EPA-230-12-85-021
September 1985
METHODS DEVELOPMENT FOR ENVIRONMENTAL
CONTROL BENEFITS ASSESSMENT
Volume III
FIVE STUDIES ON NON-MARKET VALUATION TECHNIQUES
by
David S. Brookshire, William D. Schulze, Ralph C. d'Arge
Thomas D. Crocker and Shelby Gerking
University of Wyoming
Laramie, Wyoming 82071
Mark A. Thayer
San Diego State University
San Diego, California 92182
USEPA Grant # R805059-01-0
Project Officer
Dr. Alan Carlin
Office of Policy Analysis
Office of Policy, Planning and Evaluation
U s . Environmental Protection Agency
Washington, D.C. 20460
OFFICE OF POLICY ANALYSIS
OFFICE OF POLICY, PLANNING AND EVALUATION
US . ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, D.C. 2 0460
-------
OTHER VOLUMES IN THIS SERIES
Volume 1, Measuring the Benefits of Clean Air and Water, EPA-230-12-85-019.
This volume is a nontechnical report summarizing recent research for EPA on
methods development for better estimates of economic benefits from environmental
improvement. The report 'presents the basic economic concepts and research methods
underlying benefits estimation as well as a number of case studies, including
several fran other volumes of this series. Finally, it offers insights regarding
the quantitative benefits of environmental improvement.
Volume 2, Six Studies of Health Benefits from Air Pollution Control, EPA-230-
12-85-020.
This volume contains six statistical epidemiology studies. They show that
large associations between health and current levels of air pollution are not
robust with respect to the statistical model specification either for mortality
or morbidity. They also find that significant relationships, mostly small, oc-
casionally appear.
Volume 4, Measuring the Benefits of Air Quality Changes in the San Francisco
Bay Area: Property Value and Contingent Valuation Studies, EPA-230-12-85-022.
This volume replicates a property value study conducted in the Los Angeles
Basin for the San Francisco Bay area. A taxonomy series of air guality types
and socioeconomic typoligies are defined for cities in the area to examine how
property values vary with pollution levels. The contingent valuation method
surveys individuals, directly asking their willingness to pay for changes in
air guality. The survey method yields benefit values that are about half the
property value benefits in both the Bay area and Los Angeles.
Volume 5, Measuring Household Soiling Damages from Suspended Particulate:
A Methodological Inguiry, EPA 230-12-85-023.
This volume estimates the benefits of reducing particulate matter levels
by examining the reduced costs of household cleaning. The analysis considers
the reduced freguency of cleaning for households that clean themselves or hire
a cleaning service. These estimates were compared with willingness to pay
estimates for total elimination of air pollutants in several U.S. cities.
The report concludes that the willingness-to-pay approach to estimate parti-
culate-related household soiling damages is not feasible.
Volume 6, The Value of Air Pollution Damages to Agricultural Activities in
Southern California, EPA-230-12-85-024.
This volume contains three papers that address the economic implications
of air pollution-induced output, input pricing, cropping, and location pat-
tern adjustments for Southern California agriculture. The first paper esti-
mates the economic losses to fourteen highly valued vegetable and field crops
due to pollution. The second estimates earnings losses to field workers ex-
posed to oxidants. The last uses an econometric model to measure the reduction
of economic surpluses in Southern California due to oxidants.
-------
due to pollution. Ihe second estimates earnings losses to field workers ex-
posed to oxidants. The last uses an econometric model to measure the reduction
of economic surpluses in Southern California due to oxidants.
Volume 7, Methods Development for Assessing Acid Deposition Control Benefits,
EPA-2 30-12-85-025. . . .
This volume suggests types of natural science research that would be most
useful to the economist faced with the task of assessing the economic benefits
of controlling acid precipitation. Part of the report is devoted to develop-
ment of a resource allocation process framework for explaining the behavior of
ecosystems that can be integrated into a benefit/cost analysis, addressing
diversity and stability.
Volume 8, Ihe Benefits of Preserving Visibility in the National Parklands of the
Southwest, EPA-230-12-85-026.
This volume examines the willingness-to-pay responses of individuals surveyed in
several U.S. cities for visibility improvements or preservation in several Nation-
al Parks. The respondents were asked to state their willingness to pay in the
form of higher utility bills to prevent visibility deterioration. The sampled
responses were extrapolated to the entire U.S. to estimate the national benefits
of visibility preservation.
Volume 9, Evaluation of Decision Models for Environmental Management, EPA-230-
12-85-027.
This volume discusses how EPA can use decision models to achieve the proper role
of the government in a market economy. The report recanmends three models useful
for environmental management with a focus on those that allow for a consideration
of all tradeoffs.
Volume 10, Executive Summary, EPA-230-12-85-028.
This volunB summarizes the methodological and empirical findings of the series.
The concensus of the anpirical reports is the benefits of air pollution control ap-
pear to be sufficient to warrant current ambient air quality standards. The report
indicates the greatest proportion of benefits frcm control resides, not in health
benefits, but in aesthetic improvements, maintenance of the ecosystem for recreation,
and the reduction of danages to artifacts and materials.
-------
DISCLAIMER
This report" has been reviewed by the Office of Policy Analysis, U.S.
Environmental Protection Agency, and approved for publication. Mention in
the text of trade names or commercial'products does not constitute endorse-
ment or recommendation for use.
ii
-------
FOREWORD
This volume is one of the reports prepared by research institutions under
cooperative agreements with the Economic Research Program of the united states
Environmental protection Agency (EPA). The pyrpose of the Program is to carry
out economic research, that will assist EPA in carrying out its mission. Until
very recently, most research sponsored by the Program sought to improve the
methods and data available for determining the economic benefits of pollution
control, thereby assisting EPA and other Federal Agencies responsible for
preparing benefit-cost analyses of programs and regulations. Such benefit-
wst analyses are required as part of the Regulatory Impact Analyses mandated
for most major Federal regulations by Executive Order 12291. The availability
of improved methods and data will make it possible for EPA and other Agencies
to determine more accurately the economic efficiency of their regulations and
programs. Very recently, the scope of the Program has been expanded to in-
clude a broader range of research on increasing the economic efficiency of
pollution control.
The Economic Research Program was a part of the Office of Research and
Development (GRD) until early 1983, when it was transferred to what is now the
Off ice of Policy, Planning and Evaluation. Hie cooperative agreements under
which this volume was prepared were concluded while the Program was still in
ORD; accordingly, CORD'S should be recognized.
This volume is one of a series under the title Methods Development for
Environmental Control Benefits Assessment prepared mainly under cooperative
agreement R805059 with the University of Wyoming, although several of the
individual volumes were completed under later cooperative agreements or under
subagreements with other institutions. Each of the other volumes in the series
is 1 isted on the f rent and back inside covers of this volume. The overall
purpose of the series is to report significant research results achieved under
the cooperative agreement. The purpose of the agreement was to develop im-
proved methods for assessing environmental benefits, with emphasis on air
pollution benefits. An earlier series of interim reports prepared under the
same cooperative agreement was published by EPA in 1979 under the series title
of Methods Development for Assessing Air Pollution Control Benefits with
report numbers EPA-600/5-79-001a through OOle.
This volume contains five analytical and empirical studies of alternative
techniques for valuing goods that are not marketed, with emphasis on some of
the difficulties with using benefit-cost techniques in analyses of air pollu-
tion control programs and measures. These studies are important to EPA be-
cause of the importance of determining the economic benefits of air (and
other) pollution control programs and measures and the present difficulties of
doing so. Only by solving these cliff iculties can EPA make reliable benefit-
cost estimates of the many benefits of its programs and regulations which are
not goods sold in markets.
Alan Carlin
off ice of Policy, Planning
and Evaluation
Washington, D.C.
iii
-------
ADS TRACT
This volume presents analytical and empirical comparisons of alternative
techniques for the valuation of nonmarketed goods. The methodological base of
the survey approach--directly asking individuals to reveal their preferences
in a structural hypothetical market--is examined for bias, replication and
validation characteristics. Upon finding in an experiment in the South Coast
Air Basin that the survey approach does not appear to be bias ridden,
satisfies some replication tests and was erossvalidated by the property value
hedonic technique, a simplified benefit-cost analysis was conducted. The
results imply that ambient air quality standards in the South Coast Air Basin
are probably economically justified, though uncertainty concerning the benefit
and cost calculations exists. To provide a third basis for comparison, the
wage--hedonic technique--where it is assumed that higher wages must be paid,
everything else held equal, to induce people to live in polluted communities,
was implemented on a trial basis for the Standard Metropolitan Statistical
Areas of Denver and Cleveland. The purpose was to explore if a relative low
cost technique could be utilized in achieving a national benefit estimate.
Given the research presented in this volume, it appears the three techniques
could be utilized in constructor.g a national benefit estimate.
iv
-------
CONTENTS
Foreword iii
Abstract , iv
Figures vi
Tables vii
Chapter 1 - Introduction 1
Chapter 2 - Valuing Experimental Commodities: Some
Recent Experiments 6
Chapter 3 - Valuing Public Goods: A Comparison of
Survey and Hedonic Approaches 37
Chapter 4 - The Advantages of Contingent Valuation Methods
For Benefit-Cost Analysis 60
Chapter 5 - An Examination of Benefits and Costs of Mobile
Source Control Consistent with Achievement of
Ambient Standards in the South Coast Air Basin .... 82
Chapter 6 - Effects of Air Pollution and Other Environmental
Variables on Offered Wages 117
-------
FIGURES
Numb er Page
3-1 Figure 1 ...... 41
3-2 Figure 2 .... • 44
4-1 Effect of an Improvement in Information on
Consumer's Surplus .......... 65
4-2 Effect of Costly Exchange. 70
5-1 Optimal and Excessive Standards. - 85
5-2 The Case of an Undesirable Standard. • 85
5-3 Value of Improved Air Quality (4)..-•>>•••>•> > 94
5A-1 Ozone Isopleths 112
vi
-------
¦¦ TABLES
Numb e r Page
2-1 Comparison of Results for Southwest Visibility Studies . . 20
2-2 Overview of Non-Market Valuation Experiments 28
3-1 Estimated Hedonic Rent Gradient Equations. ........ 48
3-2 Tests of Hypotheses 49
5-1 Base Year Emissions - 1975 and 1976 By Major Source
Category (Tons/Day) 88
5-2 1975-76 Emissions - Major Sources By County 90
5-3 Benefit Equation Coefficient 93
5-4 Annualized Benefit Estimates for Achieving the Federal
Ambient Standard (1978 $000) 96
5-5 Hydrocarbon Cost Estimates for Mobile Source Control ... 101
5-6 Carbon Monoxide Cost Estimates for Mobile Source Control . 103
5-7 Oxides of Nitrogen Cost Estimates for Mobile Source Control 105
6-1 Variable Definitions 121
6-2 Restricted Sample Regression 130
6-3 Pooled Sample Regression - Replacement with Means 131
6-4 Male, White, White Collar Worker, Household Heads
Aged 17-29 132
6-5 Male, White, White Collar Worker, Household Heads
Aged 30-49 133
vii
-------
6-6 Male, White, White Collar Worker, Household Heads
Aged 50-69 134
6-7 Male, White, Blue Collar Worker, Household Heads
Aged 17-29 135
6-8 Male, White, Blue Collar Worker, Household Heads
Aged 30-49 136
6-9 Male, White, Blue Collar Worker, Household Heads
17-69 137
6-10 Male, Non-White, white Collar Worker, Household Heads
Aged 30-49 138
6-11 Male, Non-White, White Collar Worker, Household Heads
Aged 17-69 ......... 139
6-12 Male, Non-White, Blue Collar Worker, Household Heads
Aged 17-29 ................... 140
6-13 Male, Non-White, Blue Collar Worker, Household Heads
Aged 30-49 . . 141
6-14 Male, Non-White, Blue Collar Worker, Household Heads
Aged 50-69 142
6-15 Male, Non-White, Blue Collar Worker, Household Heads
Aged 17-69 143
6-16 Female, Non-White, White Collar Worker, Household Heads
Aged 17-69 144
6-17 Female, White, White Collar Worker, Household Heads
Aged 17-69 145
6-18 Female, White, Blue Collar Worker, Household Heads
Aged 17-69 146
6-19 Female, Non-White, Blue Collar Worker, Household Heads
Aged 17-69 147
6-20 Pooled Sample Regression 148
6-21 Regression to Construct SOXM 149
viii
-------
6-22 Regression to Construct TSPM 150
6-23 Regression to Construct NOXM 151
6-24 Correlation Matrix 152
6-25 Means and 'Standard Deviations of Variables 153
6-26 Means of Non-Pollution Variables Used in Benefit
Calculations 154
6-27 National Air Pollution Standards 154
6-28 1978 Pollution Concentrations in Denver and Cleveland. . . 154
6-29 Cross-Tabulation of Incidence of Actual Pollution Data
by Partition 155
ix
-------
CHAPTER 1
INTRODUCTION
Benefit-cost analysis is a well established mode of applied economics
extensively used for the evaluation of public investment projects and most
recently environmental policies. This volume deals with some of the special
difficulties in the use of benefit-cost analyses in programs designed to
preserve or maintain air resources. The specific task is to estimate the
benefits associated with alternative levels of air quality. If benefit-cost
analysis is to be employed for decisionmaking, techniques need to be devised
to impute economic values for changes in the quality of air resources.
Two approaches have been proposed for measuring the value of non-market
goods . The most widely accepted approach has been the use of hedonic prices,
where it is assumed, for example, that either wages or housing values reflect
spatial differences in the quality of air resources. Alternatively, using
survey techniques, one may directly ask households or individuals to state
their willingness to pay for alternative levels of visibility. The necessity
for an alternative approch, to the hedonic, lies in the spatial nature of air
resources. In a well developed housing market, the hedonic approach is
appropriate. However, consider the case of a remote and unique scenic vista,
valuable to recreators, which is threatened by air pollution from a proposed
coal fired power plant, a typical situation in the western United States.
Although it is possible, in principle, to impute the value of clean air and
visibility from the relative decline in local visitation which might follow
construction of a power plant, information on the value of visibility at the
site is needed prior to construction for socially optimal decisionmaking. The
hedonic approach is unavailable both because the scarcity of local population
makes use of wage or property value data impossible and because scenic vistas
may themselves be unique.
The empirical implementation of the survey approach, however, raised
questions of bias, replicability, validation by other techniques, and appro-
priateness for benefit-cost analysis given the hypothetical nature of the
technique. Before incorporation of survey approach results into benefit-cost
analysis these questions require answers. Accordingly, the chapters that
1
-------
follow address the following topical areas.
In Chapter 2--Valuing Environmental Commodities: Some Recent
Experiments, we evaluate the results of six recent experiments which have
utilized the survey approach for estimating a nonmarket attribute associated
with the environment. Where possible, the issue of replication of results is
addressed. The range 'of environmental attributes valued in the six
experiments was quite large--noise, wildlife, strip mining, and visibility.
Four out of six attempted some internal methodological cross check. Biases,
within the survey approach, do not appear to be an overriding problem.
However, the studies indicate the need to establish a precise market,
hypothetical in nature, for the survey approach to be useful.
In Chapter 3--Valuing Public Goods: A Comparison of Survey and Hedonic
Approaches, we take up the central issue of validating the survey approach.
Although the results of Chapter 2 suggest the survey approach is internally
consistent, replicable and consistent with demand theory, no external
validation had been undertaken whereby a comparative analysis using another
approach independent of the survey had been conducted. Thus , the purpose of
this chapter is to report on an experiment designed to validate the survey
approach by direct comparison to a hedonic property value study.
The Los Angeles metropolitan area was chosen for the experiment due to
the well defined air pollution problem and because of the existence of
detailed property value data. Twelve census tracts were chosen for sampling
wherein 290 household interviews were conducted during March, 1978.
Respondents were asked to provide their willingness to pay for an improvement
in air quality at their current location. Air quality was defined as poor,
fair, or good based both on maps of the region (the pollution gradient across
the Los Angeles Metropolitan Area is both well defined and well understood by
local residents) and on photographs of a distant vista representative of the
differing air quality levels. Households in poor air quality areas were asked
to value an improvement to fair air quality while those in fair areas were
asked to value an improvement to good air quality. Households in good air
quality areas were asked their willingness to pay for a region-wide
improvement in air quality.
For comparison to the survey responses, data was obtained on 634 single
family homes sales which occurred between January, 1977, and March, 1978,
exclusively in the twelve communities used for the survey analysis.
Households, in theory, will choose to locate along a pollution-rent gradient,
paying more for homes in clean air areas based on income and tastes. However,
ceteris paribus, we show that the annualized cost difference between homes in
two different air quality areas (the rent differential for pollution) will in
2
-------
theory exceed the annual willingness to pay for an equivalent improvement in
air quality for a household in the lower air quality area. Thus, the rent
differential associated with air quality improvement from hedonic analysis of
the property value data must exceed estimates of household willingness to pay
for the survey responses, if the survey responses are a valid measure of the
value of air quality improvements. The theoretical model described predicts
that survey responses will be bounded below by zero and above by rent
differentials derived from the estimated hedonic rent gradient. The empirical
results do not allow the rejection of either of the two hypotheses, thereby
providing evidence towards the validity of survey methods as a means of
determining the value of nonmarket goods.
In Chapter 4--The Advantage of Contingent Valuation Methods for Benefit-
Cost Analysis, we address why the survey approach is especially useful in pro-
viding information to be utilized in environmental decisions. The chapter is
taxonomic in nature discussing why survey methods may often be a superior
means of generating data with which to value nonmarket goods. Specifically
the issue of the hypothetical nature of the survey technique is addressed. We
argue that within the constructs of economic theory, it is wrong to view
hypothetical responses as fictional, and that the survey approach is quite
often the only technique which can address future events without going through
the costly exercise of actually constructing a market.
Chapter 5--An Examination of Benefits and Costs of Mobile Source Control
Consistent with Achievement of Ambient Standards in the South Coast Air Basin,
is an examination of the benefits and costs of the national ambient air
quality standards as applied to all portions of Los Angeles, Orange, Riverside
and San Bernardino Counties in southern California. The results set forth are
based on the qualified arguments presented in Chapters 24 suggesting that both
the survey approach and property value approach are valid techniques of
benefit-cost analysis. Based upon modeling contained in the region's
Air Quality Management Plan, achievement of the ambient standards in 1979
would require emission reductions of the 974 tons/day, 5963 tons/day and 503
tons/day of reactive hydrocarbons, carbon monoxide and nitrogen oxides. It is
the share of these emission reductions attributable to on-road mobile source
control which was evaluated using benefit-cost analysis.
Benefits were calculated through an examination of housing value differ-
entials attributed to air quality (see Chapter 4) . Achieving the ambient air
quality standards was consistent with improving the "fair" and "poor" air
quality regions to the "good" category as specified in Chapter 3. In effect,
this constituted an approximate 30 percent improvement in the fair areas and a
45 percent improvement in the poor air quality areas. Corresponding benefits
were estimated to fall between 1,6 and 3.0 billion dollars per year,
3
-------
independent of any benefits accruing to agriculture and ecosystems. The share
of these benefits associated with on-road mobile source control was estimated
to be 1.362.55 billion dollars.
Cost estimates were developed from existing data sources, primarily from
manufacturer statements and government publications. Given the variation in
control cost options and the uncertain nature of the cost figures, it was
found that on-road mobile source controls consistent with a policy sufficient
to achieve the ambient standards in 1979 would involve a cost of between .61
and 1.32 billion dollars, with a best estimate of 1.02 billion dollars.
The benefits from on-road mobile emissions reductions consistent with
satisfying the ambient standards are of the same order of magnitude as the
cost estimates. This implies that the ambient air quality standards are not
without some economic justification, though the uncertainty concerning the
benefit and cost calculations prevents one from accepting the controls
outright. However, on-road mobile controls consistent with the air quality
standards cannot be rejected as economically inefficient either. Therefore,
although the mid-range benefit estimate exceeds the mid-range cost estimate,
the situation is best characterized as highly uncertain. Further, the static
analysis performed does not answer significant questions concerning the
behavior of the benefit and cost functions over time. Stronger statements
could only be made in the context of a much more detailed analysis supported
by a solid cost data base.
In Chapter 6--Effects of Air Pollution and Other Environmental Variables
on Offered Wages--we report on some exploratory estimates of the effect of
changes in air pollution levels on offered wage rates. This approach is
appropriate for a national benefits study where it is assumed that higher
wages must be paid, everything else held equal, to induce people to live in
polluted communities.
Annual benefit estimates from pollution abatement in the two cities are
positive according to the calculations made here. For Denver, meeting the
national secondary standards for TSP results in a reduction in the offered
real wage, from $4.1758/hr. to $3.9626/tir. Multiplying this difference of
$.2136/hr. by the number of persons affected times 2000 hours yields an esti-
mated annual benefit for Denver of $92,968,935. A similar calculation for
Cleveland reveals that meeting the national secondary air quality standards
causes the real wage to fall from $3,8756/hr. to $3.7693/hr. implying a
benefit of $81,360,489. Note that benefits per household head in the two
cities are $426.35 for Denver and $212.60 for Cleveland. This preliminary
research suggests the wage hedonic technique is viable for estimating air
pollution control benefits for standard metropolitan areas across the nation.
4
-------
A national benefit estimate for air pollution control based on consumer
perceptions as reflected in wages and property values appears possible.
Further, the use of the survey approach to assess the value of perceived
benefits, such as visibility improvements, not captured by wages and property
values appears feasible.
5
-------
Chapter 2
VALUING ENVIRONMENTAL COMMODITIES: SOME RECENT EXPERIMENTS
INTRODUCTION
During the past few years, economists have been attempting to apply a
variety of techniques to reveal preferences of individuals on nonmarket
environmental commodities [Bradford (1970); Bohm (1971); Randall, et al.
(1974a); Brookshire, et al. (1976)]. These techniques, in general following
Davis (1963), have attempted through a set of questions to obtain bids from
individuals which would represent their maximum willingness to pay for a non-
market commodity. Almost simultaneously, other economists have made sub-
stantial contributions to conceptually assessing the demand for nonmarket
commodities and public goods [Rosen (1974); Muellbauer (1974); Hori (1975)].
The consumer of nonmarket commodities in these studies is viewed as a utility
maximizer who combines purchase of private goods (and use of public goods),
constrained by a household technology, to produce a set of desired character-
istics [Lancaster (1966)]. Given this basic structure, methods were suggested
for calculating implicit prices for the household characteristics and non-
market goods used or produced by the consumer [Hori (1975)]. This paper is an
assessment of six recent experiments which have attempted to reveal prefer-
ences for environmental goods, where each experiment in some way utilized a
mix of both the techniques and the theory of demand for nonmarket commodities.
Each experiment was designed to estimate a nonmarket attribute associated with
the environment and also analyze potential biases in the techniques employed.
