of



discharge reductions.




     To calculate benefits, dissolved oxygen is translated into sustainable



recreation activities using the table of equivalents developed by Vaughan [1981]




and displayed here schematically as Figure 2-A.  Then the three alternative




methods of benefit C3l.-cu3 it*.c*? '•••r* applied as summarized Irr Table ''-A, where the



per capita per day values of the alternative sustainable activity measures of




quality are drawn from (Smith and Desvousges [1986]).



     What we have not done is to associate numbers of people with particular




receptor locations along the river.  ("Receptor location" is usually called




"reach" in the water pollution field.  It means a stretchof river within which




ambient quality is assumed the same.)  This is difficult to do in any case




without a study to measure the recreational suitability as determined by non-




water quality characteristics.  But it is even more difficult to dp within a




massive urbanized agglomeration such as that from Wilmington, Delaware, to




Trenton, New Jersey, that surrounds the Delaware Estuary.  The figures in Table  5

-------
                                                                           5.62
are therefore simply the sums of the relevant per capita benefits over all the




reaches.  These figures exaggerate the penalty for ignorance of the environment




to the extent that more individuals could easily travel to and recreate on the




middle reaches.  These are the most heavily polluted, and therefore benefits




associated with their cleanup show up in Methods 1 and 2, while any benefits




associated with further cleanup of the most upstream and most downstream reaches




tend to be ignored in those methods.




     Note that the use of Vaughan's equivalence in essence begs an important




question:  Do we have an environmental quality indicator that is connectable both




to discharges and to valued human uses of the environment?  Dissolved oxygen is




only one of the elements of & vector of water quality characteristics that




determine how a body of water can be used.  It may be the key element for fish




populations but is certainly much less important  in  determining whether water is




"beatable" (that is to say, pleasant to boat on)  or  swimmable (where bacterial




counts are turbidity are much more important).

-------
                                                                           5.63
         TABLE 3-A.  BASE CASE AND PREDICTED LEVELS OF DISSOLVED OXYGEN:
         TWO ALTERNATIVE STANDARDS APPLIED TO THE DELAWARE ESTUARY (PPM)
Reach
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Average
Base
Situation
8.3
7.0
5.6
4.9
4.6
4.4
3.8
2.7
1.8
1.3
1.2
1.2
1.3
1.5
1.8
2.3
2.8
3.5
4.2
5.0
5.8
6.6
3.7 Standard
3 . 5 ppm
Standard
8.6
7.7
6.6
6.0
5.7
5.9
5.9
5.8
6.1
5.3
3.6
3.7
3.6
4.0
4.5
5.2
3.0a
3.7
4.8
5.8
6.2
6.6
3.5 .Standard
5.0 ppm
Standard
8.6
7.7
6.9
6.3
5.9
6.0
5.9
5.9
6.4
6.8
5.3
6.1
5.7
5.7
6.1
6.4
5.0
5.1
5.7
6.1
6.2
6.6
5.0
Source:  V. 0. Spofford, Jr., C. S. Russell, and R. A. Kelley, 1976,

Environmental Quality Management:  An Application to the Lover Delaware Valley

(Washington, DC:  Resources for the Future).



     The standard actually imposed by Vaughan et al. was 3.0 ppm.  But 3.5 is a

     lower bound for boatable quality qwater in the Vaughan scale, so we treat

     this run as though the standard were 3.5 for purposes of Method 2

     calculations.

-------
                                                                          5.63
         TABLE 3-A.  BASE CASE AND  PREDICTED  LEVELS  OF DISSOLVED OXYGEN:
         TWO ALTERNATIVE STANDARDS  APPLIED  TO THE  DELAWARE ESTUARY  (PPM)
          Reach
  Base
 Situation
3.5 ppm
 Standard
5.0 ppm
 S tandard
          1
          2
          3
          4
          5

          6
          7
          8
          9
         10

         11
         12
         13
         14
         15

         16
         17
         18
         19
         20

         2i
         22

      Average
8.3               8.6
7.0               7.7
5.6               6.6
4.9               6.0
4.6               5.7

4.4               5.9
3.8               5.9
2.7               5.8
1.8               6.1
1.3               5.3

1.2               3.6
1.2               3.7
1.3               3.6
1.5               4.0
1.8               4.5