In order to evaluate these studies, a rather general model of the consumer
behavior is proposed. Potential biases are then discussed for various methods
used to discover environmental preferences. Following this, the six
experiments are examined on the basis of their methodological structure and
types of biases encountered. Where possible, the issue of replication of
results is addressed. In this paper, we do not attempt to evaluate all types
of environmental effects or possible ways of measuring them. Rather, a
limited set of possibilities is examined both as to technique and type of
effect
Many environmental policy issues involve changes in environmental
6
-------
attributes resulting from population growth and energy development. For
example, operation of coal fired electric power plants may significantly
reduce visibility and disturb landscapes in addition to inducing possible
health effects. Strip mining coal may have substantial detrimental effects on
wildlife populations in addition to the expanded demand for wildlife arising
from a larger local population. The construction of geothermal plants
adjacent to existing forest recreation areas may, through siting and noise,
disturb an otherwise pristine, quiet recreation area. Essentially,
recreational use and benefits would be changed by these developments but there
are no existing markets to adequately price the changes. Similarity,
population growth in some urban areas has caused significant problems with
photochemical air pollution. If benefit-cost analysis is to be employed for
decisionmaking, techniques need to be devised to impute economic values for
these and other environmental changes. The six experiments evaluated here are
tests to determine the feasibility of deriving implicit prices and/or
valuations for the types of changes mentioned above.
The techniques to be examined range from purely hypothetical direct
evaluations asking for dollar bids to hypothetical questions asked of house-
holds and recreators concerning changes in behavior to enable the imputing of
their preferences. In each case, the household was confronted with a possible
change in an environmental attribute and asked for a valuation. Since the
valuation was contingent on the specific hypothetical change identified
(through photographs, brochures and other ans), we propose that such
approaches be called contingent valuations. Individuals can be queried as to
willingness to pay, minimum compensation, evasive behavior, past experiences,
current experiences, potential site or activity substitutions, potential
expenditure adjustments, income compensation coupled with potential behavioral
adjustments, etc. which can be utilized with appropriate theoretical
structures to estimate demand curves for environmental attributes.
In some of the experiments, the household valued the change in environ-
mental attribute directly by bidding for alternative provision levels [Brook-
shire, et al. (1976)]. In others, the individual was not only asked to bid,
but also provide information on behavioral adjustments and sale of the
environmental attribute [Randall, et al. (1974a); Rowe, et al. (1980);
Brookshire, et al. (1980)].
To obtain accurate information for individual valuation of nonmarket
environmental commodities can be costly. In many cases to actually derive
true values, a "market" must be set in a place where one did not exist and
operated to record prices and demands where the environmental attribute is
actually purchased or sold. However, to construct and operate such a market
may be extremely costly, especially if there are irreversibilities associated
7
-------
with its operation. A less costly approach is to use a contingent valuation
study where prices can be imputed without the actual operation of an organized
market, but a hypothetical market is structured. However, because of its
hypothetical nature, several potential biases may occur. The major types of
biases are: (1) strategic bias whereby the individual may attempt to in-
fluence the outcome or result by not responding truthfully; (2) information
bias, which is a potential set of biases induced by lack of, or type of,
information given to the consumer in the contingent market; (3) instrument
bias, which is bias introduced by the process or procedures employed to dis-
cover preferences; (4) hypothetical bias, which is the potential error induced
by not confronting the individual with an actual situation, i.e., an organized
market with well-defined prices; or sampling, interview or nonrespondent
bias. Clearly, asking someone what they will do or pay a priori is not the
same as confronting them with a recognized and well-understood market and
observing what they actually pay. It is more analogous to the individual
making decisions on contingent events, e.g. , if air quality deteriorates, I
will move to a cleaner community. Of the list of studies summarized herein,
the list directly compares a contingent valuation study with a more
traditional property value study to assess the magnitude of these potential
biases and attempts to resolve the actual versus hypothetical payment question
[Brookshire, d'Arge, Schulze, and Thayer (forthcoming (d) )] .
A THEORETICAL FRAMEWORK FOR VALUING ENVIRONMENTAL AMENITIES
The variety of empirical approaches used to value environmental ameni-
ties, whether examining contingent or actual behavior or market prices, have
typically been based on a particular ad hoc theoretical structure. This
section attempts to provide a common theoretical basis for the variety of
approaches outlined earlier and serve as a focus in evaluating the six
experiments presented in later sections. Both Freeman (1979a,b) and Maler
(1974) have also examined available approaches from a consistent theoretical
perspective.
A general modeling structure must include the possibility of consumer
substitution across activities and locations; and must include site or ac-
tivity specific levels of environmental quality. Individual utility is thus
specified as a function of levels of activities, A , . . ., A,, . . ., A ; as
a function of a composite commodity X, "unaffected^ by activity specific
environmental quality; and (where the subscripts denote different activities)
as a function of environmental quality for each activity, Q , . . . , Q, , . .
., Qn, where we take increases in Q as increasing environmental quality.
Note we can allow possibly different environmental quality levels both by
varying Q for a specific activity A which can occur over many sites or by
i "i
defining a site specific activity in which case different Q/s are associated
8
-------
with different sites. Utility is then a quasiconcave function,
. . . , A^; Q^s . . .» Q ; X) (1)
i 2.
where Wl*h± = U^O, SU/SQ^. = U„>0, and 3U/3X = ^0 so utility is increasing
in A^, Q., and X, Of course, a Sumptions on the separability of U are
obvious, g^Lven that'environmental quality is related to specific activities in
the model." However, we do not pursue that issue here. Rather, we focus on
the form of an individual's marginal willingness to pay for environmental
quality.
The budget constraint necessary to specify the consumer's optimization
problem is given as:
n
Y-f PA-X>0 (2)
11 ~~
i= 1
n
or income Y minus the sum of expenditures on activities^ £ PA. (P istaken
as the price of activity i which may, in fact, represent joint consumption of
several market commodities; for example, activities might include driving to
work, recreating, shopping, etc.) minus expenditures for the composite
consumption commodity X (price for X is taken as unity to simplify the
analysis) must be nonnegative.
For a given vector of environmental quality, the household will then
choose to allocate activities such that (1) is maximized subject to (2) which
in turn implies that:
• Ai « 0^ > 0 i - 1,2, . . .,n (3)
or the marginal rate of substitution between activity i and the composite
commodity X equals the price of activity i, if that activity is chosen (A >
o) . We, of course, assume X > 0.
To determine the marginal willingness to pay for environmental quality
for a particular activity, for example i = 1, we set utility as given in
equation (1) equal to a constant and totally differentiate the resulting
expression. By then taking the total differential of equation (2) , setting
dQ =0 for all i # 1, dP =0 for all i, and by using (3) we obtain
i i
U
A < p
U
X
i
Ca_
u
¦pi:
-------
u
-2. " - il (4)
U dQ
x 1
as the change i£ income necessary to offset a change in environmental quality
for activity 1. If the objecti^ is to determine the marginal willingness to
pay for environmental quality » one obvious approach is to simply
postulate in a surVey' questionnaire that Q increases by a small amount dQ ,
where market prices are hypothetically held constant, and request information
on the contingent willingness of the individual to give up income for an
increase in quality (so dY would be negative in this case) . This direct
approach, however, is open to questions of bias, a topic we take up in more
detail later.
A second approach is to actually assume that prices of activities do not
change in response to a change in environmental quality. For many recreation
situations this may well be a reasonable approximation. For example, if an
energy development such as a power plant disrupts a recreation site,
recreators may respond by driving further to other alternative sites. If no
entrance fees are employed or if such fees are institutionally fixed, if
driving costs--the price of gasoline, etc. --and prices of recreation equipment
do not change, then the assumption that dP = 0 for all i appears to be a good
one. In that case the marginal willingness to pay can be determined by again
setting utility in equation (1) equal to a constant and totally
differentiating the resulting expression, by using equation (3) and by
assuming dQ, =¦ 0 for all i # 1 and that dP, = 0 for all i, to obtain:
U dA,
Q = - \ P. l + dX
U " 1 dO dQ
1=1 Hl H1
Where prices are known, an estimate of the value of environmental quality can
then be obtained empirically by collecting data on dA./dQ , the compensated
change in the pattern of, for example, recreation activit Jes in response to a
change in quality, and on dX/dQ , the compensated change in expenditures not
related to recreation activities. Note that here we assume Q^ is tied to a
specific recreation site. Of course, the change in environmental quality can
be hypothetical, resulting in contingent changes in activities, or actual
crosssectional or time series data can be employed where environmental quality
varies over space or time. In any case, all studies to date focusing on
substitution of activities or commodities in response to changes in
environmental quality that we are aware of have assumed prices to be fixed.
In contrast to the above approaches, the hedonic approach, focusing on
price effects of changes in environmental quality, assumes that P , the price
10
-------
associated with and in turn 0,, varies, but that all other prices still
remain fixed. Thus, again assuming utility is constant and by totally
differentiating equations (1) and (2) where dP^/dQ^ = 0 for all i # 1, using
(3), we obtain:
1
< • . dP*
(6)
nX dqi
where we also assume dY = 0 since compensation is achieved at the margin
through the hedonic price gradient, dP dQ^. Thus , individuals are compen-
sated for lower levels of environments! quality by a lower price. As an
example of this approach, consider a study which uses differences in property
values to value air quality. Serious questions must be raised concerning the
reality of the assumptions that other prices remain unchanged in response to
differences in air quality. For example, if wages or golf fees vary with air
quality levels, property values may not fully capture the willingness to pay
for air quality. Note that in this case we assume that is environmental
quality associated with an activity or activities.
In summary, the marginal willingness to pay of consumers for environ-
mental quality can be determined as shown in our theoretical context by three
approaches. First, consumers can be directly asked to provide their marginal
willingness to pay, dY/dC^, Second, assuming no price changes occur,
information can be collected on dA^/dQ, and dX/dQ^, the substitution of
activities and expenditures which occurs in response to a change in
environmental quality. From these data one can impute a marginal willingness
to pay. Third, assuming all prices but one are invariant, the change in the
single remaining price, dPi , can be used to impute environmental benefits. Of
the three approaches, the one which requires the fewest a priori assumptions
and minimal data collection is the first, contingent bids derived utilizing
survey instruments. However, serious questions of possible bias remain. The
next section discusses possible biases in the survey questionnaire approach.
CONTINGENT VALUATION AND BIAS
Economists have argued that valuing public goods through a direct demand
revealing process such as a contingent market would yield biased results. The
principle theoretical support for this contention is the possibility of
strategic bias. However, as survey techniques to elicit contingent behavior
or bids have come into use--in part because development of energy resources in
formerly pristine environments allows no other techniques to be used-- other
types of bias have come to be regarded as just as important. These include
information bias, instrument bias, hypothetical bias and traditional problems
11
-------
of sampling, interviewer, and non-respondent bias. This section reviews our
current understanding of such biases.
Strategic Bias
Beginning with Samuelson's seminal work on public goods, it has been
supposed that direct revelation of consumer preferences for such goods--and,
of course, environmental quality is a public good--would be impossible
[Samuelson (1954)] . In particular, the free-rider problem would give indivi-
duals incentives to misstate their preferences. For example, if nearby resi-
dents were asked how much they were willing to pay to clean up the air near a
power plant and if they suspected that control costs would be borne by con-
sumers and owners elsewhere, local residents would have an incentive to over-
state their willingness to pay. On the other hand, if residents suspected
that they would be individually taxed an amount equal to their own willingness
to pay, then a clear incentive would exist to understate their own true value,
hoping that others would bid more.
Each approach for eliciting willingness to pay will potentially generate
its own bias. Thus if recreators are told that the average of their bids to
prevent construction of a nearby power plant will be used to set an entrance
fee, those individuals who suspect their bid to be greater than the average
bid will have an incentive to overstate their willingness to pay. They, in
fact, have an incentive to raise the average bid as close as possible to their
own true bid. In other words, individuals will have incentives to misstate
their own preferences in an attempt to impose their true preferences on
others. This will require a substantial amount of information to actually
behave in this manner [See Brookshire and Eubanks (forthcoming (a))]. Of
course, if the respondents to such a survey do not believe the survey will
have any impact on policy or outcomes, then no incentives for bias exists. The
hypothetical nature of such surveys may then, in actuality, aid in eliciting
bids which are not strategically biased. Alternatively, since payment is not
required, a tendency to exaggerate willingness to pay for a preferred outcome
might also exist.
Empirical evidence thus far does not support the existence of strategic
bias among consumers. Bohm (1971) in an experimental approach utilizing
actual payments for public television failed to find strategic bias signifi-
cantly affecting the outcome. Scherr and Babb (1975) utilized three different
mechanisms for valuing public commodities and found little evidence supporting
the existence of strategic bias. Smith (1977) in laboratory experiments also
failed to find strategic bias "as a significant problem. The case studies to
be reported in the next section, where tested for, also do not find strategic
bias to be a problem.
12
-------
Information Bias
Since contingent behavior or valuation is hypothetical, it is clear that
answers obtained through surveys are not based on information similar to that
which would apply if consumers based answers on real expediences. One is an
ex ante response while the other is an ex post statement. Typically,
consumers do reevaluate decisions on the basis of experience and gained know-
ledge. Thus, an individual or household might respond to a hypothetical de-
crease in environmental quality at one location with a low bid, thinking that
other nearby sites would make good substitutes. However, in a real situation
the individual might have found that other sites involved more travel costs
and were less satisfactory than imagined. The information presented to the
respondent in a survey situation relating to substitution possibilities and
alternative costs may well change the stated willingness to pay relative to
other types of information. Thus information bias can refer to the structural
content of the contingent market being different than the valuation problem at
hand. That is, the respondent must be made aware of proposed alternatives in
terms of quality or quantity. Other variants of information bias might
include giving the respondent information as to how other respondents behaved,
whether in the aggregate their bid was sufficient to achieve (or not achieve)
the stated goal (i.e., possibly prevention of visibility deterioration) or
alternative sequencing of questions.
Instrument Bias
Related to information bias is instrument bias whereby characteristics of
the mechanism for obtaining willingness to pay possibly influence the outcome.
Two characteristics of the survey bidding approach are vehicles for payment
and a starting point for initiation of the bidding process. Studies have
recognized that the mechanism used to collect the bid or pay compensation may
influence its magnitude [Randall, et al. (1974a)]. That is, if the recreator
pays a higher park entrance fee rather than another type of tax, his bid for
an environmental attribute may differ. From economic theory, the bid should
differ, if the price of the commodity represented by the bidding vehicle
changes, provided the recreator's substitution possibilities associated with
alternative payment mechanisms are different. When a payment vehicle allows
the individual to substitute over a wider range of current commodities
purchased, then the bid should be higher or compensation lower than where the
range is smaller. Ideally, the bid or compensation should be related to
adjustments in disposable income or wealth, where the individual has the
greatest latitude for potential substitution. Practically, however, a
believable payment mechanism related to income adjustment, in general, cannot
be applied. For example, surveys are often taken at recreation sites away
from the individual's locale or state. In this case, a wage tax may not be
13
-------
viewed as realistically payable by the recreator. Thus, there is a tradeoff
between accuracy associated with a less than ideal method of payment and the
believability of the vehicle for payment or compensation. The reduction in
substitution possibilities for a more believable payment mechanism is likely
to reduce the contingent expenditure or increase the compensation estimate.
A second type' of• instrument bias is starting point bias. The contingent
valuation approach commences with questions on payment (and/or compensation)
for hypothetical changes in environmental attributes. Contingent bidding sur-
veys to date have asked the recreator (or any type of interviewee) a question
with a "yes" or "no" answer rather than a question requiring explicit calcu-
lations [See Randall, et al. (1974a), Brookshire, et al. (1976)]. It is pre-
sumed the recreator can more accurately respond to the yes/no question frame-
work, although to our knowledge, this proposition has not been formally tested
for individuals responding to contingent valuation questions. Given the pro-
position that yes/no responses are desirable, often a starting bid or minimal
level of compensation has been suggested. The potential bias arises in sug-
gesting a starting point from at least two possible sources. First, the bid
itself may suggest to the individual the approximate range of "appropriate
bids." Thus the individual may respond differently depending on the magnitude
of the starting bid. Second, if the individual values time highly, he may
become "bored" or irritated with going through a lengthy bidding process. In
consequence, if the suggested starting bid is substantially different from his
actual willingness to pay, the bidding process may yield inaccurate or only
roughly approximate results. The effect of these two types of starting point
biases may substantially influence the accuracy of contingent valuation and
therefore the usefulness of this approach for assessment of environmental
preferences.
Hypothetical Bias
The discussion on information bias suggested that the contingent valu-
ation approach will give answers dependent upon the information or "state of
the world" described. The contingent valuation approach requires postulating
a change in environmental attributes such that it is believable to the indivi-
dual and accurately depicts a potential change. The change must be fully
understandable to him, i.e. , he must be able to understand most, if not all,
of its ramifications. The individual also must believe that the change might
occur and that his contingent valuation or behavioral changes will affect both
the possibility and magnitude of change in the environmental attribute or
quality. If these conditions are not fulfilled, the hypothetical nature of
contingent valuation approaches will make their application utterly useless.
A test of hypothetical bias would require that the perturbation proposed would
occur and then the respondents actual reaction would be evaluated in terms of
14
-------
the previous hypothetical statements of willingness to pay. This, however,
makes it extremely difficult to measure the extent of hypothetical bias within
a contingent experiment since it depends not only on the structur^ of the
experiment, but also on the "uncontrolled" factors of the future.
Other Bias
Any survey approach, including the contingent valuation approach, is
subject to sampling bias, non-respondent bias and interviewer bias. Any of
these certainly can subject the results of an experiment to question even if
all previously mentioned bias are non-existent. Given the acknowledgement of
these biases, we will not discuss them in detail here given their wide recog-
nition in the survey literature. However, in discussing the case studies in
the next section, the possible existence of these biases will be discussed in
each study, where the information is available.
VALUING ENVIRONMENTAL QUALITY: RECENT CASE STUDIES
There have been numerous efforts to apply a variety of techniques for
valuing non-marketed goods; public television [Bohm (1971)]; land-form alter-
ations due to strip mining [Randall, et al. (1978)]; air pollution-induced
health effects [Loehmati, et al. (1979)]; wildlife [Hammack. and Brown (1974),
Bishop and Heberlein (1979)] ; water pollution [Gramlich (1977)] ; presentation
of river headwaters [O'Hanlen and Sinden (1978) and Sinden and Wyckoff
(1976)]; urban infrastructure allocations for expenditures and taxes [Strauss
and Hughes (1976) and Cummings et al. (1978)]; airplane safety [Jones-Lee
(1976)]; and recreation [Davis (1963)].
This section will summarize in chronological order six studies which have
in common the use of a survey technique which had its first empirical
application by Randall, et al. (1974a,b). (The Randall, et al. study was the
first systematic presentation and empirical implementation of the contingent
bidding survey approach which set the stage for further inquiries.) Tracing
the methodology development which has occurred through these six studies aids
in understanding issues relating to bias problems, replication issues and
methodological cross checks. The last study discussed, the South Coast Air
Basin Experiment, addresses the question of validation of the contingent mar-
ket approach by direct comparison of contingent results with a hedonic--market
data based-study.
The Four Corners Experiment
The Four Corners Experiment [Randall, et al. (|974a,b)] represented the
first empirical application of the survey approach. The roots of the effort
15
-------
can be traced to Davis (1963) and Bohm (1971) . The focus of the study was to
investigate the impacts of Navajo coal strip mine and the Four Corners elec-
tric generating plants in the Southwest region. Specifically, aesthetic ben-
efits of abatement of environmental damage resulting from air pollution (visi-
bility) , power lines and land disturbance from mining activities were esti-
mated. As such, the study laid the framework for future contingent valuation
studies. " ¦
The analysis focused on the design of survey instruments exploring
alternative mechanisms within the instruments for eliciting willingness to
pay. No bias tests (i.e., hypothetical, information, instrument, interviewer,
non-respondent sampling bias tests) were formally reported.
The Lake Powell Experiment"*"^
Lake Powell, with an annual visitation now approaching two million
visitor days, is an excellent example of the tradeoff between preservation and
development. The lake was formed by the filling of Glen Canyon but retains
the steep cliffs, rugged terrain features, and scenic vistas one associates
with the Grand Canyon, and is now accessible to pleasure boaters and other
recreators. Construction of the Navajo coal-fired generating station located
at the southern end of Lake Powell was completed in 1976. Another larger
plant, the Kaiparowitz Project, was also proposed for construction near Lake
Powell and became an issue of substantial public concern.
As part of the Lake Powell experiment, during the summer of 1974,
recreators at Lake Powell were interviewed in an attempt to determine the
aggregate willingness to pay to prevent construction of the proposed
Kaiparowitz plant [See Brookshire, et al. (1976)]. Photographs of the
existing Navajo power plant which all of the recreators had seen stacks remain
visible more than 20 miles up the lake were shown to recreators both with
visible pollution emanating from the stacks and with the stacks alone.
Recreators were then asked what entrance fee they would be willing to pay to
prevent construction of another similar plant, first, where only pollution
would be visible from the lake itself, and second, where both stacks and pol-
lution would be visible.
The analysis of the data focused on strategic bias. As noted above, if
recreators believed that a uniform entrance fee might actually be set on the
basis of the average bid of the sample to prevent construction or believed
that construction plans might be affected by the research results, then
"environmentalists" might well bid very high, a^ "developers" might well bid
zero dollars in an attempt to bias the results. A theoretical model of
strategic bias was constructed to explain the distribution of observed bids
16
-------
4
which would likely be bimodal rather than normally distributed if strategic
bias was present. The fact that the actual distribution of bids was normally
distributed was thus taken as evidence that strategic bias was not present.
It was suggested by Brookshire, et al. (1976), that the absence of strategic
bias might be due to the hypothetical nature of the experiment--few
respondents felt that their answers would affect real world outcomes.
Hypothetical, information and instrument bias were not addressed in this
experiment. Experimental biases such as interviewer, non-respondent bias and
sampling bias did not appear significant. The interviewers taken separately
had means and a distribution of bids that corresponded to the sample popula-
tion as a whole. In sampling which was randomly conducted for the four prin-
cipal users of Lake Powell, on the lake, in campgrounds, at motels and in the
town of Page, the highest refusal rate for residents was less than one
percent.
The remainder of the research was devoted to specifying an econometric
model of the bidding game results to estimate income effects by groupre-
creators were divided into four categories, developed and remote campers, and
visitors to and residents of the nearby town of Page, Arizona. Although the
effect of individual income by group on bids was statistically significant at
least 99% level, the income effects were all very small. It was demonstrated
that both theoretically and empirically the small income effect implied: (1)
that a compensated surplus measure would not differ practically from the equi-
valent surplus measure used in the experiment; and, (2) that income redistri-
bution between groups would not significantly affect the aggregate bid.
The average bid per family or recreator group was $2.77 in additional
entrance fees in 1974 dollars, and the total annual bid--which can be inter-
preted as an aggregate marginal willingness to pay to prevent one additional
power plant near Lake Powell--was over $700,000. An important point is that
the results show impressive consistencies both with the one previous study
[Randall, et al. (1974a)] in the region as well as with the succeeding
Farmington experiment discussed below.
12
The Farmington Experiment
This study reported in Blank, et al. (1977) and Rowe, et al. (1980)
attempted to establish the economic value of visibility over long distances
for Farmington residents and recreators at Navajo Reservoir. Clearly, the
ability to observe long distances is almost a pure public good. In addition,
efforts were made to examine the extent of certain biases which the Brookshire
et al. (1976) study identified. These were information, strategic, starting
point, and instrument biases on compensating and equivalent surplus measures
17
-------
of consumer surplus.
Recreators and residents in the Four Comers Region of New Mexico and
Arizona were interviewed. The interviewee was shown a set of pictures de-
picting visible ranges. Picture set C had a visible range of 25 miles and
picture sets B and A were 50 and 75 miles respectively. The pictures repre-
sented views in different directions from the same location, the San Juan
Mountains and Shiprock.
The first part of the experimental bidding game was structurally similar
to that of Randall, et al. (1974a»b) and Brookshire, et al. (1976). A
sequence of questions on maximum willingness to pay and minimum compensation
were asked via a survey instrument. The second method followed that of Rosen,
(1974) , Muellbauer (1974) , and Hori (1975) in attempting to utilize the house-
hold production function. The motivation was to attempt a methodological
cross check by collecting market type information via a survey instrument.
The contingent behavior component of the questionnaire attempted through con-
tingent changes in time allocation to infer an expenditure function and com-
pensated demand curve, primarily by postulating an exact form of a utility
function and estimating a time related household technology [Blank, et al.
(1977)]. Thus, the first approach bidding game was an attempt to measure the
right-hand-side of equation (4), while the second contingent behavior based on
contingent behavioral changes, attempts to measure the components of the
right-hand-side of equation (5) . These estimates from the contingent bidding
and contingent behavior portions of the experiment are not directly comparable
because the contingent behavior estimates include residents in addition to
recreators which should increase the magnitude of the estimate.
As part of the contingent bidding approach, direct tests were made for
strategic bias, information bias, and instrument bias. First, for strategic
bias investigation, the survey instrument was structured so tlj individual was
told that he would have to pay the "average" bid, not his own. The
presumption was that if his bid were below the mean bid provided by the inter-
viewer and he desired to increase the magnitude of the final aggregate bid
strategically, he would bid higher in order to shift the final bid upward.
Alternatively, if his goal in bidding strategically was to reduce the final
mean bid, he would revise his bid downward. Only in the unlikely case when
the individual's maximum bid is identical to the mean bid would there be no
incentive for the individual to change. In only one case was an individual
observed acting strategically and he turned out to be an economics professor
from the local Junior College! This additional indication along with the
results of Brookshire, et al. (1976) suggests that individuals generally do
not act strategically, at least in a meaningful manner to bias the outcome of
the results.
18
-------
For information bias, it was suggested to the individual that his or her
bid was too low--that the bid was not sufficient to keep power plant emissions
at present levels for sustained high quality ambient air. The individual was
then asked if he or she would revise the bid. Fully one-third revised their
bid when confronted with the possibility that their bid was insufficient.
This latter result is indicative of the effect that new information possibly
has on bidding behavior.
Analysis was made of various forms of instrument bias, essentially trying
to establish influences of various aspects of the contingent market structure.
It was observed that the higher the starting bid suggested by the interviewer,
the higher the maximum willingness to pay (equivalent surplus) estimates
derived from the study. Thus, if the interviewer suggested a bid of $1.00
higher, on the average, individuals would "bid" about $.60 more. Also,the
choice of method of payment influenced the magnitude of the bid significantly.
Individuals were willing to bid higher when confronted with a "payroll tax"
than with an increase in entrance fees. Finally, it was observed that whether
or not the individual was given previous information on average bids, has a
substantial impact on the maximum bid. We do not wish to suggest these
results indicate any final conclusions with regard to the information bias
problem with the contingent valuation approach, but they are suggestive that
for these approaches to be accurate, one must be very careful with the
instrument used for payment and the amount and quality of information given to
the interviewee upon initiation of the interview.
Other potential biases--sampling, non-respondent bias and interviewer
bias--are also of interest. The sample design attempted a stratified sample
with respect to household income, ethnic background, age, sex and resident/
nonresident. After identifying neighborhoods with certain characteristics and
times of day appropriate for finding males and females at home, two approaches
were utilized in obtaining interviews: randomly going door to door and
telephoning to set up an interview time. A significant non-respondent bias
might exist for the Farmington resident interviews. Up to 75% of the phone
call requests for an interview were rejected and up to 50% of the door to door
requests were declined. However, for the recreators' interviews at Navajo
Reservoir, less than 5% of the requests for interviewing were declined. Why
this disparity for responses between residents and recreators is not known.
Finally, no records were kept that would enable an investigation of
interviewer bias.
It is interesting to compare results of the Farmington study with pre-
vious studies. Randall, et al. (1974a) only reported, and Brookshire, et al.
(1976), only obtained equivalent surplus bids. The following comparisons
which are presented in Table 1, are, therefore, limited to the equivalent
19
-------
Table 2.1
COMPARISON OF RESULTS FOR
SOUTHWEST VISIBILITY STUDIES®
Non-Market Valuation
Studies
Public
Good
Vehicle
Employed
Yearly Mean
Bids
Bid Per Day
1.
Four Corners Experiment
(A. Randall, et. al.,
197'ia.b)
Visibility
Spoil banks
transmission
lines (Aesth-
etics of the
above. )
Sales
Tax
me
[4,3119
$50
13.02]
(N/A )f($l,79)d
1.191
2.
Lake Powell Experiment
(D. Brookshire, et. al.,
1976)
Visibility
(Aesthetics
only)
Access
fee
N/A
N/A
$2..95b ($1.52)
[.20] 1.291
3.
Farmington Experiment,
(F. Blank, et. al., 1977
and Rowe, et..?]. , 1980)
Visibility
(Aesthetics
only)
utility
bills or
wage tax
$82
19.10)
$57
[4.63]
$2.'t'tC (N/A)
1.23]
"The Four Corners Experiment and the Lake Powell Experiment only obtained equivalent surplus
bids, thus comparisons between studies are limited to sub-samples of the data sets from each study.
Adjusted for 6.6% inflation.
CMean bid for $1.00 starting points it) the Farmington Experiment which is the starting point used
in the Lake Powell Experiment.
d
The comparison between the Four Corners Experiment and the Lake Powell Experiment required
different comparisons with the Fa/mlngton Experiment.
'The comparisons between the Four Corners Experiment and the Farmington
Experiment is for two alternative levels of environmental quality changes.
'n/A - No comparison can be constructed.
'Standard errors in [ ].
-------
surplus bids. Using the sales tax as the^n trument, Randall, et al. (1974a),
reported yearly mean bids of $85.00 [$4.31] for moves from the highest level
of environmental damage, situation (A) , to situation (C) representing lowest
levels of environmental damage; situation (B) represented an intermediate
level of damage. A yearly mean bid of $50.00 [$3.02] per household was
reported for moves from situation (B) to situation (C) . The Farmington
experiment yearly mean bids for the most comparable situations were $82.20
[$9.10] and $57,00 [$4,631. If one considers that the Randall, et al. (1974a)
figures should be higher as respondents are also bidding on soil banks and
transmission lines, these figures are comparable.
The overall mean for situation (A) (good visibility) to (C) (poor visi-
bility) in the Lake Powell Experiment, [Brookshire, et al. (1976)], was $2.77
[$.19] per day. Adjusted for the 6.6% inflation between the time periods of
the studies, these values become $2.95 [$.20]. The overall mean for recrea-
tionists for the comparable situation in the Farmington Experiment was $4.06
[$1,111, which is considerably different. However, the mean bid was $2.44
[$1.23] when $1.00 starting bids were used in the Farmington Experiment, which
corresponds to the Lake Powell starting bid. Thus, while still statistically
different, for the same starting bids, the results are much closer. The
Farmington Experiment, while not designed as a replication, demonstrated
reasonable consistency with other studies. Finally, a comparison of values
for similar subsamples between the Four Corners and the Lake Powell
Experiments, respectively of $1.79 [$.19] and $1.52 [$.29], also suggest
consistency.
15
The Geothermal Experiment
The Jemez Mountains of New Mexico are both scenic-- characterized by
colored rock outcropping and forest areas--and a major recreation resource
with fishing, campgrounds, hiking trails, and hot springs all located on U.S.
Forest Service lands. However, the Jemez Mountains also contain one of the
major geothermal resources in the Southwest. Geothermal leases have been let
by the U.S. Forest Service on land which is now used solely by recreators.
Both a contingent bidding and a contingent site substitution approach
were used to estimate environmental damages to recreators from possible geo-
thermal development [Thayer (forthcoming)]. Recreators were shown both photo-
graphs of geothermal development in similar mountainous terrain and a map of
the location of possible development relative to recreation areas. Noise
levels and emission characteristics were described in detail. A bidding game
was then conducted using a uniform entrance fee as the vehicle to prevent
development. Additionally, respondents were asked to indicate what their con-
tingent recreation plan would be (what sites would they visit including new
21
-------
substitute sites and how often) if development were to occur. The subsample
which responded to the site substitution question, was then also asked what
they would bid in the form of a uniform entrance fee to prevent development.
Finally, starting point for the bidding game was varied from $1.00 to $10.00
in various sub samples. Thus, the study was structured to test: (1) if con-
tingent bidding and site substitution results were consistent; (2) if infor-
mation on alternative new substitute sites would affect bidding results; and
(3) for starting point bias.
A set of theoretical models were constructed to estimate a consistent
measure of willingness to pay to prevent development from two measures: (1)
the contingent valuation bidding and; (2) additional travel costs associated
with alternative recreation plans. This was an attempt at a methodological
cross check.
The interviews were conducted randomly amongst recreators in the Jemez
area. It is not known if this resulted in sampling bias. A simple distribu-
tional analysis of the data indicated no interviewer bias.
More surprising, however, were the results for information and starting
point bias experiments. Neither bias was statistically significant. The
obvious question is: Why are these results different from those of the
Farmington Experiment, which indicated that both information and starting
point would likely be serious problems? The best explanation that can be
given at this point is that the value of the change in environmental quality
proposed in the two studies was more precisely perceived by respondents in the
Geothermal Experiment than in the Farmington Experiment. In other words,
respondents could more easily relate the costs to themselves of "losing," in
part a recreation area than they could determine the costs of a change in
visibility.
The results of the experiment were as follows: thirty-two percent of the
respondents indicated they would no longer visit the Jemez area if development
occurred. This resulted in about a 40% contingent decrease in visitation.
About 65% of the respondents indicated they would visit alternative sites more
frequently, usually the Pecos Forest area. Bids averaged $2.54 per visitor
party day while the site substitution measure yielded a range of $1,852.59
depending on the assumed driving cost per mile. The results appear to be
consistent for the two approaches and imply an annualized aggregate bid to
prevent construction of about $300,000 for a 50 megawatt plant.
16
The Wildlife Experiment
Through contingent bidding and site substitution approaches, this study
22
-------
attempted to develop a methodology for valuing wildlife experiences. The
valuations were developed to enable policymakers to judge which sites should
be reserved from energy developments so that energy development would not ser-
iously impinge on wildlife. Hunters and wildlife observers were queried as to
their willingness to pay for "encounters" with various types of wildlife.
Encounters was chosen as the variable of perturbation. The hypothesis was
tirat itrinee animals' sighted the greater the satisfaction from the hunting
experience. The species examined were elk, cottontail, coyote, grizzly
bear, bighorn sheep, trout, dipper, and brown creeper. The assumed utility
function had as arguments the number of encounters and length of activity.
Thus, the study attempted to measure both the left and right-hand-side compo-
nents of equations (4) and (5) . Prices for purchase of private goods for the
hunting, fishing, or observation experience were presumed to be constant,
which appears, except for inflationary factors, to be a reasonable assumption.
A wide variety of surveys were tested utilizing alternative formats and struc-
tural components.
A type of instrument bias was observed in that bids were recorded through
license fees, access fees, and utility bill adjustments, but difficulties were
encountered in convincing some respondents that competition between energy
development and wildlife herds would be sufficient reason for utility bill
adjustments. Starting point bias was tested for, but was not found to
substantially affect the bids on species commonly hunted. Thus, this
experiment appears to substantiate the comparison between the Geothermal and
Farmington Experiments which led us to propose that the more clearly identi-
fied the change in the environmental attribute is, the lower the probability
of starting point bias.
Sampling was carried out in Laramie, Wyoming, drawing from hunting and
fishing license lists provided by the Wyoming Game and Fish Department. Add-
resses were drawn randomly from the lists. Refusals by individuals to
actually participate in all parts of the study was about 9%.
Interviewer bias was not present at the .05 level of significance. Non-
response rates to individual bidding games where an individual permitted an
interview but refused to play a particular bidding game under the stated rules
ranged from 2% for willingness to pay games to 30%' for some willingness to
accept compensation games.
Results indicate that, for elk, the average willingness to pay equivalent
surplus measure is $54.00 per year to increase expected encounters (i.e. ,
sightings) from 1 to 5 per day for elk hunting in Wyoming. The average will-
ingness to accept compensating surplus measure for a reduction of 5 to 1 en-
counters per day of elk was $142.00. Some private clubs which specialize in
23
-------
elk hunting in Wyoming charge entrance fees ranging from $85.00 to $150.00 per
year, roughly in the range of the compensating surplus measure for elk en-
counters obtained though the contingent valuation approach.
Before turning to the last case study, we would like to discuss an issue
that has arisen in empirically implementing contingent bidding games. The
issue that continually arises is the observed differences between willingness
to pay (WTP) and willingness to accept (WTA) measures of welfare change.
Willig (1976) derived^gonditions that suggest upper and lower bounds exist
between the measures. However, Gordon and Knetsch (1979) suggest that WTP
and WTA differentials are, in fact, substantial. Empirically to date the
results have been mixed. In the Four Corners Experiment [Randall, et al.
(1976b)] it was noted that "the number of 1 infinity' responses is striking"
and that WTA answers "generally exceeded the willingness of respondents to pay
for environmental improvement." It is suggested that this was not indicative
that no amount of compensation was sufficient, but that abatement by the
energy industry might be preferred. The Lake Powell Experiment [Brookshire et
ai. (1976)] derived a WTA measure from WTP responses and found the measures to
be close. The Farmington Experiment [Blank, etai. (1977) and Rowe, et al.
(1980)] again directly asked compensating measures, finding the WTP and WTA
measures statistically different. Over 50% of the respondents in the Farming-
ton study either refused to cooperate or bid infinity. Finally, the Wildlife
Experiment [Brookshire, et al. (1976) and Brookshire, et al. (1980)] utilized
different formats for obtaining WTP and WTA measures of consumer surplus.
Again the results were statistically different. However, when the WTA
measures were derived, similar in context to Brookshire, et al. (1976) the
values were statistically the same.
What conclusions and explanations can be given for the above results?
Differences between a WTP and WTA welfare measures potentially could be due to
income constraint consideration, differing property rights structures, failure
of the respondent to relate and be able to respond to the contingent market
presented, and/or protest "votes" based on ethical considerations. To date,
we know of no experiment that has been performed to attempt an explanation or
identify which of the above reasons might be correct.
19
The South Coast Air Basin Experiment
In some Los Angeles neighborhoods, deterioration in air quality has been
slight, e.g. , communities adjacent to the Pacific Ocean, while in others, the
deterioration has been relatively severe, as measured by concentrations of NO^
or total oxidants.
The previous case studies reported here, while internally consistent,
24
-------
failed to provide a methodological cross check to actual market data. Thi s
experiment, in contract, attempted to compare both the left-hand-side (contin-
gent bid) and right-hand-side (hedonic measure of equation (8)). Thus, both a
traditional property value study and a contingent valuation were conducted in
an attempt to determine if people will actually pay (as exhibited by property
values) what they say they are willing to pay. Finally, site substitution
information pertaining to activities, location, duration, frequency and ex-
penditures was collected as well.
A shortcoming of the visibility case studies discussed earlier was the
potential confounding between health effects of air pollution and aesthetic
effects. The contingent bidding and substitution approaches employed in this
experiment attempted to value each of these components separately. Aesthetic
considerations were represented by alternative levels of visibility, acute
health effects by eye irritation and chronic health effects by reduction in
life span. Additionally, the population of the South Coast Air Basin has be-
come well informed through the years of the causes of air deterioration, the
potential effects, and scope of the problem. Thus in valuing the non-market
good , "air quality," the experiment was conducted with reasonably well devel-
oped market information for individuals.
In order to insure comparability of results and aid in aggregation, six
pairs of neighborhoods were selected at the census tract level. The pairings
were made on the basis of similarities of housing characteristics, socioeco-
nomic factors, distance to beach and services, average temperature, and sub-
jective indicators of the "quality" of housing. Thus, for each of the six
pairs, an attempt was made to exclude effects on property values of factors
other than differences in air quality. Each of the methodologies were imple-
mented in the paired areas. The bidding game was conducted by randomly
choosing homes within the paired areas. The air quality levels for the paired
areas were determined using monitoring station data in the South Coast Air
Basin. Focusing on total oxidants, nitrogen dioxide and total suspended par-
ticulates, isopleths were constructed for each pollutant. This allowed
"good, " "fair," and "poor" air quality regions to be designated for purposes
of the experiment.
Th|^ata for the property value study, obtained from the Market Data
Center, pertained to 719 homes sold in the 12 paired communities from
January, 1977 to March, 1978 (note the interviewing was conducted during the
latter part of this time interval) and contains information on most important
structural and/or quality attributes. Thus , the data was micro level in
detail and yielded valuation estimates at the household level. The property
value analysis encompassed three separate, and increasingly complex
approaches. First, a comparison of average housing values in the sample paired
25
-------
communities, standardizing only for living space was conducted. Second, a
linear relationship between a home's sale price and its supply of housing and
community attributes was estimated. The value of an improvement in air
quality was then deduced from the resulting hedonic housing value equation.
Third, following Harrison and Rubinfeld (1978) , a hedonic housing equation
allowing for nonlinearities was estimated from which the willingness to pay
equation, as a function of income and other household variables, was again
estimated. This last procedure partly overcomes some of the strict
assumptions of the more simplistic approaches such as identical preferences of
all individuals.
The contingent bidding and site substitution data of the experiment were
collected via a survey questionnaire. The survey questionnaire yielded
valuations by individual for aesthetic and health effects. The survey ques-
tionnaire was designed to test for strategic, information and starting point
biases. The postulated change in air quality was represented both through
regional maps showing good, bad and fair air quality areas as-well as by
photographs showing typical visibility levels. Two specific forms of infor-
mation bias were investigated via a health pamphlet. The health pamphlet
attempted to determine for a subsample of the respondents if detailed in-
formation about health effects would affect bidding and substitution behavior.
Strategic (as in Brookshire, et al. (1976)), information and instrument bias
were not statistically significant influences upon the results. Also
interviewer bias was not present. No records were kept that would enable the
testing for nonrespondent bias.
Accounting for factors such as distance to beach and differences in
preferences, the property value study gave an estimated average bid of $40.00
per month per household for a 30% improvement in air quality. The bidding
results gave an average bid of slightly less than $30.00 per month. Thus ,
reasonable comparability was obtained between the survey and property value
estimates. Given various assumptions of location, income, aggregation by
areas, specific housing characteristics and knowledge on health effects of air
pollution, both the bidding game and property value studies yielded estimates
ranging from $20.00 to $150.00 per month per household for a 30% reduction in
air pollution. These results indicate that air quality deterioration in the
South Coast Air Basin has had substantial effects on housing prices and that
these negative price effects on housing are comparable in magnitude to what
people say they are willing to pay for improved air quality.
CONCLUSION
The six case studies summarized above have shown some consistency in
results and hopefully further the evaluation of problems in structuring
26
-------
contingent market experiments.
Table 2 presents a brief summary of the characteristics of each experi-
men t. The range of environmental attributes valued is quite large--including
visibility, wildlife, health and noise. Four out of six attempted some in-
ternal methodological cross check, however, only the South Coast Air Basin
Experiment utilized an observed set of market prices for the comparison.
Biases do not appear to be an overriding problem. Strategic bias was not
observed in any experiment. Vehicle and starting point biases were highly
significant in the Farmington Experiment. Starting point bias was not found
in any other study. Vehicle bias was significant in the Wildlife Experiment.
A probable explanation for these results, which offers advice for future
experiments, is that the linkage within the contingent market between the
environmental attribute, institutional setting and the bidding instrument must
be realistic and be accepted by the respondent or biased results will be
obtained. The studies further indicate the need to establish a precise
contingent market--the "good" must be well defined.
Possibly the most important result of the studies summarized here is the
replication of results utilizing a traditional property value study and a
contingent bidding approach. At least for this first test case, individuals
do appear to provide contingent valuations comparable to what actual market
behavior implies they are willing to pay for an environmental attribute.
Finally, the studies reviewed in this paper are part of what has become
an ongoing research tradition. It is thus worthwhile to place these efforts
in the context of other recent comparable research. First, both the
experimental research reported by Grether and Plott (1979) and that reported
by Smith (1977) supports the general conclusion that strategic bias in re-
vealing consumer preferences is not likely to be a major problem. Second, a
rather different attempt at validation of a survey approach has recently been
conducted by Bishop and Heberleln (1979) . A market for repurchasing hunting
permits was structured in a "bidding context" and the results are compared to
a traditional travel cost methodology. Since no similar efforts have been
undertaken utilizing mail surveys and repurchasing plans, the research is not
directly comparable to that reported here, Bishop, et al. conclude, somewhat
pessimistically, that since their survey approach might overvalue or
undervalue goose hunting permits by as much as 60 percent and 55 percent
respectively, while the travel cost methodology undervalues by 67 percent,
that all of the available techniques show considerable bias and are thus of
limited use. We, rather, take an opposing position, and view these results as
quite encouraging for the following reason: In many cases, decisionmakers
quite simply have absolutely no idea as to the economic value of preserving
environmental quality. All evidence obtained to date suggests that the most
27
-------
Table 2.2
OVERVIEW OF NON-MARKET
VALUATION EXPERIMENTS
Environmental
Methodol-
Instrument Biases
Non-Market Valuation
Studies
Attribute Being
Valued
Location
ogical Cross
Check
Strategic
Bias
Vehicle
Bias
Starting
Point Bias
Information
Bias ,
1.
Four Corners Experiment
(A. Randall, et. a I.,
1974).
Visibility,
spoil banks,
transmission
lines.
Four Corners
Area, South
west.
Mo
N/A"
N/A
N/A
N/A
2.
Lake Powell Experiment
(D. Brookshire, et. al.
1976).
Visibility
Four Corners
Area, South
West.
No
No
N/A
N/A
N/A
3.
Farmington Experiment
(F . Blank, et.al., 1577
and F.o..c e t. a
Visibility
Four Corners
Area, South
Wes t,
Yes
No
Yes1
Yes1
Yes1
4.
Geothermal Experiment
(M. Thayer, et. al.,
forthcoming).
Noise^ L«^<±
Dist u. t b «.»%e e.
Jemez Moun-
tains.
Yes
No
N/A
No"
No'
5.
Rocky Mountain Wild-
life Experiment (3.
Brccksr.i re, e t. a 1.,
i 5 7 7 and forthcoming).
Encoun' ers
with w Idl ife
Wyoming
Yes
No
Yes'
No'
N/A
6.
South Coast Air Basin
Experiment, (D. Brook-
shire, et. al., 1530),
Visibi itv.
health Effects
Los Angeles
Region,
California
Yes
No
No"
No9
No"
'uoc Available - The experiment did notconsider either structurally or analytically this form of bias.
bias tests were defined in Brookshi re,et.al. (1976).
cut i li ring estimated bid curves the t rat ios for these variables were respectively (3.05), (7 98) and (-4.54) where the vehicle
variable was O "utility bill, 1= payroll deduction; starting bid variable was either $ I, $5, or $10 anc information variable
j vias 0 " no prior information, 1 - prior information. See Rowe, et.al. (ijSO).
Utilizing an estimated bid curve the t ratio was .689 on the start i ng po'i nt variable Indicat ng no significant influence.
elnfor.;at ion bias in this study pertained to whether tha suggestion of alternative recreation cities would influence the bid.
c A 5 tandard r-tes t was ut i I ized with no statistical influence beinjg observed.