2.3               5.2
2.8               3.0*
3.5               3.7
4.2               4.8
5.0               5.8

5.8               6.2
6.6               6.6

3.7      Standard 3.5
                 8.6
                 7.7
                 6.9
                 6.3
                 5.9

                 6.0
                 5.9
                 5.9
                 6.4
                 6.8

                 5.3
                 6.1
                 5.7
                 5.7
                 6.1

                 6.4
                 5.0
                 5.1
                 5.7
                 6.1

                 6.2
                 6.6

        .Standard 5.0
Source:  W. 0. Spofford, Jr., C.  S.  Russell, and R. A. Kelley,  1976,

Environmental Quality Management:  An Application to the Lower Delaware Vallev

(Washington, DC:  Resources for the Future).
     The standard actually imposed by Vaughan et al. was 3.0 ppm.   But 3.5 is a

     lower bound for boatable quality qwater in the Vaughan scale,  so we treat

     this run as though the standard were 3.5 for purposes of Method 2

     calculations.

-------
                                                                                5.64
              TABLE 4-A.  CALCULATING SURROGATE BENEFITS FOR DISSOLVED OXYGEN
                       IMPROVEMENTS  IN THE DELAWARE  ESTUARY  BY METHOD
Method 1
              Base Case Average:  3.7 ppm (B)
              3.5 ppm Standard    3.5 ppm (B)  Benefit - 0 x 22 - 0
              5.0 ppm Standard    5.0 ppm (G)  Benefit - $19.1 x 22 - 420.6
Methods 2
and 3
Marginal Benefits
Method 2
Base Case
Reach Sustainable Use 3.5 Standard
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Totals
S
S
G
B
B
B
B
U
U
U
U
TJ
U
U
U
U
U
B
B
G
G
S

-
.
.
-
.
-
B (20.5)
B (20.5)
B (20.5)
B (20.5)
B (20.5)
B (20.5)
B (20.5)
B (20.5)
B (20.5)
U
.
-
-
.
-
184.5
5.0 Standard
-
-
G (19.1)
G (19.1)
G (19.1)
G (19.1)
G (19.1)
G (19.1)
G (19.1)
G (19.1)
G (19.1)
G (19.1)
G (19.1)
G (19.1)
G (19.1)
G (39.6)
G (19.1)
G (19.1)
-
-
-
326.1
Marginal Benefits
Method 3
3 . 5 Standard
-
S (35.4)
G (19.1)
G (19.1)
G (19.1)
G (19.1)
G (39.6)
G (39.6)
G (39.6)
B (20.5)
B (20.5)
B (20.5)
B (20.5)
B (20.5)
G (39.6)
U
-
-
•
-
•
372.7
5.0 Standard
-
-
.
-
.
-
'
-
S (35.4)
G (19.1)
" f!9.1>
u (ly.r,
G (19.1)
G (19.1)
.
G (39.6)
G (19.1)
G (19.1)
*
-
•
208.7
     A dash  (-)  indicates no  improvement  in sustainable  recreational use  over  the

     next lower  standard or over  the base case  as  appropriate.

-------
                                                                              5.65
                                      Figure 2-A
DO
ppm
7.0
6.5
6.0
5.5
5.0
4.5
4.0
3.5
Sustainable Associated Annual Marginal ,
Activity Shorthand Willingness to pay per person
Swimmable S $35.4
(plus fishing
and boating)
Game Fishable G $19.1
(plus boating)
Beatable B $20.5
Unacceptable U 0
for boating
a Source: William J. Vaughan, 1981, "The Water Quality Ladder," Appendix II In Robert C.
Mitchell and R. T. Carson. An Experiment in Determlnlne Willineness to Pav for National
Water Dual
itv Improvements (Washington, DC: Resources for the Future, draft report).
Source:  Smith and Desvousges [1986].

-------
                                                                           5.66
                                                            Russell  and  Smith
                                   CHAPTER 5

                                  REFERENCES
Bauraol, William J. and Wallace E.  Dates,  1975,  The Theory of Environmental

     Policy (Engelwood Cliffs, New Jersey:   Prentice Hall).



Berndt, Ernst R.,  1976, "Reconciling Alternative Estimates of the Elasticity of

     Substitution," Review of Economics and Statistics.  Vol 58,  February,  59-

     69.