'a T-test was conducted where the hypothesis that the final value data was influenced by the bidding veh cle (starting bids) was
rejected.
'A T-test was conducted whereby the acceptance of the hypothesis that the mean bids for all paired areas combined for different
bidding vehicles (starting points) are equal implies and higher.
Information bias in this study related to alternative sequencing of health and aesthetic information, he test was as in
footnote e.
-------
readily applicable methodologies for evaluating environmental quality--hedonic
studies of property values or wages, travel cost and survey techniques--all
yield values good to well within one order of magnitude in accuracy. Such
information, in our view, is preferable to complete ignorance.
29
-------
REFERENCES
1. Maler (1974) has classified the possibilities for measurement of environ-
mental goods or services into four broad categories: (1) asking indi-
viduals what they are willing to pay; (2) voting on the supply; (3) in-
direct methods based on observations on the relationship between private
goods purchased and environmental goods; and (4) estimation of physical
damage and evaluation on the basis of observed market prices. in this
paper, we analyze methods only within Maler's categories (1) and (3).
2. In the recent literature, one approach within this set has been called
"bidding games," [Randall, et al. (1974a); Brookshire, et al. (1976)].
However, because some types of responses are not bids but changes in
behavior, e.g., site substitutions or minimum compensation, we prefer the
more general "contingent valuations" to identify the set of approaches
that directly query the individual for information in a series of hypo-
thetical situations on markets.
3. An alternative listing of explanations for bias and other problems is
given in D. Grether and C. Plott, (1979) .
4. A detailed form for the utility function compatible with our arguments is
A), G2^2'A2i* ' ' *;X'K The G-j0311 be concave increasing
functions for each" activity and imply the utility function is weakly
separable over locations or activities.
5. This is equivalent to the compensating variation measure of consumer
surplus where the initial level of utility is maintained. See, for
example, Mishan (1971) .
6. See Brookshire and Crocker [forthcoming (b)] for a discussion of the role
of information in contingent markets and validity of the consumers
response.
7. One survey of air pollution in the late 1960's for Los Angeles which we
prefer not to cite asked the question "How much are you willing to pay
for less air pollution?" Clearly, this question is too vague and subject
30
-------
to multiple interpretations as to the change in environmental attribute.
Alternatively, a question "How much are you willing to pay for an annual
average reduction in oxidant concentrations o'f .10 parts per million in
the seven block radius around Hollywood Boulevard and Vine Street?" may
be too specific and not readily understandable by the interviewee. There
appears to be a fine line where the general public can fully understand
the question posed, yet the question is precise enough to be of scienti-
fic usefulness, i.e., be relatable to scientific measures of environ-
mental change.
8. See Brookshire and Crocker [forthcoming (b) ] for further discussion.
9. We present this extremely brief summary of Randall's work noting it was
the first effort, and to set the stage and focus the discussion for the
remaining case studies. See Randall et al. (1974a,b) for a complete
discussion of the results.
10. This research was funded by the NSF-RANN Lake Powell Research Project.
11. The average bid concept was introduced in the survey instrument in the
following manner; "Let's also assume that all visitors to the area will
pay the same daily fee as you . . . ." The use of the terms "environ-
mental" and "developers" is to distinguish two groups who might have
widely divergent preferences with respect to environmental commodities.
12. This study was supported by the Electric Power Research Institute (EPRI),
Palo Alto, California to the University of Wyoming. EPRI does not assume
any liability for the completeness of research, or usefulness of the
results.
13. For individuals to bid strategically to achieve a specific outcome when
the respondent knows everyone must pay the final bid is extremely diffi-
cult. For instance, all previous bids by others must be known, the
sample size and if the individual is not the "last" bidder, then future
bids must be known. For more discussion see Brookshire and Eubanks
[forthcoming (a)].
14. Standard errors in brackets. 15. The research reported here was
supported by a NSF grant entitled "An Economic and Environmental Analysis
of Solar and Geothermal Energy Sources."
16. Portions of this study were funded by the U.S. Fish and Wildlife Service
contract numbers 14-16-0009-77-022 and 14-16-0009-77-003 with the
University of Wyoming and parts were sub-contracted to the University of
31
-------
Kentucky.
17. For a complete discussion of the study see Brookshire, et al. (1977) and
Brookshire, et al. (1980).
18. Randall and Stoll (1979) have reformulated Willig's results from price to
surplus space."
19. This study was supported by the U.S. Environmental Protection Agency
EPA-600/6-79-0001b.
20. The Market Data Center is a computerized appraisal service centered in
Los Angeles, California.
32
-------
BIBLIOGRAPHY
Bishop, R. C. and T. A. Heberlein. 1979. "Measuring Values of Extra-Market
Goods : Are Indirect Measures Biased?" American Journal of Agricultural
Economics, Vol. 61, No. 5 (December): 926-930.
Blank, F., D. Brookshire, T. Crocker, R. d'Arge, R. Horst, Jr. and R. Rowe.
1977. Valuation of Aesthetic Preferences: A Case Study of the Economic
Value of Visibility, Research Report to the Electric Power Research
Institute, Resource and Environmental Economics Laboratory, University
of Wyoming.
Bohra, P. 1971. "An Approach to the Problem of Estimating Demand for Public
Goods." Swedish Journal of Economics 3, No. 2: 111-130.
Bradford, D. 1970. "Benefit Cost Analysis and Demand Curves for Public Goods."
Kyklos 23, >°» 4: 775-791.
Brookshire, D., B. Ives and W. Schulze. 1976. "The Valuation of Aesthetic
Preferences ." Journal of Environmental Economics and Management 3, No. 4
(December): 325-346.
Brookshire, D., A. Randall, et al. 1977. Methodological Experiments in
Valuing Wildlife Resources: Phase I Intern Report to the United States
Fish and Wildlife Service.
Brookshire, D., A. Randall and J. Stoll. 1980. "Valuing Increments and Decre-
ments in Natural Resource Service Flows." American Journal of Agricultu-
ral Economics 62, No. 3 (August): 478-488.
Brookshire, D. and L. Eubanks. "Contingent Valuation and Revealing the
Actual Demand for Public Environmental Commodities." Public Choice
[forthcoming (a)].
Brookshire, D. and T. Crocker. "The Advantages of Contingent Valuation Methods
for Benefit-Cost Analysis." Public Choice [forthcoming (b)].
33
-------
Brookshire, D., R. d'Arge, W. Schulze and M. Thayer. "Experiments in Valuing
Public Goods." Advances in Applied Macroeconomics, ed. V. Kerry Smith,
JAI Press Inc. [forthcoming (c)].
Cummings, R. G., W. Schulze and A. F. Mehr. 1978. "Optimal Municipal Invest-
ment in Boontowns: An Empirical Analysis." Journal of Environmental
Economics and*Management 5, No. 3 (September): 252-267.
Davis, R.' 1963. "Recreation Planning as an Economic Problem." Natural Resources
Journal 3 (May): 239-249.
Freeman III, A. Myrick. 1979a. "Hedonic Prices, Property Values and Measuring
Environmental Benefits: A Survey of the Issues." Scandinavian Journal
of Economics 81, No. 2: 154-173.
Freeman III, A. Myrick, 1979b. The Benefits of Environmental Improvement:
Theory and Practice, Baltimore: Johns Hopkins Press.
Grether, D. M. and C. R. Plott. 1979. "Economic Theory of Choice and the
Preference Reversal Phenomenon." American Economic Review, Vol. 69,
No. 4 (September): 623-638.
Gorden, 1. M. and J. L. Knetsch. 1979. "Consumer's Surplus Measures and the
Evaluation of Resources." Land Economics, Vol. 55, No. 1 (February).
Gramlich, F. W. 1977. "The Demand for Clear Water: The Case of the Charles
River." National Tax Journal 30, No. 2 (June): 183-194.
Hammack, J. and G. Brown. 1974. Waterfowl and Wetlands: Toward Bioeconomic
Analysis, Baltimore: Johns Hopkins University Press.
Harrison, D., Jr., D. L. Rubitifeld. 1978. "Hedonic Housing Prices and the
Demand for Clean Air." Journal of Environmental Economics and Management
Vol. 5, No. 1 (March): 81-102.
Hori, H. 1975. "Revealed Preferences for Public Goods." American Economic
Review, Vol. 65, No. 5 (December): 978-991.
Jones-Lee, M. W. 1976. The Value of Life. Chicago: The University of Chicago
Press.
Lancaster, K. 1966. "A New Approach to Consumer Theory." Journal of Political
Economy, Vol. 74, No. 2 (April): 132-57.
34
-------
Loehman, E., S. Berg, A. Arroyo, R. Hedinger, J. Schwartz, M. Stiaw, R. Fahien,
V. De, R. Fishe, D. Rio, W. Rossley, and A. Green. 1979. "Distributional
Analysis of Regional Benefits and Costs of Air Quality Control." Journal
of Environmental Economics and Management, Vol. 6, No. 3 (September):
222-243.
Maler, K. G. 1974. 'Environmental Economics: A Theoretical Inquiry, The Johns
Hopkins University Press.
Mishan, E. J. 1971. Cost-Benefit Analysis: An Introduction, New York:
Praeger Publications.
Muellbauer, J. 1974. "Household Production Theory, Quality, and the ^Hedonic
Technique'." American Economic Review, Vol. 64, No. 6 (December):
977-994.
O'Hanlon, P. W. and J. A. Sinden. 1978. "Scope for Valuation of Environmental
Goods: Comment." Land Economics, Vol. 54, No. 3 (August): 381-387.
Randall, A., B. Ives and C. Eastman. 1974a. "Bidding Games for Valuation of
Aesthetic Environmental Improvements." Journal of Environmental Economics
and Management, Vol. 1, No. 2 (August): 132-149.
Randall, A., B. Ives and C. Eastman. 1974b, "Benefits of Abating Aesthetic
Environmental Damage from the Four Corners Power Plant, Fruitland, New
Mexico." New Mexico State University Agricultural Experiment Station
Bulletin 618.
Randall, A. and J. Stoll. 1980. "Consumer's Surplus in Commodity Space."
American Economic Review, 70 (June): 449-455.
Randall, A., 0. Grtmewald, S. Johnson, R. Ausness and A. Pagoulatos, 1978.
"Reclaiming Coal Surface Mines in Central Appalachia: A Case Study of
the Benefits and Costs." Land Economics 54, No. 4 (November): 472-489.
Rosen, S. 1974. "Hedonic Prices and Implicit Markets: Product Differentiation
in Pure Competition." Journal of Political Economy , Vol. 82, No. 1
(January/February): 34-55.
Rowe, R., R. d'Arge, D. Brookshire. 1980. "An Experiment on the Economic
Value of Visibility." Journal of Environmental Economics and Management,
Vol. 7, No. 1 (March): 1-19.
Samuelson, P. A. 1954. "The Pure Theory of Public Expenditures." Review of
35
-------
Economics and Statistics, Vol. 36, No.4 (November): 387-389.
Scherr, B. A. and E. M. Babb. 1975. "Pricing Public Goods: An Experiment with
Two Proposed Pricing Systems." Public Choice 23 (Fall): 35-48.
Sinden, J. A. and J. B. Wyckoff. 1976. "Indifference Mapping: An Empirical
Methodology for Economic Evaluation of the Environment." Regional
Science and Urban Economics 6, No. 1 (March): 81-103.
Smith, V. K. 1977. "The Principle of Unanimity and Voluntary Consent in Social
Choice." Journal of Political Economy 85, No. 6 (December): 1125-1139.
Strauss, R. P. and G. D. Hughes. 1976. "A New Approach to the Demand for
Public Goods." Journal of Public Economics, Vol. 6, No. 3 (October):
191-204.
Thayer, M. "Contingent Valuation Techniques for Assessing Environmental Impact:
Further Evidence." Journal of Environmental Economics and Management,
(forthcoming).
Willig, R. D. 1976. "Consumer's Surplus without Apology." American Economic
Review, Vol. 66, No. 4 (September): 589-597.
36
-------
CHAPTER 3
VALUING PUBLIC GOODS: A COMPARISON OF SURVEY AND HEDONIC APPROACHES
INTRODUCTION
Although the theory of public goods has progressed rapidly since
Samuelsoti's seminal article (1954) , the empirical measurement of the value of
(demand for) public goods only recently has received increased attention.
Perhaps the best known and most widely accepted empirical approach has been
the use of hedonic prices wherein, for example, it is assumed that either
wages or housing values reflect spatial variation in public good characteris-
tics of different communities. This indirect approach, based on theoretical
work of Tiebout (1956), Lancaster (1966) , Rosen (1974) and others has proven
quite successful. Among public goods or bads which have been valued using the
hedonic approach are climate [Hoch (1974)], air pollution [Anderson and
Crocker (1971) and Harrison and Rubinfeld (1978)], social infrastructure
[Cummings, et al. (1978)] and other community characteristics such as noise
level [Nelson (1979)] and ethnic composition [Schnare (1976)].
An alternative approach is to directly ask households or individuals to
state their willingness to pay for public goods using survey techniques.
Despite arguments that strategic bias will invalidate survey results, there
exists the need for an alternative to the hedonic approach. As an example,
consider the case of a remote and unique scenic vista, valuable to recreators,
which is threatened by air pollution from a proposed coal fired plant--a
typical situation in the Western United States. Although it is possible, in
principle, to impute the value of clean air and visibility from the relative
decline in local visitation which might follow construction of a power plant,
information on the value of visibility at the site is needed prior to con-
struction for socially optimal decisionmaking on plant location and pollution
control equipment. The hedonic approach is unavailable both because the
scarcity of local population--as opposed to recreators--makes use of wage or
property value data impossible and because scenic vistas may themselves be
unique. For these reasons, Randall et al. (1974) first applied survey methods
for valuing visibility and other environmental effects of large coal fired
37
-------
power plants in the Four Corners region of New Mexico. Since this initial
application, the survey approach has been widely used to value environmental
commodities where market data for hedonic analysis is difficult to acquire
[see, for example, Brookshire, Ives and Schulze (1976), Rowe, et al. (1980),
and Brookshire, et al. (1980)]. Other early attempts to value public goods
using the survey approach include Davis (1963) , Bohm (1972) and Hammack and
Brown (1974) . 1
Although results of using the survey approach for estimating the value of
public goods appear to be internally consistent, replicable and consistent
with demand theory [see Schulze et al. (forthcoming)], no external validation
has been reported (i.e., a comparative analysis using another approach
independent of the survey has not been conducted) . Thus , the purpose of this
paper is to report on an experiment designed to validate the survey approach
by direct comparison to a hedonic property value study.
The Los Angeles metropolitan area was chosen for the experiment because
of the well defined air pollution problem and because of the existence of
detailed property value data. Twelve census tracts were chosen for sampling
wherein 290 household interviews were conducted during March, 1978. Respon-
dents were asked to provide their willingness to pay for an improvement in air
quality at their current location. Air quality was defined as poor, fair, or
good based both on maps of the region (the pollution gradient across the Los
Angeles Metropolitan Area is both well defined and well understood by local
residents) and on photographs of a distant vista representative of the
differing air quality levels. Households in poor air quality areas were asked
to value an improvement to fair air quality while those in fair areas were
asked to value an improvement to good air quality. Households in good air
quality areas were asked their willingness to pay for a region-wide im-
provement in air quality. The region-wide responses are reported elsewhere
[Brookshire, et al. (1980)].
For comparison to the survey responses, data was obtained on 634 single
family home sales which occurred between January, 1977 and March, 1978 ex-
clusively in the twelve communities used for the survey analysis. As we show
in the next section, households, in theory, will choose to locate along a
pollution-rent gradient, paying more for homes in clean air areas based on
income and tastes. However, ceteris paribus, we show that the annualized cost
difference between homes in two different air quality areas (the rent
differential for pollution) will in theory exceed the annual willingness to
pay for an equivalent improvement in air quality for a household in the lower
air quality area. Thus, the rent differential associated with air quality
improvement from hedonic analysis of the property value data must exceed es-
timates of household willingness to pay for the survey responses, if the
38
-------
survey responses are a valid measure of the value of air quality improvements.
Section 3 describes the data analysis and experimental design in more detail.
We also conjecture that the willingness to pay for air quality improve-
ments is greater than zero for residents in our sample communities based on
statewide political support for air quality regulation. The State of
California, principally in response to the air pollution problem in the Los
Angeles Metropolitan area, has led the nation in imposing automobile emissions
standards. The automobile industry, under pressure from the California
Legislature, installed the first pollution control devises on California cars
in 1961. This initial step was followed nationally in 1963. Again, Califor-
nia imposed the first exhaust-emission control regulations in 1966, leading
the nation by two years. Over the decade of the 1970's, California has had
more stringent automotive emission standards than Federal levels, resulting in
higher initial costs and sacrifices in both performance and fuel economy. In
spite of these difficulties, political support, as reflected both in the State
Legislature and in several administrations, has remained strong for auto
emission controls.
In Section 4 the results of the hypotheses tests are presented. As Table
2 illustrates, results of the experiment can be summarized as follows: In the
nine census tracts where air quality improvements are possible (poor and fair
communities) , we cannot reject our dual hypotheses that, in each census tract,
household willingness to pay for air quality improvements, as estimated by
surveying households, falls below equivalent property value rent differentials
and lies above zero. We view these results as a qualified verification of the
survey approach for estimating the value of public goods. Further
interpretation of the results is contained in the concluding remarks offered
in Section 5.
A THEORETICAL BASIS
The property value and the survey approaches for valuing public goods
have received considerable theoretical scrutiny. Property value studies are
conceptually based on hedonic price theory as developed by Rosen (1974) and
recently summarized by Freeman (1979) . The survey approach has been modeled
using standard concepts of consumer surplus by Randall et al. (1974), Bohm
(1972), and Brookshire et al. (1976) where the latter two analyses also focus
on the possibility of strategic behavior. The considerable empirical evidence
now available suggests that strategic bias may be of little consequence both
in survey work [See Brookshire et al. (1980) and Rowe et al. (1980)] and in
experimental economics [See Grether and piott (1979), Scherr and Babb (1975)
and Smith (1977)]. However, other types of bias may still invalidate a survey
approach for valuing public goods. It has even been suggested that the survey
39
-------
approach produces "noise" since responses are purely hypothetical and have no
necessary connection to actual budgetary decisions.
In this section, a simple theoretical model is developed for comparison
of survey responses to a property value study for valuing air quality im-
provements in the Los Angeles region in order to determine if valid public
good measures can "be obtained from survey data.
We use the following notation:
Let P = the level of air pollution
x = consumption of a composite commodity excluding housing
c = unit cost or price of the composite commodity X
R = rent or periodic cost of housing
Y * household income
and U(P,X) = household utility, a decreasing function of ^ llution Up < 0
an increasing function of consumption U < 0.
Each household maximizes utility, U(P,X), subject to the budget constraint:
Y - GX - R(P) =0
where we assume the existence of a continuous differentiable rent gradient
R(P) . [See Rosen (1974)] for a complete discussion of the generation and
existence of rent gradients. Our model is a simple adaptation of Rosen's, so
we will not elaborate here.) Two distinct choices are modeled: consumption
of the composite commodity, X, and that of housing location by pollution
level, P. Presuma|ly, lower rents will be paid for homes in more polluted
areas, so R'(P)<0. The first order conditions for choice of P and X imply
that
C J? = R' (P)
U
X
or that the marginal rate of substitution between pollution, P, and the
composite commodity, X, valued at the cost of the composite commodity, C,
equals the slope of the rent gradient R'(P) at equilibrium location and
consumption levels.
Figure 1 illustrates the solution graphically and allows us to structure
hypotheses for testing the validity of survey results in comparison to the
property value approach. The vertical axis measures the quantity of the
composite commodity, X, where we assume that the cost, C, of the composite
40
-------
Figure 3.1
Composite
Commodity
y'-R(P)
R(P)
AR
AP
Air Pollution
With identical housing attributes the identical rent differential, AR, exceeds
individual willingness to pay, WA and wB.
41
-------
commodity is unity; i.e., the vertical axis measures dollars as well.
Pollution is on the horizontal axis. Given household income Y°, the budget
constraint, shown as Y" - R(P) in Figure 1, is obtained by vertically sub-
tracting the rent gradient, R(P). Thus, household A with preferences shown by
indifference curve 1° would maximize utility at point "a", choosing to locate
at pollution level P° , consume X° and pay rent R°. If household A's intome
were to increase to Y , the budget constraint would shift vertically to r -
R(P) and the same^household would relocate, goosing point "b", at a lower
pollution level" P" w th" higher consumption, X , given tastes as represented by
indifference curve T ^ ^lternat;>-vely' another household, B, with income y",
but tastes as shown by I w^uld choose point "d", locating at P as well, but
choosing lower consumption X . Thus, both tastes and income enter location
decisions over pollution levels.
The survey approach used in the Los Angeles metropolitan area to obtain
an estimate of the value of air quality asked households how much, at most,
they would be willing to pay for an improvement in air quality at the site
where they presently live. Thus , the household in equilibrium at point "a" in
Figure 1 was asked how much X it would forego to experience"! rather than P"
while maintaining the same utility level. Presumably, household A would be
A
indifferent between points "a" and "c" and be willing to pay W dollars (or
units of X) to achieve a reduction in air pollution of AP. Unfortunately, as
is illustrated in Figure 1, the budget constraint, Y" - R(P), obtainable by
estimating the rent gradient function, R(P), does not provide information on
the bid for improved air quality, tf^. Rather, the change in rent between
locations with air quality levels P" arid~P ," AR in Figure 1, must, for any
household located at "a", equal or exceed the bid, w , if the second order
conditions for the household optimization problem are generally satisfied.
Thus, we can establish an upper bound on the willingness to pay for air
quality improvement by |xaminixxg the rent gradient. For example, if household
B had a lower income, Y , it would locate at point "e". Even though household
B is now located at pollution level P" like household A, its bid for an air
quality improvement AP would be W^, smaller than_J^f yet still less than AR.
Thus , if survey bids are a valid measure of willingness to pay for air quality
improvements then AR > W.
This hypothesis holds for each household even if we consider the case of
multiple housing attributes. Including other attributes such as square
footage of the home, bathrooms, fireplaces, neighborhood characteristics,
etc., denoted by the vector Z, the model is revised as follows:
max U(Z, P, x)
St. Y - Cx - R (Z, P) - 0
42
-------
3
with first order conditions
cup= R (z,p)
and
C X " Rt(Z,P).
°r
These first order conditions constitute, along with frequency distributions
for housing characteristics and household j>ref|rences, a system of partial
differential equations which solve for R(Z,F). Thus, a hedonic rent gradient
is defined for pollution, P, and other household characteristics, Z, as well.
As is illustrated in Figure 1, in which housing characteristics other
than pollution are not incorporated, budget constraints for different house-
holds are obtained by vertically shifting the same rent gradient. Thus, all
households face the same rent differential AR for a change in pollution level
AP even though willingness to pay for that change may differ, i.e., # WB.
However, turning to Figure 2, household A, located at P°, may occupy a house
with attributes-^ while household Bgalso located at P° may occupyAa house
with a different set of attributes Z . Household A, with income Y. ^ would
then face a rent gradient like that shown in Figure 2 defined by R(Z , P) and
choose point "a
'' b"S household B with income Y , would now face a different
rent gradient of R(Z , P) and choose to locate at point "b". Therefore,
households, with different housing characteristics may face different rent
gradients over pollution when projected in the (X, P) plane. In general, AR,
unlike the case shown in Figure 1, will no longer be constant across house-
holds at the same location. However, for each household i (i = A, B in Figure
2), it is still true that the rent differential, AR , for a change in
pollution AP, calculated for the fixed vector of housing characteristics ZT ,
will exceed that household's willingness to pay, W , for the same change in
pollution level at the same location. Note that households were asked their
willingness to pay with the specific assumption that they remained in the same
house and location. Thus, Z , for a particular household was truly fixed -
allowing the simple analysis in the (X,P) plane as shown in Figure 2.