Bockstael, Nancy E.,  W. Michael Hanemann, and Ivar E. Strand, Jr., 1987,

     Measuring the Benefits of Water Quality Improvements Using Recreation

     Demand Models. Vol II (Department of Agricultural and Resource Economics,

     University of Maryland).



Cheuner. Andrev and Ian Jewitt, 198,', "Tne Zius ot a Heteroskedasticit>

     Consistent Covariance Matrix Estimator, Econometrica. Vol. 55, September,

     1217-1222.



Cicchetti, Charles J., Joseph J. Seneca and Paul Davidson, 1969, The Demand and

     Supply of Outdoor Recreation (New Brunswick, New Jersey: Bureau of

     Economic Research Rutgers University).



David, Elizabeth L.,  1971, "Public Perception of Water Quality," Water

     Resources Research. Vol 7, June 453-457.

-------
                                                                            5.67
Du Mouchel, William and Jeffery E.  Harris,  1983,  "Bayes Methods for Combining




     the Results of Cancer Studies  in Humans and Other Species," Journal of the




     American Statistical Association.  Vol.  78,  June,  293-308.








Fisher, Anthony C., 1981, Resource  and Environmental Economics (Cambridge:




     Cambridge University Press).









F^rsund, Finn R, 1972, "Allocation  in Space and Environmental Pollution."




     Swedish Journal of Economics.  Vol. 74,  19-34.









Freeman, A. Myrick III, 1984, "On the Tactics of Benefit Estimation Under




     Executive Order 12291," in V.  Kerry Smith,  editor, Environmental Policy




     Under Reagan's Executive Order (Chapel Hill, North Carolina:  University




     of North Carolina Press).








Freeman, A. Myrick III, 1982, Air and Water Polltoction Control:	A Benefit-^st




     Assessment (New -York:  John Wiley).








Freeman, A. Myrick III, 1979, The Benefits of Environmental Improvement:




     Theory and Practice (Baltimore:  Johns Hopkins University).








Hazilla, Michael and Raymond J. Kopp, 1986, "Systematic Effects of Capital




     Service Price Definition on Perceptions of Input Substitution," Journal of




     Business and Economic Statistics. Vol 4, April, 209-224.

-------
                                                                           5.68
Liroff, Richard, 1986, Reforming Air Pollution Regulation (Washington,  D.C.:



     The Conservation Foundation).








MacKinnon, Jams? G. and Halbert White,  1985, "Some Heteroskedasticity




     Consistent Covariance Matrix Estimators With Improved Finite Sample



     Properties," Journal of Econometrics. Vol. 29, 305-325.








McGartland, Albert M., Wallace E. Gates and Paul R. Portney,  1988, "The Net




     Benefits of Least-Cost Controls:  Setting Environmental Standards in the



     Real World," unpublished paper, Washington, D.C. , Resources for the




     Future.








Mitchell, Robert Cameron and Richard T. Carson, 1986, "The Use of the



     Contingent Valuation Data for Benefit/Cost Analysis in Water Pollution



     Control," unpublished report to U.S. Environmental Protection Agency,




     Resources for the Future, September.








Mitchell, Robert Cameron and Richard T. Carson, forthcoming, Using Surveys to



     Value Public Goods:  The Contingent Valuation Method (Washington, D.C.:




     Resources for the Future).








Naughton, Michael C.t George R.  Parsons, and William H. Desvousges, 1988,




     "Benefits Transfer:  Conceptual Problems  in Estimating Water Quality




     Benefits Using Existing Studies," unpublished paper. February.

-------
                                                                           5.69
Newey, Whitney K. and Kenneth D.  West,  1987, "A Simple,  Positive Semi-Definite,




     Heteroscedasticity and Autocorrelation Consistent Covariance Matrix,"




     Econometrica. Vol. 55, May,  703-708.









Portney, Paul R. , 1984, "The Benefits and Costs of Regulatory Analysis," in V.




     Kerry Smith, editor, Environmental Policy Under Reagan's Executive Order:




     The Role of Benefit-Cost Analysis (Chapel Hill, N.  C.:  University of




     North Carolina Press).








Revelle, Roger,  1985, "The Scientific History of Carbon Dioxide." in E. T.