The first hypothesis for testing the validity of the survey approach can
be constructed as follows: for each household i in a community, AR" >$ . It
then follows that in each community the average rent differential across
households, AR, must equal or exceed the average willingness to pay W for an
improvement in air quality. In other words, if survey bids are a valid mea-
sure of willingness to pay, then for each community in our sample, AR>W, i.e.,
average willingness to pay cannot exceed the average rent differential. Our
second hypothesis is that, given the political history of air pollution
control in the State of California as described in the introduction, mean bids
43
-------
Figure 3.2
Composite
Commodity
X
AR
AR
B
LA
a
yA-R(Z*P)
/i
/®/^y®-R(z?P)
/ i
/ i
/ 1 ^
r"/" ~ ¦" "*/• ™ ¦" 1 1
f WB{]V
— i /
i/
v. — u
0
p- p-
AP
Air Pollution
With differing housing attributes across households each individual rent differential
exceeds that households willingness to pay.
44
-------
in each community are nonnegative, W > 0.
Our dual test of the validity of survey measures must remain somewhat
imprecise because hedonic rent gradients themselves only provide point
estimates of the marginal rates of substitution (slopes of indifference
tunes) between pollution and other goods (money) for individuals with pos-
sible differing tastes and income. One does not have information necessary to
estimate, for example, the shape of 1° in Figure 1 solely on the basis of the
slope of the budget constraint, R.' (P°)» at point "a". Attempts to estimate
individual willingness to pay (w in Figure 1) from hedonic rent gradients
must thus introduce strong assumptions about the nature of preferences. (See,
for an example of an hedonic approach which derives willingness to pay by
making such assumptions, Harrison and Rubinfeld [1978] .
SAMPLING AND DATA ANALYSIS
The previous section has presented a theoretical framework for a com-
parison between the survey technique and the property value approach for
valuing public goods. In order to empirically implement the comparison, the
two approaches require a consistent sampling procedure. This section de-
scribes the sampling procedure and results of the separate studies.
Sampling was restricted to households within the Los Angeles metropolitan
area. The first concern was air pollution data. Air monitoring stations are
located throughout the Los Angeles area providing readings on nitrogen dioxide
(NO^), total suspended particulate matter (TSP) and other pollutants. The
objective was to relate as closely as possible the readings of two con-
stituents of air pollution (NO and TSP) to census tracts used both for the
2 . ...
property value and survey studies. The air shed was divided into the follow-
ing air quality ^regipjnu: "good" (NO^ < 9 pphm) (TSP < 90 yg/m ) ; (NO^
911 pphm) (TSP 9110 yg/m ); and "poor" (N02 > 11 pphm) TSP > 110 yg/m 1 .
Improvements from poor to fair and fair to good across the region are each
associated with about a 30% reduction in ambient pollution levels. Consid-
eration was given to wind patterns and topography of the area in making these
distinctions.
Many variables may affect the value households place on air quality. To
control for as many of these as possible in advance of the actual experiment,
the sample plan identified six community pairs where each pair was relatively
homogeneous with respect to socioeconomic, housing and community
5
characteristics, yet allowed for a significant variation in air quality.
The property value analysis attempts to provide external validation for
the survey approach. The absence of such validation explains in our view, the
45
-------
lack of general acceptance of survey techniques. The objective, then, is to
estimate the hedonic rent gradient R(Z, P) and calculate rent differentials
associated with the poor-fair and fair-good air quality improvements for
sample census tracts. These results are then utilized for comparison to the
survey results.
A hedonic rent gradient was estimated in accordance with literature as
recently summarized by Freeman (1979) . Housing sale price is assumed to be a
function of housing structure variables (living area, bathrooms, fireplaces,
etc.), neighborhood variables (crime rate, school quality, population density,
etc.), accessibility variables (distance employment to centers and beach) and
air quality a^ measured by total suspended particulate (TSP) or nitrogen
dioxide (NG^). The primary assumption of the analysis is that variations in
air pollution levels as well as other household, neighborhood and
accessibility attributes are capitalized into home sale price. Implicit or
hedonic prices for each attribute are then determined by examining housing
prices and attribute levels.
The property value analysis was conducted at the household level in order
to provide an appropriate comparison to the survey instrument. Thus, the
household data used were a^ the micro level of aggregation and include a large
number of characteristics. Data was obtained for 634 sales of single family
homes which occurred between January, 1977 and March, 1978 in the communities
used for the survey analysis. In addition to the immediate attributes of the
household, variables which reflected the neighborhood and community were
included to isolate the independent influence of air quality differentials on
home sale price.
As indicated by Maler (1977) even under the presumption of correct model
specification, estimation of a single equation hedonic rent gradient may be
hindered by severe empirical difficulties, primarily multi-collinearity. With
respect to this problem, in each of three data categories--household,
neighborhood, and air quality—multicollinearity forced the exclusion of
variables and the usage of proxy variables. For instance, collinearity
between number of rooms, number of bedrooms and living area as quantitative
measures of house size allowed the use only one--living area which serves as a
proxy for all. Further, since housing density and population density measure
essentially the same phenomenon, only the former is used in the estimated
equations. The estimation procedure was not able to separate out the
independent influence of each air pollutant. Thus, only one pollution mea-
sure, either NO or TSP, was utilized to describe the level of air quality.
In order to provide information concerning the sensitivity of our analysis,
results are presented for each of these pollutants. Finally, contrary to
expectation a collinearity problem did not exist between distance from beach
46
-------
and air pollution. This can be attributed, in part, to the success of the
sample plan in isolating the effects of air quality.
ternatively using N02 or TSP to represent pollution level. A number of as-
pects of the equations are worth noting.
First, approximately 90% of the variation in home sale price is explained
by the variation in the independent variable set. Second, with only a minor
exception, all coefficients possess the expected relationship to the dependent
variable and are statistically significant at the one percent level. The
exception is the crime rate in both the NO and TSP equations. Third, in
their respective equations, the log form of the pollution variables have the
expected negative influence on sale price and are highly significant. The
estimated relationship between house sale price and pollution is therefore
consistent with the graphical analysis of Section 2; that is, the rent
gradient is convex from below in the pollution/dollars plane. Finally, the
stability or relative insensitivity of the regression coefficients to the
particular pollution variable indicates that individuals have an aversion to
pollution in general rather than to any one pollutant.
Estimation of the rent gradient was also completed using other forms of
the pollution variables (linear, squared, cubic). Whereas the squared and
cubic terms did not demonstrate statistical significance, the first order
terms performed only marginally worse than the log formulation. Rent dif-
ferentials have also been calculated for these and other forms with results
nearly identical to those presented here.
The next step was to estimate the rent differential AR. for each indi-
vidual household for each census tract. The rent differential specifies the
premium an individual household would have to pay to obtain an identical home
in the next cleaner air region (poor to fair for six communities, fair to good
for three communities) . Due to the estimated functional form of the rent
gradient, the calculated rent differential is dependent upon the value of all
other variables. The average home sale price change based on individual
data in each census tract associated with an improvement in air quality,
ceteris paribus, is shown in column two of Table 2 of the next section.
Column one of Table 2 lists communities by air quality level. The table only
shows for the log-linear NO equation since, as noted above, other
specifications give nearly identical results. The figures shown are derived
by evaluating the hedonic housing expression, given the household's charac-
teristics, for a pollution change from poor to fair or fair to good as the
case may be. The resulting sale price differential is then converted to an
equivalent monthly payment through the standard annualization procedure and
Two alternative nonlinear specifications
47
-------
Table 3.1
Estimated Hedonic Rent Gradient Equations3
Dependent Variable = Log (Home Sale Price in $1,000)
Independent
N 0, Equation
TSP Equation
Variable
Housing Structure Variables
Sale Date
.018591
.018654
(9.7577)
(9.7727)
Age
-.018171
-.021411
(2.3385)
(-2.8147)
Living Area
.00017568
.00017507
(12.126)
(12.069)
Bathrooms
.15602
.15703
(9.609)
(9.6636)
Pool
.058063
.058397
(4.6301)
(4.6518)'
Fireplaces
.099577
.099927
(7.1 705)
(7.1866)
Neighborhood Variables
Log (Crime)
-.0838 1
-.1 0 4 0 1
(-.5766)
("l .9974)
School Quality
.0019826
.001771
(3.9450)
(3.5769)
Ethnic Composition
.027031
.043472
-(Percent White)
(4.3915)
(6.2583)
Housing Density
-.000066926
-.000067613
(9.1 277)
(-9.2359)
Public Safety Expenditures
.00026192
.00026143
(4.7602)
(4.7418)
Accessibility Variables
Distance to Beach
-.011586
-.011612
(-7.8321)
(7.7822)
Distance to Employment
-.285 1 4
-.26232
(-14.786)
(14.158)
Air Pollution Variables
log (TSP)
-.22183
log (NO )
(-3.8324)
-.22407
c
(4.0324)
Constant
2.2325
1.0527
(2.9296)
(1.4537)
R*
.89
.89
Sum of Squared Residuals
18.92
18.97
Degrees of Freedom
619
619
a
t - Statistics in Parentheses
48
-------
Table 3.2
Tests of Hypotheses
roperty Value Results3
Survey
Results
I-
I
Tests uf Hypotheses
Commun i ty
A R
(Standard
)ev i a t i on)
'lumber of
Observations
W
(Standard
Deviation
Number of
Observations
-statistics
«g > Ob
-statistics
W4R - "WC
Poor - Fair
El Monte
15.44
(2.88)
22
11.10
(13.13)
2 0
3.78
1.51
Montebel lo
30.62
(7.26)
49
11.42
(15.15)
19
3.28
7.07
La Cafiada
73.78
(48.25)
51
22.06
(33.24)
17
2.74
4.10
Sample
Population
45.92
(36.69)
122
14.54
(21.93)
5 6
4.96
5.54
Fair - Good
Canoga Park
33.17
(3.88)
22
16.08
(15.46)
34
~
6.07
5.07
Huntington
Beach
47.26
(10.66)
44
24.34
(25.46)
3 8
5.92.
5.47
Irvine
48.22
(8.90)
196
22.37
(19.13)
27
6.08
5.08
Culver City
54.44
(16.09)
64
28.18
(34.17)
30
5.42
11..85
Encino
128.46
(51 .95)
45
16.51
(13.38)
37
7.51
12.75
Newport
Beach
77.02
(41.25)
22
5.55
(6.83)
20
3.63
7.65
Sample
Population
59.09
(34.28)
393
20.31
(23.0)
186
12.02
14.00
'Rent differentials for the hedonic housinq equation in which Jog (NO?) is the relevant
pollution variable are presented here. Essentially identical results are obtained using
N02, TSP or log (TSP).
"The hypotheses to be tested were Ho : = > ®- All test statistics indicate
rejection Of the null hypothesis at the l'£ significance level.
cThe hypotheses to be tested were Ho : Uj-r- > ; Hj : Vj-r- < All Test statistics indicate
that the null hypothesis could not be rejected even at the I OS significance level.
-------
division by twelve. Since our hypothesis test is posed in terms of the
average rent differential in the relevant communities, then a community mean
and standard deviation are calculated. Column three of Table 2 shows the
number of homes for which data was available to calculate average rent dif-
ferentials and standard deviations for each community. Monthly rent differ-
entials ranged from $15.44 to $45.92 for an improvement from poor to fair air
quality and $33.17" to $128,46 for an improvement from fair to good air qual-
ity. The higher figures in each case are associated with higher income com-
munities. Again, these average differentials should provide an upper bound
for the survey results.
The survey approach followed the work of Davis (1963) and Bohm (1972) in
gathering the information necessary for estimating a Bradford (1972) bid
curve. The approach involves the establishment of a hypothetical market via a
survey instrument. Through the work of Randall, et al., (1974) and
Brookshire, et al., (1976), the necessary structure for constructing a hypo-
thetical market for the direct determination of economic values within the
Hicksian consumer surplus framework has been developed. The survey reported
here is consistent with this previous literature.
The hypothetical market was defined and described both in technical and
institutional detail. The public good (air quality) was described by the
survey instrument to the respondent in terms of ea !lly perceived levels of
provision such as visual range through photographs and maps depicting good,
fair and poor air quality levels over the region. Respondents had little
difficulty understanding the levels of air quality represented to them because
of the sharp pollution gradient across the region.
13
Payment mechanisms were specified within the survey instrument and the
respondent was asked to react to alternative price levels posited for
different air quality levels. In every case the basis for the bid for better
air quality was the existing pollution situation as determined by location of
their home shown on a map of the Log Angeles metropolitan area which depicted
regional air quality levels. Various starting points for the bidding prices
and differing information structures were included in the survey format.
Biases from alternative starting points and information^|itructures were not
present in the results [See Brookshire, et al. (1980)].
The survey was conducted over the period of March, 1978. A total of 290
completed surveys were obtained^for the above mentioned areas. Sampling was
random within each paired area.
Table 2 in the next section presents the mean bids and standard devia-
tions and number of observations in Columns four and five respectively for
50
-------
each community for an improvement in air quality. Two types of bids are pre-
sented: proposed improvements from poor to fair air quality and from fair to
good air quality. In poor communities--El Monte, Montebello and La
Canada--the mean bids ranged from $11.00 to $22.06 per month. For the fair
communities—Canoga Park, Huntington Beach, Irvine, Culver City, Encixio and
Newport Beach communities --the mean monthly amounts range from $5.55 to $28.18
to obtain good air "quality.
TEST OF HYPOTHESES
The previous sections have described a theoretical structure and two
different empirical estimation techniques for determining the value of urban
air quality improvements in the Los Angeles metropolitan area. The theoreti-
cal relationship between the valuation procedures (AR W) and the hypothesis
that survey bids are non-zero (W > 0) are tested in this section.
Table 2 presents the community average survey bids (column four) and
corresponding rent differentials (column two). As is indicated, in each com-
munity the sample survey bids are non-zero and less than the calculated rent
differentials in absolute magnitude. This establishes that the survey bid
bounds are consistent with our theoretical arguments but does not indicate
statistical significance, which is provided below.
With respect to the test of equality of mean survey bids to zero, Table 2
(column six) presents the experimental results. The calculated t-statistics
indicate rejection of the null hypothesis (that the population mean, \i- equals
zero at the one percent level in every community sampled.) These resuYts are
in accordance with the political situation of the region and indicate that
individual households are willing to pay amounts significantly greater than
zero for an approximate 30% improvement in air quality.
The comparison of the survey bids to the estimated rent differentials is
presented in Table 2 (column seven). In this instance the compound hypothesis
that population average rent differential (u—) equals or exceeds the
population average survey bid (u-) is again tested using the t-statistic.
Rejection of the null hypothesis requj^es that the calculated t-statistics be
negative and of sufficient magnitude. The standard t-test calculations
(column seven, Table 2) imply that the hypothesis y— >_ u~ cannot be rejected
for the population means y- and even at the 10% critical level. Although
we present only the results for the hedonic housing equation in which log
(NO^) is the pollution measure, these results remain essentially unchanged for
all communities, for all estimated hedonic rent gradients, regardless of the
variable (N02or TSP) utilized as a proxy for the general state of air qual-
ity. The results then are quite insensitive to the particular hedonic model
51
-------
specification, providing a degree of generality to the results.
The hypotheses tests indicate that the empirical analysis is entirely
consistent with the theoretical structure outlined above. This conclusion,
when combined with the absence of any identified biases [see Brookshire,
et al. (1980)] suggests that survey responses yield estimates of willingness
to pay for environmental improvements in an urban context consistent with a
hedonic- market analysis. A further implication is that individual households
demonstrated a non-zero willingness to pay for air quality improvements rather
than free riding. This conforms to the previous survey resuits of Brookshire,
et al. (1976) and Rowe, et al. (1980) as well as the experimental work of
Scherr and Babb (1975) , Smith (1977) and Grether and Plott (1979) concerning
the role of strategic behavior. This seems to indicate that the substantive
effort to devise a payment mechanism free of strategic incentives for con-
sumers [see Groves and Ledyard (1977)] has been directed towards solving a
problem not yet empirically observed. However, the conclusions of this
experiment are not without qualifications. In the next section possible limi-
tations of survey analysis and conclusions concerning the efficacy of
employing surveys to value a wide range of non-market commodities are
discussed.
CONCLUSION
There are a number of limitations in generalizing our results to all
survey work. First, this experiment was conducted in the South Coast Air
Basin where individuals have both an exceptionally well-defined regional pol-
lution situation and a well-developed housing value market for clean air. The
effect of clean air on housing values appears to be exceptionally well under-
stood in the Los Angeles metropolitan area. Thus, the Los Angeles experiment
may be a special case in which an informed populace with market experience for
a particular public good allowed the successful application of the survey
approach. In particular, situations where no well-developed hedonic market
exists may not be amenable to survey valuation. Biases due to lack of exper-
ience must then be considered a possibility. However, existing studies by
Randall et al. (1974) and Brookshire et al. (1976) and Rowe et al. (1980) of
remote recreation areas certainly suggest that survey approaches provide
replicable estimates of consumer's willingness to pay to prevent environmental
deterioration, without prior valuation experience.
In summary, this paper set out to both theoretically and empirically
examine the survey approach and to provide external validation for survey
analysis. The theoretical model described in Section 2 predicts that survey
responses will be bounded below by zero and above by rent differentials de-
rived from the estimated hedonic rent gradient. In order to test the dual
52
-------
hypotheses a survey and a traditional analysis of the housing market were
undertaken. Each was based upon a consistent but random sampling procedure in
the Los Angeles Metropolitan area. The empirical results do not allow the
rejection of either of the two hypotheses, thereby providing evidence towards
the validity of survey methods as a means of determining the value of public
goods .
53
-------
REFERENCES
1. Alternatively we could define the utility function U(-P, X) which
would be an increasing quasi-concave function of both arguments.
2. Primes or subscripts denote derivatives or partial derivatives
respectively throughout the paper.
3. The second expression is, of course, a vector of conditions, one
for each attribute.
4. For a continuous model one could specify a taste parameter in the
utility function and specify a distribution of households over that
parameter. To complete a closed model one also needs the distribution
of housing units over characteristics.
5. The paired areas with associated census tract marker and air qual-
ity level are respectively (1) Canoga Park - #1345 - fair/El Monte -
#4334 poor, (2) Culver City - #2026 - fair/Montebello - #4301.02
and part of #5300.02 - poor, (3) Newport Beach - central #630.00 -
fair/Pacific - northeast portion of //2627.02 and southwest inter-
section good; (4) Irvine - part of #525 - fair/Pales Verdes -
portion of good; (5) Eticino - portion of #1326 - fair/La
Canada - south-central portion of #4607 - poor; (6) Huntington Beach
central portion of #993.03 poor/Redondo Beach - eastern portion
of #6205.01 and #6205.02 - good. For a map showing the monitoring
station locations in relation to the paired sample areas and the air
quality isopleths see Brookshire, et al. (1980).
6. The estimation of a hedonic rent gradient requires that rather re-
strictive assumptions are satisfied. For Example, Kaler (1977) , has
raised a number of objections to the hedonic property value approach
for valuing environmental goods. These include the possibility that
transaction costs (moving expenses and real estate commissions) might
restrict transactions leaving real estate markets in near constant
disequilibrium; and that markets other than those for property alone
might capture part of the value of an environmental commodity. The first
54
-------
of these criticisms is mitigated by the extremely fluid and mobile real
estate market of the late 1970's in Los Angeles, where rapidly escalating
real property values increased homeowner equity so quickly that
"house jumping" became financially feasible. The second of Maler's
concerns, that other prices, e.g., golf club fees and wages capture
part of the willingness to pay can be addressed empirically. For
example* attempts to test if wages from our survey data across the
Los Angeles area reflected differences in pollution level produced
negative results.
7. Note that we use sale price or the discounted present value of the
flow of rents rather than actual rent as the dependent variable.
Given the appropriate discount rate the two are interchangeable.
8. Housing characteristic data was obtained from the Market Data
Center, a computerized appraisal service with central headquarters
in Los Angeles, California.
9. Although the nonlinear equations provide large t values on the air
pollution coefficients, the coefficients on the pollution variables
in the linear equations possessed the expected relationship and were
significant at the 1% level. Also, the calculated rent differentials
associated with the linear specifications were larger than those from
the nonlinear equations.
10. it should be noted that the nonlinear estimated equations will give
biased but consistent forecasts of rent differentials. However, the
linear estimated equations in all cases forecast larger rent differentials
than the nonlinear estimated equations presented here.
11. A capital recovery factor equal to .0995 which corresponds to the
prevailing .0925 mortgage rate in the January, 1979 - March, 1978
period is used.
12. In developing photographs, two observational paths from Griffith
Observatory in Los Angeles were chosen: (1) toward downtown Los
Angeles, and (2) looking down Western Avenue. The approximate visi-
bility (discernible objects in the distance, not visual range) for
poor visibility was 2 miles, for fair visibility 12 miles, and for
good visibility 28 miles.
13. Payment mechanisms are either of the lump sum variety, or well
specified schemes such as tax increments or utility bill additions.
The choice in the experimental setting varies according to the
55
-------
structure of the contingent market.
14 Questions have been raised as to problems of biases in the survey
approach. Strategic bias (i.e., free rider problems), hypothetical
bias, instrument bias all have been explored. Generally speaking,
problems of bias within the survey approach have not been prevalent.
For a general* review of the definition of various biases and results
of different experiments see Schulze et al. (forthcoming) and for
investigations of strategic bias utilizing other demand revealing
techniques see Scherr and Babb (1975) and Smith (1979) .
15. Interviewer bias was not present. No records were kept that would
enable the testing for non-respondent bias.
16. For instance, rejection of the null hypothesis x. y-) at the
one percent level would require a calculated t-statistic less than
-2.326 given a large number of observations. Since none of the
calculated t-statistics are negative the null hypothesis cannot be
rejected [See Guenther (1973)].
56
-------
BIBLIOGRAPHY
Anderson R., and T.Crocker, "Air Pollution and Residential Property Values,"
Urban Studies, October 1971, 8, 171-80.
Bohm, P., "Estimating Demand for Public Goods: An Experiment," European
Economic Review, 1972, 3, 11-130.
Brookshire, D., R. d'Arge, W. Schulze and M. Thayer, "Experiments in Valuing
Public Goods," Advances in Applied Macroeconomics, cd., V. Kerry Smith,
JAI Press, 1980.
Brookshire, D., B. Ives and W. Schulze, "The Valuation of Aesthetic Prefer-
ences," Journal of Environmental Economics and Management, December
1976, 3, 325-346.
Bradford, D., "Benefit Cost Analysis and Demand Curves for Public Goods,"
Kyklos, November 1972, 23, 775-782.
Cummings, R., W. Schulze and A. Meyer, "Optimal Municipal Investment in Boom-
towns : An Empirical Analysis," Journal of Environmental Economics and
Management, September 1978, 5, 252-267.
Davis, R., "Recreation Planning as an Economic Problem," Natural Resources
Journal, October 1963, 3, 239-249.