     Sundquist and W. S. Broecker (eds.),  The Carbon Cycle and Atmospheric Co.:




     Natural Variations Archean to Present  (Washington,  D.C.:  American




     Geophysical Union).








Richmond, Harvey, 1983, "Criteria for Specifying Alternative Primary




     Standards," Memojardum, AmMent Standards Branch, U.S. Environmental




     Protection Agency, May 3.








Russell, Clifford, William J. Vaughan, and  Therese Feng, 1983,  "The Potential




     for Application in Benefit Measurement," in K. W. Turgeon,  (ed.) Marine.




     Ecosystem Modeling (Washington, D.C.:  U.S. Department of Commerce,




     National Oceanic and Atmospheric Agency).








Shaw, Daigee, 1988, "On-Site Samples Regression:  Problems of Non-Negative




     Integers, Truncation and Endogenous Stratification," Journal of




     Econometrics. Vol. 37 February, 221-224.

-------
                                                                           5.70
Siebert, Horst, 1985, "Spatial Aspects of Environmental Economics,"  in A.  V.




     Kneese and James L.  Sweeney (eds.),  Handbook of Natural  Resource  and




     Energy Economics  (New York:   North-Holland).








Smith, V. Kerry, 1988, "Selection and Recreation Demand," American Journal of




     Agricultural Economics. Vol.  70, February,  29-36.








Smith, V. Kerry, editor,  1984, Environmental Policy Under Reagan's Executive




     Order:  The Role of Benefit-Cost Analysis (Chapel Hill,  N.  C.:  University




     of North Carolina Press).








Smith, V. Kerry and William H. Desvousges, 1986, Measuring Water Quality




     Benefits (Boston: Kluwer Nijhoff).








Smith, V. Kerry and Yoshiaki Kaoru, 1988, "Signals or Noise?:  Explaining the




     Variation in Recreation Benefit Estimates," unpublished paper,  North




     Carolina State University, May.








Smith, V. Kerry and Yoshiaki Kaoru, 1987, "Recreation Benefits Transfer




     Project," Third Quarterly Report to U. S. Environmental Protection Agency,




     July 17.








Spofford, W. 0., Jr., C.  S. Russell, and R. A. Kelley, 1976, Environmental




     Quality Management:   An Application to the Lower Delaware Valley




     (Washington, DC:  Resources for the Future).

-------
                                                                           5.71
Tietenberg, Thomas, 1987, "Marketable Emission Permits in the United States:   A




     Decade of Experience," paper presented at conference Public Finance and




     Performance of Enterprise, International Institute of Public Finance,




     Paris, August 24-29.









Tietenberg, Thomas, 1985, Emission Trading:  An Exercise in Reforming Pollution




     Policy (Washington, D.C.:  Resources for the Future).








Tietenberg, Thomas H., 1978, "Spatially Differentiated Air Pollutant Emission




     Charges:  An Economic and Legal Analysis," Land Economics. Vol. 54, 265-




     277.









U.S. Environmental Protection Agency, 1987a, EPA's Use of Benefit-Cost




     Analysis:  1981-1986 (Washington, D.C.:  Office of Policy Planning and




     Evaluation, 1987).








U.S. Environmental Protection Agency, 1987b, Unfinished Business:  A




     Comparative Assessment of Environmental Problems.  (Washington, DC;




     USEPA, February).








Vaughan, William J. and Clifford S. Russell, 1982, Freshwater Recreational




     Fishing.  (Washington, D.C.:  Resources for the Future).

-------
                                                                           5.72
Vaughan, William J.,  1981, "The Water Quality Ladder," Appendix II in Robert




     Cameron Mitchell and Richard T.  Carson, An Experiment in Determining




     Willingness to Pay for National  Water Quality Improvements,  draft report




     to U.S. Environmental Protection Agency (Washington, D.C., Resources for




     the Future).








Ward, Frank A. and John B. Loomis, 1986, "The Travel Cost Demand Model as an




     Environmental Policy Assessment Tool:  A Review of Literature,"  Western




     Journal of Agricultural Economics. Vol II, December 164-178.








White, Halbert, 1980, "A Heteroskedastlcity-Consistent Covariance Matrix




     Estimator and a Direct Test for Heteroskedasticity," Econometrica. Vol.




     48, 817-83.

-------