Freeman III, A. Myriek, "Hedonic Prices, Property Values and Measuring Environ-
mental Benefits: A Survey of the Issues," Scandinavian Journal of
Economics, 1979, 81, 154-173.
Grether D., and C. Plott, "Economic Theory and the Preference Reversal
Phenomenon," American Economic Review, September 1979, 69, 623-638.
Groves T., and J. Ledyard, "Optimal Allocation of Public Goods: A Solution to
the ^Free Rider' Problem," Econometrics, May 1977, 45, 783-809.
Guenther, W., Concepts of Statistical Inference, McGraw-Hill 1973.
5 7
-------
Hammack J. , and G. Brown, Waterfowl and Wetlands: Toward Bioeecroomic Analysis,
Baltimore: John Hopkins University Press 1974.
Harrison, D., Jr. and D. Rubinfeld, "Hedonic Housing Prices and the Demand for
Clean Air," Journal of Environmental Economics and Management, March
1978, 5, 81-102.
* • V
Hoch, I. with T. Drake, "Wages, Climate, and the Quality of Life," Journal of
Environmental Economics and Management, December 1974, 1, 268-295.
Lancaster, K., "A New Approach to Consumer Theory," Journal of Political
Economy, April 1966, 74, 132-157.
Maler, K., "A Note on the Use of Property Values in Estimating Marginal
Willingness to Pay for Environmental Quality," Journal of Environmental
Economics and Management, December 1977, 4, 355-369.
Nelson, J., "Airport Noise, Location Rent, and the Market for Residential
Amenities," Journal of Environmental Economics and Management, December
1979, 6, 320-331.
Rowe, R., R. d'Arge and D. S. Brookshire, "An Experiment in the Value of
Visibility," Journal of Environmental Economics and Management, March
1980, 7, 1-19.
Randall, A., B. Ives and C. Eastman, "Bidding Games for Valuation of Aesthetic
Environmental Improvements," Journal of Environmental Economics and
Management, August 1974, 1, 132-149.
Rosen, S., "Hedonic Prices and Implicit Markets: Product Differentiation in
Pure Competition," Journal of Political Economy, January/February 1974,
82, 34-55.
Samuelscm, P., "The Pure Theory of Public Expenditures," Review of Economics
and Statistics, November 1954, 36, 387-389.
Scherr B., and E. Babb, "Pricing Public Goods: An Experiment with Two Proposed
Pricing Systems," Public Choice, Fall 1975, 23, 35-48.
Schnare, A., "Racial and Ethnic Price Differentials in an Urban Housing
Market," Urban Studies, June 1976, 13, 107-120.
Schulze, W., R. d'Arge and D. S. Brookshire, "Valuing Environmental Commod-
ities: Some Recent Experiments," Land Economics' (forthcoming subject
58
-------
to revisions).
Smith, V., "The Principle of Unanimity and Voluntary Consent in Social Choice,"
Journal of Political Economy, December 1977, 85, 1125-1140.
Tiebout, C., "A Pure Theory of Local Expenditures," Journal of political
Economy, October '1956, 65, 416-424.
59
-------
CHAPTER 4
THE ADVANTAGES OF CONTINGENT VALUATION METHODS FOR BENEFIT-COST ANALYSIS
INTRODUCTION
Historically, policy decisions regarding the alteration or manipulation
of natural systems have relied to some extent upon the methods of benefit-
cost analysis to provide information about the efficiency attributes of se-
lected alternatives. Construction programs of the Army Corps of Engineers and
the Bureau of Reclamation are probably the best examples. These programs have
usually had explicit market price information on the value of additional water
or electricity that could be used to analyze the benefits and costs. When
non-marketed goods, such as loss of wildlife habitat, were to be influenced by
the project, they were not formally incorporated into the benefit- cost
analysis.
A developing emphasis, however, on valuing non-marketed goods and incor-
porating these values into formal benefit-cost analyses can be traced in part
to recent Federal and State legislation oriented toward environmental quality
regulation and preservation. Quantification of non-market benefits to estab-
lish the economic efficiency of regulatory decisions is required, for example,
under the Occupational Safety and Health Act of 1970, the Safe Drinking Water
Act, the Clean Air Act Amendments of 1977, the Toxic Substances Control Act,
the Endangered American Wilderness Act, and others.
The implications of this requirement for policy and for benefit-cost
analysis are severe. First, many of the key components of the benefits are
several steps removed from a direct relation with a marketed good. When con-
sidering potential degradation of a Class I visibility area such as the Grand
Canyon National Park, how is value to be placed on the scenic beauty of the
colors and the pristine visibility? What is the value of being able to see
120 miles versus 90 miles? Additionally, what are the benefits of permanently
preserving ancestral habitat for bighorn sheep versus utilization of the area
for a natural resource development when both preservation and development can
benefit current and future generations? What are the benefits of reduced risk
60
-------
to life? What are three less years of life worth, or 20 illness days per
year? These types of questions either implicitly or explicitly are raised by
today's legislative mandate. Thus, recent legislative history asks of
benefit-cost analysis to assess various trade-offs for which many of the key
value components of the tradeoff process are not readily observable in the
market place. Further, many of the valuations necessary do not have readily
observable market surrogates available to impute the value of non-market com-
modities. For instance, property value studies have been proposed as a manner
to impute the value of air quality in urban regions. How could such an
approach possibly work in the Four Comers region of the sparsely populated
Southwest? How could travel cost methodologies possibly impute the value of
critical habitat preservation? The answer, of course, is that they cannot.
An additional issue in valuing non-market environmental goods is that the
policy alternatives frequently involve the provision of some quantity of the
good or the restructuring of property claims on the good in a fashion outside
the realm of recent historical experience. If behavior and valuations are
sensitive to these institutional and quantity perturbations, retrospective
observations have little to offer to the benefit-cost analyst. For example,
the development of a massive synthetic fuel industry in the Rocky Mountain
area could, if atmospheric emissions are uncontrolled, cause major
deteriorations in the area's atmospheric visibility. However, because there
has historically been little degradation of visibility in the area, that re-
cord which could allow the economic value of any change to be empirically
determined does not exist. To acquire the record, one must either develop the
synthetic fuels industry, hoping that the development can be reversed if the
value of atmospheric degradation proves excessive, or undertake small scale
experiments that generate data by artificially perturbing the essential
features of the problem. On the presumption that the former course can be
exceedingly expensive., we present heuristic arguments for the use of experi-
ments. Our attention is focused upon contingent valuation studies of complex
natural processes rather than upon carefully controlled laboratory studies.
Plott (1979) has recently written a valuable review and defense of laboratory
studies.
Contingent valuation studies are distinguished from traditional benefits
assessment practices by their use of survey questionnaires to acquire the data
for analysis. Despite a paucity of empirical evidence to support or deny its
significance, the systematic misrepresentation of preferences is widely
recognized among economists as being potentially a serious disadvantage of
using survey questionnaires for valuation purposes. Our purpose is to raise
the possibility that economists, by their near-exclusive devotion to the
strategic behavior problem, may, at their own apparently unrecognized cost,
have neglected many of the analytical and empirical advantages to be reaped
61
-------
through the use of survey instruments to acquire valuation information. We
will examine these advantages in terms of the contribution survey question-
naires can make to filling the informational void the policymaker now often
faces, and in terms of the conformity of their data-generating process with
the economic-theoretic foundations of benefit-cost analysis. Any thorough
assessment of the relative reliabilities and validities of data generated by
survey questionnaires'and by observed behavior must weigh these advantages.
CONTINGENT VALUATION APPROACHES
The key to contingent valuation approaches to valuing a non-marketed good
is the construction of a hypothetical market for that good. The procedure is
as follows:
a. The non-market commodity is described in quantity,
quality, location and time dimensions. Various types
of supplementary information including maps and photo
graphs are introduced when appropriate.
b. The rules of operation of the hypothetical market are
established. Then a representation of the available
quantity of the environmental good is perturbed and the
respondent is asked to state willingness-to-pay or
required compensation, or the activity substitutions and
expenditure adjustments he would make. Both a status quo
quantity of the good and price are explicitly stated by
the interviewer prior to any respondent statements. The
first is a direct approach, while the second provides infor
mation for using the indirect techniques commonly employed
with data on actual observed behavior.
c. The market rules of operation, bidding vehicles, and status
quo prices and quantities may differ across respondents.
Each respondent is presented a status quo price and/or
quantity of the non-marketed good; the price and/or quantity
of the good is then altered by the interviewer until a com-
bination is reached to which the respondent is indifferent.
Thus, a series of contingent markets are established with a mechanism of
payment suggested for the alternative levels of the non-market good in
question. For instance, a proposed power plant of 1000 kilowatt capacity
located ten miles from a site is said to result in a 25 mile reduction in the
visual range, and the respondent is asked whether he would be willing to pay
perhaps fifty dollars over some specific time period to prevent the reduction.
62
-------
An important element in the process is clearly defining the non- marketed good
in a manner that establishes a clear linkage to physical parameters. For
atmospheric visibility, this would include linking power plant emissions to
ambient concentrations, and ambient concentrations to the representation of
ambient concentrations used in the interview.
Bradford (197?)) has set forth the analytical basis of the direct version
(bidding games) of the contingent valuation technique. Davis (1963) and
Randall, et al. (1974), made the first empirical applications to environmental
goods. Instruments that collect information on time and budget adjustments
and then employ this information to infer valuations of a non-marketed good,
have the bulk of their analytical foundations presented in Hori (1975) and
Freeman (1979) .
Published papers employing these contingent claims games to acquire
information have valued non-marketed goods as diverse as public television
programming [Bohm (1972)]; atmospheric visibility [Randall, et al. (1974,
Brookshire, et al. (1976), and Rowe, et al. (forthcoming)]; land-form alter-
ations due to strip mining [Randall, et al. (1978)]; air pollution-induced
health effects [Loehman, et al. (forthcoming), and Brookshire, et al. (forth-
coming (a))]; wildlife [Hammock and Brown (1974) and Brookshire, et al.
(forthcoming (b))]; water pollution [Gramlich (1977)]; preservation of river
headwaters [O'Hanlon and Sinden (1978), and Sinden and Wyckoff (1976)]; urban
infrastructure allocations for expenditures and taxes [Strauss and Hughes
(1976)]; and airplane safety [Jones-Lee (1976)]. In addition, there are a
number of as yet unpublished reports and papers that have used the technique
to value atmospheric visibility [Horst and Crocker (1978)1; power plant
cooling towers [Curry, et al. (1979)]; boomtowti infrastructure [Cummings and
Schulze (1978), Brookshire and d'Arge (1979)]; urban public parks [Vaughn
(1974)]; odors [Loehman, et al. (1978)]; and geothermal steam development in
wilderness areas [Ben-David, et al. (1977)].
One might reasonably conclude from this listing that in spite of the
persistently held belief that valuations established through contingent
(hypothetical) claims games are systematically biased, there have neverthe-
less been some economists who have overcome their skepticism. However, they
have not yet offered a coherent presentation of the advantages of their
technique. In succeeding sections, we present some of the elements on which
advantages might stem.
CONTINGENT VALUATIONS AND THE CONSUMER SURPLUS FRAMEWORK
Buchanan (1969) distinguishes between ex ante and_ex post costs. He
argues that it is the former that is relevant to choice. We employ the
63
-------
distinction to establish the place of contingent valuations in a consumer
surplus framework. Contingent valuations are seen as providing a means for
the potentially affected individual to participate in the choice of the
provision The choice to be used for valuation purposes is based upon "what
could be," rather than upon "what might have been."
In making a decision about cost (either a bid in direct valuation, or
reallocation of time and budget components in indirect valuation), an indivi-
dual in the contingent valuation approach is setting forth his evaluation of
the prospective sacrifices or gains in utility as a result of the proposed
contingencies. Thus, cost is a choice-bound concept, and choices are based on
prospects referenced in the type of information provided. Cost, then, in its
relationship to choice must be based on expectations, not experience. This
viewpoint suggests that : (1) the oft-discussed discrepancies between observed
and proposed behavior [e.g., Fromm (1968) and Mills and Feenberg (1977)] are
not an issue in valuing non-market commodities unless the information
underlying the proposed behavior is identical to the information leading to
the actual behavior; (2) for given information, the contingent valuation
framework provides valuations in terms of expected value to the individual,
i.e. , willingness-to-pay for the prospective outcome.
Let us consider an example that illustrates these points in the context
of a contingent valuation market. Assume that a respondent's demand for a
marketed activity (e.g., camping in a national park) is weakly complementary
in the non-marketed commodity (e.g., visibility |s measured by the distance
that can be seen in and around a national park) . Participation in camping is
assumed to have an invariant opportunity cost of P, which is independent of
the level of availability of atmospheric visibility in the national park. In
Figure 1, the D curve represents the individual's income-compensated demand
function for the camping activity (A), averaged over all possible levels of
atmospheric visibility.
The ability to observe distant mountains from the camping site enhances
the utility of camping. The efficient plan for the camper with no forecast of
the availability of clear vistas is to undertake the activity at activity
level a in Figure 1, where a represents average visibility versus a pristine
or murky level of visibility. The marginal value attached to an additional
planned unit of camping just equals the individual's opportunity cost. The
consumer surplus expected from camping, once the activity begins, is the area
above the opportunity cost line and beneath the "average demand" function D.
The latter is the individual's mathematical expectation of the valuation
attached to camping levels, once realized.
Now suppose a formal contingent market is constructed where the non-
64
-------
Figure 4.1
Effect of An Improvement in Information
on Consumer's Surplus.
(D | M)
(D|c)
(D)
M
C
65
-------
marketed commodity, atmospheric visibility in and around the national park, is
described to the would-be camper in the requisite detail. For instance, coal
fired power plants will either be installing control devices or shutting them
down for maintenance. Visibility in and around the park will thus be clear
(C) or murky (M) during the camping trip. The manner in which the camper will
revise his estimates about the probability of clear or murky conditions can be
described by Bayes*' (1764) rule. If the information leads to the conclusion
of clear visibility, the camper's subjective evaluation of his average
compensated dgmand function will be (D/C), with a planned increase in camping
activity to a „ The area (b-d-e-f) gives the increase in expected utility if
"clear" is the forecast. ^f the forecast is "murky," the planned activity
level will be reduced to a0, and the area (b-d-h-g) gives the loss in expected
utility.
K
Now suppose that M is forecast, implying a planned activity level of a *
and an expected consumer surplus of g-p-h. If, instead, C is realized and the
camper is unable to adjust his activity level accordingly, he will have
obtained a consumer surplus of p-h-f-n, an amount greater by g-h-n-f than the
consumer surplus on the basis of which he made his decision to go to the park.
This latter consumer surplus, which is established from observed behavior, has
no correspondence to the basis of the camper's choices. In fact, according to
the opportunity the camper has to adjust^his activities, any activity level
from the origin to a. might be observed. "Only if clear skies had been
forecast and actually realized would the camper's expected and realized
consumer surpluses coincide, thus allowing the investigator to infer the
utility basis of the camper's choices from his observed behavior. In
contrast, contingent valuation techniques place the individual in a
representation of the context in which he actually makes choices. Unless
policy maker decisions about levels of provision of non-marketed goods are to
be only randomly connected to the nexus the individual confronts, the
appropriate state for measuring consumer surplus is that corresponding to the
instant of the individual's decision.
HYPOTHETICAL BEHAVIOR AND MARKETS
If different answers can be anticipated based upon alternative infor-
mation structures, what "state" is the appropriate one for measuring consumer
surpluses for benefit-cost analysis? Can a contingent market be developed
that is "appropriate" to the policy question at hand? What happens if infor-
mational content of an observable market is identical to that of a contingent
market.
Frotraa (1968) and many other economists believe that hypothetical ques-
tions generate fictional and, therefore, inaccurate responses. The dictionary
66
-------
defines a hypothetical proposition as a conditional proposition, i.e., an "if
X, then Y" statement. A hypothetical question would then be a conditional
statement in the subjunctive mood, an "if X were . \ ., then . . . state-
ment. In the contingent valuation setting, a hypothetical market is con-
structed, perturbed and then the respondent states conditional behavior based
on the specified market structure or events. Fundamentally, the problem is
not hypothetical, bfct ^one of the relation between information and choice as
set out for the camper in the immediately preceding section.
The individual's ultimately realized benefits and his prospective eval-
uations are neither jointly instantaneous nor coincidental. Frequent discrep-
ancies should then be expected between response to contingencies embodying one
form of information and eventual observable behavior carried out upon the
basis of altered information. The key point is that the contingent answer is
still acceptable given the well-defined circumstances that were presented to
the respondent. The question of inaccuracy is then not whether, given a
change in circumstance, the observable behavior pattern changes, but whether
the contingent answer can be observed when the defined circumstances have not
changed. Only if the answers relate to past rather than intended behavior
will a simple comparison of answers with actual behavior suffice to ascertain
the accuracy of the answers. Otherwise, one must explain how the individual
responds to new information and circumstances in order to perform the com-
parison.
An empirical rebuttal of these points would require evidence that the
provision of additional information about future states does not change con-
tingent values and that contingent values and observed market values fail to
coincide when the defined circumstances in the hypothetical market and the
actual market are similar. In this section, we offer brief summaries of two
studies that contribute to this empirical evidence.
The first study was performed in Farmington, New Mexico, where a hypo-
thetical market for alternative levels of visibility due to additional energy
development [Rowe, et al. (forthcoming)] was developed. The appropriate
"states" corresponded to energy development scenarios for the Four Corners
area. To investigate the role of information, after direct valuation res-
ponses had been received, a subsample of respondents was told either that
others had bid a certain amount or that the bid of the subsample was so low
that the proposed change in the allocation of the non-market good was im-
possible. In both instances, respondents were given the opportunity to revise
their valuation. The results indicated that the valuation measures were af-
fected by the structure and the information content of the contingent market.
Thus, at least in this case, information about the behavior of other market
participants affected valuations. This behavior is, of course, consistent
67
-------
with the strategic behavior predictions of the free-rider decision problem in
public choice theory.
The second study was conducted in the South Coast Air Basin of southern
California [Brookshire, et al. (forthcoming (a))]. A residential property
value study based on sales of individual properties in a sample of paired
communities where mcst of the variation in physical attributes within a pair
was due to air pollution was performed. Similarly, during the same time per-
iod that the property sales occurred, a contingent valuation study within the
same paired community sample was undertaken. The set of circumstances de-
picted in the contingent valuation study was those actually prevailing in the
Basin at the time of the property value sales. Within a factor of less than
two, the two independent studies produced similar valuations. For an approx-
imate 20% improvement in the ambient air quality of the Basin, the property
value study gave an average dollar bid per household per month of $42, while
the bidding game study yielded a mean bid of $2 9 per household per month.
COSTLESS VERSUS COSTLY EXCHANGE
Even if the information available to participants in an everyday actual
market and in a contingent valuation exercise were identical, there remains at
least one reason why the two types of valuations might still diverge: the in-
stitutional structures of the contingent valuation market and the everyday
market may differ. Plott (1979) reviews several empirical laboratory and
field experimental studies indicating that market outcomes are highly sensi-
tive to differences in institutional structures. Given this sensitivity, if
meaningful measures of the gains and losses from the provision of a non-mar-
keted environmental good are to be established, the measures must be derived
within an institutional structure conforming to that posited in the welfare-
theoretic basis of benefit-cost analysis. In this section, we argue that this
conformity is often more readily achieved with the use of contingent valuation
techniques.
Benefit-cost analysis is an attempt to ascertain the quantity of some
numeraire (e.g., current dollars) that the gainers and losers from some pro-
posed public investment will consider equivalent in value to their respective
gains and losses. The price structure, where price is a sufficient measure of
social as well as private value, represents the only terms with which the
world with or without a public investment is evaluated. Prices, as generated
by market exchange and adjusted in proportion to excess demand, embody all
relevant information about relative economic scarcities and are a sufficient
means of allocating resources to their socially most highly valued uses. The
benefit-cost analyst is trying to ascertain what individuals are willing to
pay and/or would have to be paid for the public investment in a world where
68
-------
markets are pervasive.
If realized market behavior is used as the data base toestablish these
valuations, the analyst uses propositions from economic theory for two pur-
poses: (1) toinfer what the price structure would be in a world of pervasive
markets; and (2) to reason from the pervasive market price structure to the
implied consumer valuations. When contingent valuation responses are employed
for the data base, the first step can be avoided, if the conditions posited in
the questionnaire instrument correspond to a world of pervasive markets. One
might reasonably question whether the conditions corresponding to a world of
pervasive markets are sufficiently close to a respondent's experiences to be
meaningful to him. This justifiable doubt must be weighed however, against
the difficulties of carrying through the analytical exercises necessary to
construct a pervasive market price structure from initial knowledge of the
price structures of a world where markets for many goods are not pervasive.
The way in which this difficulty is customarily avoided when using observable,
realized prices is to assume (for simplicity?) that the observed prices
correspond to those in a world of pervasive markets.
It is a relatively easy task to construct examples that make apparent the
difficulties of reasoning topervasive markets from observations on non-
pervasive markets. Consider costs of exchange, a phenomenon present whenever
valuable resources (e.g. , time, information, legal and police services, etc.)
must be expended toperform the exchange process.
In Figure 2, the individual's initial endowment of Y and Y_is _Q %
When exchange processes become costly, the individual's budget constraintwiii
vary according to his initial endowment. This is because the costs of the act
of exchanging Y and Y2 differ from the costs of exchanging Y for Y . For
example, from tie perspective of a single individual, the cost of engaging in
a transaction in which he is to exchange automobiles that he owns for clean
ambient air may differ from these same costs in a transaction where he is
exchanging clean air for automobiles. If the exchange act is costly, an
initial endowment of Q implies a budget constraint of VQV, whereas if the
exchange act is costless, the budget constraint is MM, the customary form
which isan integral part of derivations of demand functions and their assoc-
iated consumer surpluses. Then the individual completes his exchanges during
the period, he will select, Y° and"%2 as an optimum if MM is operative. If VQV
is the operative budget constraint, he will select Y' and Y'^. If some point
on MM other than Q constitutes the initial endowment, costly acts of exchange
will mean that a budget constraint different from either VQV or mm may be
operative because the costs of exchange acts may differ by the relative
quantities of the goods in the initial endowment as well as by types of goods.
Thus, the individual's budget constraint may vary according to the form in
69
-------
Figure 4.2
Effect of Costly Exchange
2
M
v
0
2
1
2
Y?
y:
V
70
-------
which his initial endowment was accumulated, although the market value of this
endowment may be identical for many combinations of Y and Y . Since costs of
L 2
the exchange act differ according to the original (Y^.Y^) combination, each
combination will result in a different and generally nonlinear budget
constraint. It follows that, from the individual's perspective, a dollar is
not an invariant pecuniary measure. Instead, the subjective value of an
additional dollar &ep6nds on the fm of the income change, i.e., on the good
in which the increment is embodied! Moreover, it appears that realized
market behavior is dependent not only on money incomes and relative market
prices of goods, but also upon the combination of goods the individual starts
with and the relative and absolute costs of exchange associated with those
goods . These costs of exchange acts are probably neither trivial nor similar
across individuals.
If realized market behavior depends on the costs of the exchange act for
the bundle of goods an individual holds, if, for the same bundle of goods,
these costs differ across individuals, and if individuals do not hold similar
goods bundles, then the analytical effort required to infer what the price
structure would be in a world of pervasive markets must clearly be greater
(probably much greater) than when all individuals have no exchange act costs
and when budget constraints are therefore invariant with respect to the bundle
of goods held. Rather than facing these and similar analytical complexities
directly in order to construct the price structure of a world of pervasive
markets, or rather than simply dismissing the problem as an offensive bother,
it may often be more effective to construct, using contingent valuation tech-
niques, an artificial market for the environmental good to be valued. For the
contingent valuation exercise participant, the number of goods for which
markets are non-existent or incomplete is thereby reduced by one. This re-
duction clearly cannot remove all sources of distortion since the
participant's valuation continues to depend in part upon the price structure
for all remaining goods. Nevertheless, it is well known that the direct
effects of a parameter change upon a variable of interest exceed the indirect
effects. This suggests that the introduction of the artificial market reduces
rather than enhances the impact upon valuation of the presence of incomplete
markets.
BENEFIT-COST ANALYSIS, PROPERTY RIGHTS, AND CONTINGENT VALUATIONS
The fact of incomplete markets says nothing about the degree of distor-
tion in the observed price structure for marketed goods. Some recent qual-
itative literature [e.g., Norgaard and Hall (1974), and Smith and Krutilla
(1979)] suggests that the extent of distortion could be substantial. The
great bulk of goods having actual market prices are thought to be primary
commodities and the goods chemically and mechanically fabricated from them.
71
-------
Because the costs of participating in direct exchanges involving the aesthe-
tic, health and ecological support system aspects of natural environments are
high, they frequently have no explicit market prices even though they are
economically scarce and contribute in a non-separable fashion to the pro-
duction of the fabricated goods. Asa result, the market prices of fabricated
goods are typically less than their opportunity costs of production. In
short, those who attach high relative values to environmental goods have
historically subsidized the consumers of fabricated goods. To use a price
structure that has evolved at the expense of environmental goods to impart
values to them has no basis in economic logic. The values must be established
in a setting where the opportunity costs of environmental goods register in
the plans of those who would use them.
Attempts to bring about this registration must generally involve the
reassignment and/or the restructuring of claims on common property or public
environmental goods. There exist analytical devices in economics allowing one
to ascertain the effects upon consumer valuations of property rights re-
assignments for goods, whether marketed or non-marketed. However, where the
conditions of use, exclusion, or alienation are altered (i.e., property rights
are restructured), there is no everyday behavior to observe, except insofar as
one is willing to draw analogies from observed behavioral responses to changes
in the property rights structures of other goods. If one knew what the
availability of the non-market good would be under the property rights
restructuring, it might seem one could, if one had everyday behavioral obser-
vations on consumer time and budget allocations at the same level of avail-
ability, determine the change in consumer valuation due to the property right
restructuring. However, if the restructuring reduces the costs of the act of
exchange this reduction can, as we argued in the previous section, alter the
value the consumer attaches to a given level of availability. Furthermore,
since consumer valuations will, through either the market or the political
process, influence the level of availability, how is one going to reason from
the level of availability to consumer valuations for the restructured property
right? Economic analysis does not yet have a sufficient understanding of the
reciprocal relations between costs of the act of exchange and property rights
structures, nor between these costs and various demand phenomena, to permit
the ready testing of detailed empirical generalizations in a wide variety of
settings. Thus, the only really sound way of obtaining an estimate of whether
the net benefits of a particular property rights restructuring are positive,
if one insists upon employing observed everyday behavior, would be to perform
the restructuring and observe the results. In some circles, this is simply
known as trial and error. Trial and error can be an extremely costly way to
perform research because the errors are real rather than hypothetical. In
contrast, contingent valuation methods allow one to investigate the behavioral
responses to a wide variety of property rights structures without involving
72
-------
the citizenry in the traumas of what often is euphemistically termed social
experimentation.
One obviously cannot directly observe everyday behavioral responses to
property rights structures that have never existed. Similarly, one cannot
directly observe the everyday behavioral responses of individuals who have
never participated'in activities involving the non-marketed good at levels at
which the good has been historically available. If some of the proposed
levels of availability have not been historically available, and if some
former nonparticipants would become participants at these new levels, the use
of data on observed behavior to ascertain valuations would mean that the
valuations of the would-be participants play no part in determining the valu-
ation. For each proposed level of availability, the use of observed, realized
behavior to establish valuation will mean that only historical participants
are to count. Those who have not participated historically have no opportun-
ity to communicate their preferences. Contingent valuation methods, because
they allow the researcher to introduce ranges of availability of the non-
marketed good that are broader than historical experience, permit the values
of historical non-participants to become relevant.
CONTINGENT METHODS AND A PRIORI ASSUMPTIONS
Previous sections have stressed the usefulness of contingent valuation
methods in traveling from prices, established within incomplete markets to
value measures that are meaningful in welfare-theoretic and policy terms. In
this section, we argue that these methods are useful even when the trip is
unnecessary as when expected and realized utility are similar and when markets
are nearly pervasive. The methods can be useful in even these cases because
they assist in reducing the dimensionality of the reality the investigator
must grasp.
Economists who have worked with problems of consumer analysis are
thoroughly familiar with three fruitful a priori restrictions (additivity,
homogeneity, and symmetry) that come from the neoclassical demand theory of
Slutsky (1915) and Hicks (1934) . Further reductions in dimensionality of the
parameter space in which estimation is to be carried out can be achieved by
judicious invocation of various separability conditions. Finally, some recent
developments in the application of mathematical duality principles (the enve-
lope theorem) to consumer theory sometimes allow one to reduce the number of
parameters to be estimated without having to impose particular monotonicity
and curvature properties upon the consumer's maximization problem (See
Diewert, 1974).
Contingent valuation methods can provide additional restrictions by
73
-------
allowing the investigator to control the number and levels of different physi-
cal contexts and adaptation opportunities to which the participant must re-
spond. Disturbances imposed by confounding variables upon the responses of
interest are therefore at least partially controlled for in the data generat-
ing exercise. This contrasts with the standard practice of placing sole
reliance in an ex post fashion upon the application of multi-variate para-
metric estimation techniques. For a given number of observations, these
methods can thus increase degrees of freedom and the efficiency of estimators.
For instance, in the South Coast Air Basin Experiment previously mentioned,
the contingent value approach was able to obtain separate dollar valuations
for aesthetic as well as acute and chronic health effects. In contrast, one
can only guess what the relative magnitudes are for the property value com-
ponent cross check.
The use of contingent valuation techniques to reduce the parameter space
may be advantageous for reasons in addition to statistical considerations.
Often, as noted above, the investigator imposes, _ex post, various separability
conditions upon market-generated data in order to make it more tractable.
These separability conditions may imply, for example, that beer drinking at
the local tavern is not a substitute for cross-country skiing. The conditions
are imposed without consulting the individuals whose responses are registered
in the market data. They are instead generated by what the investigator in-
tuitively feels to be "reasonable," and what is required for analytical con-
venience. It is not obvious that the investigator's "feelings" and the frame-
work he uses in accounting for what is and what is not important is to be
preferred to actually providing the respondent with the opportunity to state
how he would respond to alternative contingencies. The details to be ab-
stracted from are presented to the respondent rather than being left to the
investigator. In both situations, simplifications are made that will permit
the investigator to work with the data. In the contingent valuation case,
however, the respondent gets the opportunity to weigh the importance of these
confounding variables in making his choices. In the observed behavior case,
the investigator is presuming he knows as well as the respondent what, from
the respondent's perspective, is and is not an irrelevant alternative. The
closed questions employed to gather data with contingent valuation methods
allow the domain in which the response data is generated to conform to the
structures of the underlying analytical model rather than forcing, via a set
of possibly tenuous assumptions (e.g. , the absence of jointness, the presence
of perfect competition, etc. the real world generated data to conform to the
preconceptions of the model. At the same time, the user of the methods must
accept ultimate responsibility for the origin of the data, as well as the ana-
lytical model and the estimation procedures used to test hypotheses.
74
-------
SUMMARY AND CONCLUSIONS
The preceding is a taxonomie discussion of some reasons why contingent
market methods may often be a superior means of generating data with which to
value non-market commodities. We have argued that economists have erred in
viewing the situations these methods posit as necessarily fictional; that the
data generated by the'methods may, for non-marketed goods and the activities
with which they are associated, accord more closely with the conditions of
received economic theory; that the methods can make it easier to remove the
difficulties of estimation and interpretation introduced by confounding vari-
ables; and that they often permit one to deal more readily with phenomena that
have not been in the range of historical experience. Nevertheless, whatever
the advantages, a major disadvantage remains. Until detailed analytical
knowledge is acquired of the manner in which expectations are formed, there
exists no way to refute empirical propositions established from contingent
markets. Nevertheless, the previously mentioned South Coast Air Basin experi-
ment (Chapter 3), where the bids obtained for clean air conformed fairly
closely to the values implied in a residential property value study, suggest
that contingent valuations have a basis in the real decision processes of
consumers.
75
-------
REFERENCES
1 Issues of potential bias in any mechanism that elicits preferences have
long been raised (Samuelson, 1955) . It is not our purpose to address
these, as series of contingent valuation experiments suggest the problem
is not significant [Brookshire, et al. (forthcoming (a))].
2 According to Maler (1974, pp. 183-189), weak complementarily exists if
the quantity demanded of a marketed good is zero when the marginal
utility of the non-marketed good is zero. Bradford and Hildebrandt
(1977) have recently expanded the Maler (197,4) result to show that under
weak complementarily all information required for efficient provision of
the non-marketed good is imbedded in the demand functions of marketed
goods .
3 Adaptive behavior, once having committed one's self and experiencing
unanticipated regret or satisfaction thereby, can be treated as the
acquisition of further information.
4 As used here, "social" refers solely to a world in which all voluntary
gains from exchange, given the initial distribution on income, are
exhausted. Only under classical conditions (an absence of noticonvexities,
irreducible uncertainty, coordination costs leading to externalities, and
less than complete contingent claims markets), does current economic
knowledge demonstrate that market prices alone would be sufficient to
make efficient (Pareto-optimal) allocations attainable.
5 Empirical evidence to support this is widely available. Newhouse, et al.
(1974) find that the demand for health care is sensitive to modes of pay-
ment. Keeley, et al. (1978) obtain the same result in the demand for
leisure when the form of a negative income tax is altered. In contingent
valuation exercises, Rowe, et al. (forthcoming), and Brookshire, et al.
(forthcoming) have found statistically significant differences in bids
when utility bills, income changes, and hunting license fees are employed
as bidding vehicles. Indeed, the standard undergraduate problem of
whether one would prefer a housing allowance to an income subsidy of
equivalent money value implies that the former is not readily converted
76
-------
to the latter.
6 Consider the simple consumer's utility maximization problem of:
maximize U = u (Xx,X2)
" ''' subject to M = +
J *1 1 12X2
where the x^(i»l,2) are goods, the p ^are their respective unit prices.,,
and M is money' income. An interior maximum requires that l^U
_> 0, where the subscripts indicate partial derivatives taken wiui respect
to the good in question. This says that the effects upon the utility
obtained from one good due to a change in another good cannot dominate
the direct utility effects of a change in either good.
7 If, for example, there is an increase in pollution, the amount the
sufferer would have to be paid in order to be willing to accept the
increase is consistent with the polluter being liable for the damages he
causes. The amount the consumer would be willing to pay to prevent the
increase implies that the polluter has zero liability for any harm he
imposes upon the sufferer.
8 As Medawar (1979, p. 15) has remarked: "It is a truism that a ^good'
experiment is precisely that which spares us the exertion of thinking;
the better it is, the less we have to worry about its interpretation,
about what it really means."
77
-------
BIBLIOGRAPHY
Bayes, T., 1974, "An Essay Towards Solving a Problem in the Doctrine of
Chances." In Philosophical Transactions, Giving Some Account of the
Present Undertakings, Studies, and Labours of the Ingenious in Many
Considerable Parts of the World for the Year 1763. London: The Royal
Society.
Ben-David, S., Schulze, W. D., Nell, S., and Thayer, M., 1977. "The Economics
of Geothermal Energy: A Case Study." National Science Foundation
Project Report APR 75-18249.
Bohm, P., 1972. "Estimating Demand for Public Goods: An Experiment."
European Economic Review 3, No. 4, 111-130.
Bradford, D. D., 1970. "Benefit-Cost Analysis and Demand Curves for Public
Goods." Kyklos 23, No. 4, 775-387.
Bradford, D. and Hildebrandt, M., 1977. "Observable Preferences for Public
Goods." Journal of Public Economics 8 (June): 106-114.
Brookshire, D., Ives, B., Schulze, W., 1976. "The Valuation of Aesthetic
Preferences," Journal of Environmental Economics and Management, Vol 3.
No. 4, 325-346.
Brookshire, D., d'Arge, R., Schulze, W., and Thayer, M., (forthcoming (a))
"Experiments in Valuing Public Goods." ed. by V. K. Smith in
Advances in Applied Micro-Economics, JAI Press, Inc.
Brookshire, D., Randall, A., and Stoll, J., (forthcoming (b)), "Valuing
Increments and Decrements in Natural Resource Service Flows."
American Journal of Agricultural Economics.
Brookshire, D., d'Arge, R., 1979. "Resource Impacted Communities: Economics,
Planning, and Management" paper presented at 4th U.S.-U.S.S.R. Symposium
78
-------
on Comprehensive Analysis of the Environment, Jackson, Wyoming.
Buchanan, J., 1969. Cost and Choice. Markam Publishing Company, Chicago.
Cummings, R. G., and Schulze, W. , 1978. "Social Valuation of Municipal Infra-
structure in Boomtowns: A Second Look." Manuscript.
Curry, J. W., 1979. "The Visual Aesthetic Impact of Alternative Closed Cycle
Cooling Systems," Vol. 1. Report to U.S. Nuclear Regulatory Commission.
Davis, R., 1963. "Recreation Planning as an Economic Problem." Natural
Resources Journal 3: 239-249.
Diewert, W. E., 1974. "Applications of Quality Theory." In Frontiers of
Quantitative Economics, Vol. 11, eds. M. D. Intriligator and D. A.
Kendrick. Amsterdam: North Holland Publishing Company: 106-160.
Freeman, A. M. III., 1979. The Benefits of Environmental Improvement.
Baltimore: Johns Hopkins University Press.
Fronaa, G., 1968. "Comment." In Problems in Public Expenditure Analysis, ed.
S. B. Chase, Jr., Washington, D. C.: Brookings Institution: 166-76.
Gramlich, F. W., 1977. "The Demand for Clear Water: The Case of the Charles
River." National Tax Journal 30, No. 2 183-194.
Hammaek, J. and Brown, G., 1974. Waterfowl and Wetlands: Toward Bioeconomic
Analysis. Baltimore: Johns Hopkins University Press.
Hicks, J. R., 1934. "A Reconsideration of the Theory of Value." Part 1.
Economics 1 (February): 52-76.
Hori, H., 1975. "Revealed Preference for Public Goods." The American
Economic Review 65 (December): 978-991.
Horst, R. and Crocker, T., 1978. "Household Substitution Model of the Value
of Visibility," paper presented at the Western Economic Association
Meetings, Honolulu, Hawaii.
Jones-Lee, M. W., 1976. The Value of Life. Chicago: The University of
Chicago Press.
Keeley, M. C., Robins, P. K. , Spiegelman, R. G. and West, R. W., 1978. "The
Estimation of Labor Supply Models Using Experimental Data." The
79
-------
American Economic Review 68 (December): 873-887.
Loehman, E., et al., (forthcoming). "Distributional Analysis of Regional
Benefits and Costs of Air Quality Control." Journal of Environmental
Economics and Management.
Loehman, E., Ben-David, S., De, V. H., 1978. "Measuring Demand and Political
Acceptability for Non-Market Goods: A Case Study of Health Effects
Related to Air Quality." Informal Note, CEPR-IN-78-3, SRI International
(October).
Maler, K. G., 1974. Environmental Economics. Baltimore: The Johns Hopkins
Press.
Medawar, P. B., 1979. Induction and Intuition in Scientific Thought.
Philadelphia: American Philosophical Society.
Mills, E. S., and Feenberg, D. , 1977. "Measuring Benefits of Water Pollution
Abatement." A paper presented at the Annual Meetings of the American
Economic Association, New York. (December 28).
Morgenstern, 0., 1959. On the Accuracy of Economic Observations. Princeton,
N.J.: Princeton University Press.
Newhouse, J. P., Phelps, C. E., and Schwartz, W. B., 1974. Policy Options and
the Impact of National Health Insurance. R-1528-HEW/DEP. Santa Monica,
California: Rand Corporation (June).
Norgaard, R. B. and Hall, D. C., 1974. "Environmental Amenity Rights, Trans-
actions Costs, and Technological Change." Journal of Environmental
Economics and Management 1 (December): 251-267.
O'Hanlon, P. W. and Sinden, J. A., 1978. "Scope for Valuation of Environmental
Goods: Comment." Land Economics 54 /August},: 381-387.
Plott, C. R., 1979. "Experimental Methods in Political Economy: A Tool for
Regulatory Research." Social Science Working Paper. California Insti-
tute of Technology.
Randall, A., Grunewald, 0., Johnson, S., Ausness, R. and Pagoulatos, A., 1978.
"Reclaiming Coal Surface Mines in Central Appalachia: A Case Study of
the Benefits and Costs." Land Economics 54 (November): 472489.
Randall, A., Ives, B. and Eastman, C., 1974. "Bidding Games for Valuation of
80
-------
Aesthetic Environmental Improvements." Journal of Environmental Economics
and Management 1, No. 2, 132149.
Rowe, R., d'Arge, R., Brookshire, D., (forthcoming). "An Experiment on the
Economic Value of Visibility." Journal of Environmental Economics and
Management.
Rowe, R., d'Arge, R., Brookshire D., (forthcoming). "Environmental Preferences
and Effluent Charges." Resource and Environmental Economics Laboratory,
University of Wyoming Research Paper # 6.
Samuelson, P., 1955. "Diagrammatic Exposition of a Theory of Public Expendi-
tures." Review of Economics and Statistics 37 (November): 350356.
Sinden, J. A. and Wyckoff, J. B., 1976. "Indifference Mapping: An Empirical
Methodology for Economic Evaluation of the Environment." Regional
Science and Urban Economics 6 (March): 81103.
Slutsky, E., 1915. "Sulla leoria del bilancio del consuraatore." Giormale
degli Economist 51: 1-26.
Smith, V. K. and Krutilla, J. V., 1979. "Resource and Environmental Con-
straints to Growth." American Journal of Agricultural Economics 61
(August): 395408.
Strauss, R. P. and Hughes, G. 0., 1976. "A New Approach to the Demand for
Public Goods." Journal of Public Economics 6 (October): 191204.
Vaughn, R., 1974. "The Value of Urban Open Space." Working Paper 78. Pro-
gram on Environmental Pollutants and the Urban Economy, Center for Urban
Studies, University of Chicago (August).
81
-------
CHAPTER 5
AN EXAMINATION OF BENEFITS AND COSTS OF MOBILE SOURCE CONTROL
CONSISTENT WITH ACHIEVEMENT OF AMBIENT STANDARDS IN THE SOUTH COAST AIR BASIN
INTRODUCTION
Chapter 3 described an experiment conducted in the South Coast Air Basin
to quantify and validate benefit measures of air quality improvements. The
qualified success of that effort suggests that a policy application of those
benefit measures may be appropriate. Thus, the intent of this chapter is to
examine national ambient air quality standards in a benefit-cost analysis
framework as applied to the South Coast Air Basin which consists of all or
portions of Los Angeles, Orange, Riverside and San Bernardino Counties in
Southern California.
The national ambient standards for oxidant (formerly .08 and now .12 ppm
maximum hourly concentration) and nitrogen dioxide (.05 ppm annual average
concentration) are consistently violated throughout the basin with the notable
exception of the immediate coastal areas which we have described as
characterized by "good" air quality in Chapter 4 [See Figures 3.13.4 in
Brookshire, et al. (1978) for a map of air quality areas] . Thus, in a broad
context, if the entire South Coast Air Basin were to be brought into
compliance with ambient standards, areas we have characterized as having
"fair" or "poor" air quality would then be characterized as having "good" air
quality. The development of an aggregate benefit measure for achieving
ambient standards for the entire basin is then a straightforward extrapolation
(given the original experimental design) where benefits are taken to be the
aggregate willingness to pay for all households in both "poor" and "fair" air
quality areas to have "good" air quality, as defined both for the preceding
property value and survey studies. Of course, any extrapolation to a
basin-wide population of about 2.4 million households (homes) from a sample of
719 home sales or from interviews with about 400 households can come under
serious question. In particular, the communities chosen for sampling,
although characterized by considerable variation in income and social
characteristics, may not represent a random sample of communities in the South
Coast Air Basin. However, the property value study does allow calculation of
82
-------
household willingness to pay as a function of income and air pollution. it is
this relationship that we use for benefit calculations assuming, in effect,
that income and population affect willingness to pay for air quality
improvement in the same way throughout the basin as they did in our limited
sample. Note that these estimates exclude any agricultural or ecosystem
effects.
Since benefits are calculated for moving from the current (1976 emissions
inventory) level of air quality to the ambient standards, costs must be
calculated on the 'same basis. However, our preliminary analysis indicated
that costs for on-road mobile source control measures were substantially more
defensible than those associated with stationary and institutional controls.
Therefore only the benefits and costs attributable to on-road mdjjile source
control are examined to the exclusion of other control measures. Benefits are
then those corresponding to the share of total emissions reductions which are
accomplished by on-road mobile source control. Costs are calculated for only
these measures also.
Although a careful engineering-cost study for using mobile source control
to achieve ambient standards would be desirable, the objective here must be
quite limited in that we are forced to use available cost evidence which in
many cases is quite uncertain. For the most part, we have relied on manu-
facturer statements and government publications for cost calculations. In
developing control cost estimates, given the large uncertainty which exists,
we simply present available data on the range of costs per ton of reduced
emissions for hydrocarbons, carbon monoxide and nitrogen oxides and, using
these numbers, estimate a broad range for basin-wide control costs to compare
to the range of benefit measures.
In addition, we have used the Air Quality Management Plan (January, 1979)
as the basis for the calculation of required emissions reductions.
Calculations presented in the plan indicate that to achieve ambient standards
in 1979 would require reductions of 974 tons per day in reactive hydrocarbons,
5963 tons per day of carbon monoxide and 503 tons per day of nitrogen oxides.
Of these amounts we have estimated that mobile source controls are responsible
for 728 tons/day, 6023 tons/day and 397 tons/day of hydrocarbons, carbon
monoxide and oxides of nitrogen, respectively. Our principle conclusions can
then be summarized as follows:
Benefits of achieving ambient standards for air quality
in the South Coast Air Basin for 1979 fall in a range of
1.6 to 3.0 billion dollars per year. Of this total on-
road mobile source control is responsible for approximately
1.36-2.55 billion dollars.
83
-------
Assuming that to achieve the ambient standards in 1979 the on-road
mobile source emission reductions are those stated above, then
corresponding total basin-wide control costs fall in the range of .6. -
1.32 billion dollars.
Benefits of control efforts to achieve ambient air quality standards
in the South Coast Air Basin appear to be of the same order of magnitude
as control costs. Given uncertainties over benefits and costs, this
implies that ambient air quality standards cannot be rejected as
economically inefficient on the basis of benefit-cost analysis.
Continued growth of population and economic activity in the South
Coast Air Basin could well alter the relative magnitudes of benefits
and costs of achieving the ambient standard in an unknown direction by
the attainment date of 1987.
The next section briefly discusses the use of this type of benefit-cost
study in policy analysis. Section 3 describes the construction of aggregate
benefits, costs of control are presented in Section 4, and Section 5 concludes
with a comparison of benefits and costs.
THE APPLICATION OF BENEFIT-COST ANALYSIS TO ENVIRONMENTAL STANDARDS
The application of benefit-cost analysis to environmental standards has
been described in great detail in the economics literature (see for example
Kneese and Herfindahl, 1974) . An ideal or optimal standard is one where net
benefits -- the difference between benefits of improved air quality and
control costs -- are the greatest. For example, Figure 1 shows the optimal
standard as S where the degree of air pollution control provides a level of
improved air quality (as measured on the horizontal axis) such that benefits,
B , exceed control costs, C , (or both measured on the vertical axis) to the
1 1
greatest extent. Note that in Figure 1, benefits are assumed to increase at a
decreasing rate with air quality improvement while control costs are assumed
to increase at an increasing rate. The slopes of these relationships are
presumed to arise respectively from (i) the diminishing rate of increase of
value to consumers of improved air quality as air quality approaches
"perfection" and (ii) because costs of additional emissions control will rise
increasingly rapidly as a level of zero emissions (i.e., perfect air quality)
is approached. At the optimal or economically efficient standard, S » given
our assumptions, benefits strictly exceed costs (B > CI) . Thus, in
evaluating the role that on-road mobile controls p lay in achievement of the
national ambient air quality standards as applied in the South Coast Air
84
-------
Dollar Valued
Benefits 8k Costs
C
Bo --
B. >
c,
I Control
costs
Benefits
0 s, S*
"Optimal Standard" "Excessive Standard"
Figure 5.1
Optimal and Excessive Standards
IMPROVED AIR
QUALITY
Dollar Valued
Benefits 8 costs
B.
o
' 3
Cent rol
costs
Benefits
"Undesirable Standard"
Figure 5.2
The Case of an Undesirable Standard
IMPROVED AIR
QUALITY
85
-------
Basin, the first test for economic efficiency is simply to check if benefits
exceed costs. Obviously, it would be desirable to construct benefit and cost
curves as shown in Figure 1 to pick the best standards for comparison to
actual standards. However, the uncertainty over benefits and especially costs
makes such an effort of doubtful value. Rather, given the likelihood of broad
ranges for benefits and costs as calculated for the ambient air quality
standards, we are Interested, from the perspective of economic efficiency, in
avoiding either a situation like S in Figure 1 or in Figure 2. In both
these cases, costs far exceed benefits and it is clear that the standards,, S.^
or S, , are economically inefficient. in the first case (S2 in Figure 1) the
standard has been pushed too far -- to the point where costs of control have
risen above benefits, implying excessive standards (B < C ). In the second
case (S ^ in figure 2), control costs are always above benefits and any stand-
ard is undesirable. Given that control costs typically rise very sharply as
emission controls become stringent, it is worthwhile, even with uncertain
estimates, to check if benefits are at least of the same order of magnitude as
costs.
Thus, placed in this perspective our objective is not to develop precise
and defensible cost estimates for comparison to benefit measures developed in
the preceding chapter, but rather to see if claimed ranges for control cost
options to achieve ambient standards possibly allow ambient standards to be
met at costs less than benefits.
BENEFITS FROM AIR QUALITY IMPROVEMENT
3
Description of the Study Region
The study area -- the South Coast Air Basin (SCAB) -- consists of Orange
and Los Angeles Counties and portions of San Bernadino and Riverside Counties
of California. This area has a long history of air quality problems. For
instance, Spanish explorers in the sixteenth century noted smoke from Indian
campfires in the basin, trapped by inadequate horizontal and vertical air
mixing. The post World War 11 period, characterized by Southern California's
rapid population growth and industrial development, was marked by the
emergence of photochemical smog as a threat to the regional environment. In
response, air pollution abatement programs for stationary sources began in the
late 1940's. Control of mobile source emissions commenced in the early
1960's, a response to the discovery of the automobile's role in the smog
formation mechanism. Thus, air quality deterioration in the SCAB has multiple
causes: topography, meteorology, and dense population and economic activity
with correspondingly large emissions.
The SCAB is essentially a coastal plain with connecting valleys and low
86
-------
lying hills bounded by the Pacific Ocean to the south and west and mountain
ranges along the inland perimeter [Southern California Association of Govern-
ments, et al., (1979)]. Elevation varies from slightly above sea level in the
coastal areas to greater than 11,000 feet in the mountainous inland. Intra-
basin transport of air pollutants generally follows inland valley pathways.
The main meteorological characteristics of the South Coast Air Basin are
mild temperatures, limited precipitation, low wind speeds and persistent in-
version layers with low mixing heights. Annual average temperatures range
from the low to mid 60's throughout the basin. Variation in temperature is
much greater in the eastern portion of the basin due to the decreased oceanic
influence. Rainfall amounts vary little throughout the basin and are
generally small, typical of a coastal desert. Sunshine is a critical element
in the formation of photochemical oxidants, and possible sunshine is generally
high. For instance, 73 percent of possible sunshine is recorded annually in
downtown Los Angeles.
Low wind speeds with little seasonal variation are a common occurrence
throughout the basin. An average wind speed of 5.7 miles per hour has been
recorded in downtown Los Angeles over the period 1950 to 1976. The dominant
diurnal wind pattern, broken only by the Santa Ana winds and winter storms, is
a daytime sea breeze and a nighttime land breeze. Horizontal air movement is,
therefore, limited. Vertical dispersion of air pollutants is also limited due
to frequent existence of temperature inversions near the surface.
The topographic and meteorological conditions inherent in the South Coast
Air Basin imply that the region is limited in its ability to disperse
pollutants, both horizontally or vertically. Therefore, pollution emissions
have a relatively large impact upon ambient air quality. The situation is
further exacerbated since the emission of air pollutants is considerable due
to the region's dense population and prosperous economy.
Table 1 presents the air pollution emissions for 1975-76 by major source
category for an average summer weekday in the SCAB. Also included are the
relative percentage contributions by mobile and stationary sources. These
figures represent the baseline emissions for the benefit-cost analysis which
follows; that is, the reductions required to attain the federal primary air
standards are determined from these baseline statistics. As is illustrated,
on-road mobile sources (light duty autos and trucks, medium and heavy duty
trucks, heavy duty diesel trucks, and motorcycles) contribute in excess of 50%
of total emissions for all pollutants except sulfur oxides and particulate.
In these latter categories, stationary sources are the dominant contributors.
Offroad mobile sources (aircraft, railroads, ships, etc.) contribute
negligible amounts in all cases.
87
-------
Table 5.1
BASE YEAA EMISSIONS 1 9 7 5 4 1976
BY MAJOR source category (TONS/DAY)
AVERAGE SUMMER WEEKDAY
SCAB
SOURCE
THC
C
CO
I
50
X
PART
TONS/DAY
* of
* of
ONS/PAY
X of
X of
T On si ~
* of
TONS/DAY
X of
TOMS/OAY
X of
TONS/DAY
* of
Man-Made
Total
*fefft~*ad
TOTAL
TOTAL
TOTAL
TOTAL
TOTAL
STATIOHARY
(ArM ( Point)
676
38.9
23.5
510
3*.5
30. 1
215
2 . 6
464
36.2
313
81.9
1 50
56.2
On RoadNobllt
% 9
55.8
33.9
88k
59 1
52.2
7699
91.2
694
5VI
37
9.7
94
35.2
Off-bad Hob! 18
92
5.3
3.2
64
57
5.0
527
6.2
125
9.1
32
8.4
23
8.6
Subtot* I (Man-Mad*)
1737
100.0
U7i
toa.o
644 1
100.0
1 283
100.0
362
100.0
267
100.0
Natural Sources*
1132
39.5
215
TOTAL
2869
100.0
1693
00.0
8MI
100.0
126]
100.0
382
100.0
267
100.0
• lnc.lude> vegetative, landfills# nd • nimal waste.
Referent.; AQMP
-------
In Table 2 the emissions inventory is disaggregate by county. As is
indicated in the Table, Los Angeles and Orange Counties have a
disproportionate share of total emissions, but this corresponds to their
shares in population and economic activity.
The existing emissions inventory is such that on virtually every day, at
least one of the ffederal air quality standards is violated at some location in
the South Coast Air Basin. For example, the federal oxidant standard (.08
ppm) was exceeded on 252 days in 1976. In addition, the State oxidant first
stage episode level of .20 ppm was violated on 204 days in 1976, with a
maximum reading of .38 ppm. The nitrogen dioxide standard (.05 ppm) was also
consistently violated, with the greatest number of violations occurring in the
densely populated areas of Los Angeles and Orange Counties [Southern
California Association of Governments, et al., (1979)]. Therefore,
significant reductions in existing emissions levels of all pollutants, with
the exception of sulfur oxides are required if the South Coast Air Basin is to
become an attainment region.
It should be noted that reactive hydrocarbons, nitrogen oxides and carbon
monoxide are the pollutants of most importance in the South Coast Air Basin.
Significant reductions of total suspended particulate (TSP) are also required
to meet the corresponding ambient standards. However, total suspended part-
iculate pollution is primarily background [Southern California Association of
Governments et al., (1979]). For this reason, the benefit-cost analysis which
follows concentrates on the required reduction of reactive hydrocarbons,
nitrogen oxides, and carbon monoxide.
In order to determine the emissions reductions to satisfy the federal
standards, one must have knowledge of the relationship between emissions and
air quality. However, this |s an area characterized by substantial
uncertainty and controversy. The estimates used in this analysis are from
the Air Quality Management Plan.
This modelling indicated that reactive hydrocarbon emission of 506 tons/
day, nitrogen oxides emissions of 800 tons/day and carbon monoxide emissions
of 2480 tons/day would allow the federal ambient standards to be satisfied.
Therefore, the baseline emissions of reactive hydrocarbons, nitrogen oxides
and carbon monoxide would have to be decreased by 974 tons/day, 503 tons/day
and 5963 tons/day, respectively [Southern California Association of Govern-
ments, et al., (1979)].
Since, the primary concern of this exercise is the evaluation of on-road
mobile source controls then the proportion of the required reductions in
emissions attributable to these controls was necessary information. This was
89
-------
Table 5.2
197s-76 Emissions-Major Sources by County
SCAB Average Sumner Weekday
Los Angeles Orange Riverside San Bernardino
County County County County
Total Hydrocarbons"
Stationary Manmade
Natural
On-Road Mobile
Off-Road Vehicles
520.3
699.3
686.9
65.1
91.7
250.2
187.1
17.7
21.5
89.8
38.4
3.6
33.0
101.1
54.6
5.2 "
TOTAL
1971.6
546.7
153.3
193.9
Reactive Hydrocarbons
Stationary-Manmade
Natural
On-Road Mobile
Off-Road Vehicles
393.1
91.2
626.7
59.6
69.3
24.6
170.8
16.2
16.2
60.7
35.1
3.3
24.9
43.3
^9.8
4.7
TOTAL
1170.6
280.9
115.3
122.7
Carbon Monoxide
Stationary
On-Road Mobile
Off-Road Vehicles
18.9
5462.2
373.1
9.1
1451.5
9 9.5
23.2
352.1
24.2
164.2
439.0
30.2
TOTAL
5854.2
1560.1
399.5
633.4
Nitrogen Oxides
Stationary
On-Road Mobile
Off-Road Vehicles
347.8
482.2
86.4
32.7
135.5
24.3
6.7
33.0
5 . 9
50.0
42.9
7 . 7
TOTAL
943.4
192.5
45.6
100.6
Sulfur Oxides
Stationary
On-Road Mobile
Off-Road Vehicles
234.2
26.0
22.8
22.8
7.1
6.2
1.7
1.5
55.6
2.3
2.0
TOTAL
283.0
36.1
3.2
59.9
Total Suspended
Particulate
Stationary
On-Road Mobile
Off-Road Vehicles
75.5
65.5
16.2
20.7
18.3
4.5
27.0
4.3
1.1
27.2
5.8
. \A
TOTAL
157.2
43.5
32.4
34.4
Reference: AQMP
90
-------
calculated as follows. First, baseline emissions (1975-1976) were inflated to
reflect the expected growth from the present to 1987, the expected attainment
date. This yielded the emissions levels in the absence of control. The
inflated emissions were divided into the on-road mobile, off-road mobile and
stationary categories assuming that growth in each was proportional to its
existing share of the emissions inventory.
¦4 - •
Second, from these 1987 emissions levels we subtract^}the projected 1987
emissions levels which assume currently mandated controls. The result was
the impact of the control measures in each category. Therefore, on-road
mobile source controls account for .747, .789 and 1.01 of the reduction in
emissions of HC, NO , and CO from the present to 1987. Finally, these
factors were appliel to the required emissions reductions stated above. Thus
in the scenario analyzed here on-road mobile source controls are responsible
for reducing emissions approximately 728 tons/day, 397 tons/day and 6023
tons/day of reactive hydrocarbons, nitrogen oxides and carbon monoxide,
respectively. Off- road mobile, stationary and controls make up the balance
of the control effort designed to attain the Federal Ambient Standards.
The Benefits of Emissions Reductions
The benefits from air quality improvements are derivable from either the
housing value method or the survey approach detailed in the previous chapter.
However, the housing value approach, which allows the derivation of an esti-
mated relationship between pollution abatement benefits and the independent
variables income and initial pollution concentration, is more amenable to this
policy application. For this reason, the housing value approach is the
primary method employed to estimate benefits from the air quality improvement
associated with the stated emissions reductions.
The housing value analysis used is a multi-step procedure:7 (i)
estimation of a hedonic housing value equation which relates home sale price
to a set of home and neighborhood variables; (ii) derivation of marginal
willingness to pay for air quality improvement; (iii) estimation of a marginal
benefit equation which relates marginal willingness to pay to income and
existing pollution levels (i.e., this is the inverse demand curve); and (iv)
mathematical integration of the marginal benefit equation to determine total
household benefits for any stated air quality improvement. This final step is
equivalent to determining the area under the inverse demand relationship. It
is this latter relationship that is used to determine basinwide benefits for
any decrease in pollutant concentrations by applying the household benefits to
the relevant population.
The multi-step nature of the housing value approach produces a resulting
91
-------
benefit equation which is inherently dependent upon previous steps. For
instance, collinearity among the various pollution measures dictated the use
of nitrogen dioxide (NO ) as a proxy for overall pollution. Also, there
existed no significant difference in statistical^perfornMpee in the hedonic
housing equation between NO^ measured as'NO ^"-NO , jarlNf) . However, the
resulting benefit estimates were substantiall'lyaf ffessttedd by the choice of
measurement. Variation in the third procedural step estimation of the
marginal benefit relationship was found not to alter the benefit results
measurably; that is, benefit estimates were essentially invariant to the form
of the estimated relationship (linear-linear, log-log). Therefore, benefits
from air quality improvement are not determined uniquely, rather a range
results are obtained depending upon the particular estimation procedure used.
In total, six estimated benefit equations,3determined by the pollution
variable used in the initial step (NQ . NQj, -Hop and the form of the marginal
benefit equation (linear-linear, log-log), were utilized to calculate house-
hold benefits. The general structure of the benefit equations corresponding
to the linear-linear marginal benefit equations is
HB "Cl'(VV + V(pb-V-y + C3-(PB - $
where
HB = household benefits in dollars
P <¦ initial pollution (NO ) level in pphm
= pollution (N0_) level after proposed improvement in pphm
Y « income in dollars
C ,C ,C = estimated coefficients determined by integration of the appropriate
^ marginal benefit function.
The general benefit equation corresponding to the log-log marginal benefit
equations is
C2 S C3
H e = ci x • pb - pa • c_„
Table 3 presents the estimated coefficients.
In order to demonstrate the use of the benefit equations, consider Figure
3. The figure shows a family of constant benefit curves which indicate all
combinations of income and existing pollution that yield an identical
willingness to pay (dollar amount over the life of the home) to achieve the
ambient standard. As is evident, those individuals with high income and poor
air quality would be willing to pay the most for the stated air quality
92
-------
TABLE 5.3
Benefit Equation Coefficients
Pollution Variable Used in
Hedonic Housing Value Equation
Linear - Linear Marginal
Benefit Function
53.996
.11513
-31.204
-1883.1
.086995
66.605
NO g
-2294.5
.057521
95.145
Log - Log Marginal
Benefit Function
C2
c,
.024134
1 .1983
.69054
.001785
1.1985
1 .69195
.000115
1.1988
2.691
9 3
-------
Figure 5.3
VALUE OF IMPROVED AIR QUALITY ($)
50
45
*0
35
30
25
20
15
10
10,000
7 5 0 0
5000
HB= -1883-(PB-6.9) + .086995 • (PB-6.9).I + 66.605 (p2-47.6)
2500
1000
500
6.9 8 10 12
! ! I I
STANDARD Nq2 (pphm)
Present Pollution Level
94
-------
improvement. Further, an individual with relatively high income and good air
quality would be willing to pay as much as an individual characterized by poor
air quality and low income.
Benefits derivable from moving to the ambient standard can be calculated
given values for income and baseline pollution in all areas. in the context
of the earlier discussion, this constitutes improving the "fair" and "poor"
air quality areas to the "good" category. The fair and poor regions are
assigned values of 9.55 pphm and 12.38 pphm, respectively, as determined by
the sampling procedure outlined in Chapter 3. Therefore, if all regions were
to upgrade to the "good" level, it would involve an approximate 30% improve-
ment in the fair communities and a 45% improvement in the poor air quality
communities.
With respect to income data, two methods were initially utilized. In the
first method each household was allocated the county average income. The
second procedure assumed that the good air quality region was inhabited by an
income group wealthier than average. Thus, on the basis of the survey
responses the good air quality area income was determined and then separated
out from total county income. Each household in the poor and fair air quality
regions was then allocated the average of the remaining income. This second
method, although somewhat lowering average income per household in the poor
and fair communities had little effect on aggregate benefit estimation. Thus ,
results are presented for the first method only.
With all data inputs specified, household benefits are calculated using
the estimated benefit equations. These benefits which accrue over the life of
the home, represent differences in home sale price attributable to variations
in air quality. In order to transform these into annual benefits the standard
annualization procedure is employed (1978 interest rate = .10). Aggregation
is then accomplished for each county by deflating by persons per household and
multiplying by county population. This generalization to the entire county
assumes that the household sample analyzed is representative of the population
at large.
Aggregate benefits associated with achieving the federal air quality
standards in the South Coast Air Basin are presented in Table 4. As is illus-
trated, aggregate benefits range from 1.5 to 3.8 billion dollars annually.
Further, the bulk of the benefits occur in populous Los Angeles and Orange
counties. The upper bound estimate corresponds to the benefit equations
derived from the use of N02in the hedonic housing equation (initial step of
ttfe multi-step procedure) whereas the lower bound corresponds to the use of
N02. The form of the estimated marginal benefit function has no significant
impact on the benefit estimates.
95
-------
Table 5.4
Annualized Benefit Estimates for Achieving the
Federal Ambient Standard (1978 $000)
Functional Form County
Pollution Variable Functional Form San Bernardino Riverside Orange Los Angeles Total
in Hedonic Housing of Marginal
Equation Benefit Function
NO, Linear - Linear 295708 234345 638571 2711045 3879669
L.
NOg Linear - Linear 194455 153284 414687 1776702 2539128
NO; Linear - Linear 118541 93117 245573 1065688 1522919
N02 Log - Log 287443 228393 615769 2611015 3742620
NO^ Log - Log 204302 162331 413405 1789905 2569943
N0| Log - Log 129231 102681 246916 1092596 1571424
-------
For purposes of comparison, the survey approach which accompanied the
property value analysis, yields an aggregate benefit estimate of approximately
1.65 billion dollars annually, whereas a housing value analysis which utilized
total suspended particulate as the proxy variable yields estimates in the 2.2
to 2.7 billion dollar range. Based on this evidence, a narrowing of the
probable range of benefits to 1.6 to 3.0 billion dollars annually seems in
order. • •
Apportionment of these benefit figures between the on-road mobile, off-
road mobile and stationary categories is accomplished through application of
the percentage figures described above. Using the percentage averages over
reactive hydrocarbons, nitrogen oxides and carbon monoxide then on-road mobile
controls are assigned approximately 85% of the benefits. Therefore, the
benefits from air quality improvement associated with on-road mobile controls
range from 1.36 - 2.55 billion dollars annually. Again, the remainder of air
quality improvement benefits are a function of off-road and stationary and
institutional control measures.
Before proceeding to the next section two qualifications should be noted.
First, the benefit calculations are inherently tied to both the air pollution
modeling efforts contained in the Air Quality Management Plan and the esti-
mation procedures outlined in Chapter 3. Second, it should be noted that
these benefit calculations were derived assuming a one year cleanup period.
This essentially static analysis is somgwhat unrealistic given the magnitude
of the air quality problem in the SCAB. A dynamic approach which examined the
benefits resulting from a multi-year clean-up would indicate expanded benefits
due to increased population and economic growth and associated increased
emissions levels. The increased emissions would imply a larger required
emission reduction to satisfy the federal standards and a corresponding larger
benefit per household. The greater population would increase aggregate
basin-wide benefits.
In the next section, dollar per ton removed cost estimates are presented
for on-road mobile pollution control methods. These cost estimates are then
used to determine total clean-up costs for the required emissions reductions.
ESTIMATED COSTS OF AIR QUALITY IMPROVEMENT
Institutional Background
Control of vehicular emissions began in 1961, when the automakers, under
pressure from the California legislature, installed positive crankcase
ventilation (PCV) systems in order to reroute "blowby" fumes back into the
engine intake. These emissions had been discovered two years earlier to
97
-------
account for 20-25 percent of hydrocarbon emissions. Positive crankcase
ventilation was adapted nationwide in 1963 [Mills, et al. (1978), and
Wakefield, (1980)] .
The 1965 amendments to the Clean Air Act directed the secretary of HEW to
set emissions standards for automobiles effective January 1, 1968. The Clean
Air Act was further amended in 1970 setting goals of 90 percent reductions in
emissions from automobiles by 1975-76. The objective of such legislation
seemed reasonable; fewer pollutants, more efficient engines. However, the
attainment of such objectives has been a difficult process.
Emission standards were first set for hydrocarbons (HC) and carbon
monoxide (CO); the 1970 standards were 2.2 grams/mile and 23 grams/mile
respectively (7-mode test). In the early years the control of those
pollutants focused on modification of existing engines. The original
modifications (leaner air-fuel mixtures, retarded ignition timing and higher
coolant temperatures) were relatively unsuccessful, causing associated side
effects (reduced fuel economy and engine response). Later modifications
proved more successful both in combatting pollution and reducing the unwanted
side effects.
California was the first to set a limit on nitrogen oxide (NO ) emissions
-- 4.0 grams/mile for the 1971 year. This was in response to NO §eing
identified as an important element in the formation of photochemical smog.
However, the control of NO introduced an inherent conflict. Hydrocarbon and
carbon monoxide control ha2 been achieved by afterburning through air
injection, delaying spark, leaner mixtures or hotter combustion. Nitrogen
oxide control required reducing temperature since they were a byproduct of
very hot, relatively efficient combustion. This was accomplished by exhaust
gas recirculation (EGR) which had dramatic negative impacts on fuel mileage
and driveability (response to acceleration and performance under constant
speed [Wakefield, (1980)].
Even though the original 1975-76 standards were delayed considerably the
1975 federal emission regulations were so stringent (1.5/15/3.1 grams per mile
of HC/CO/NO using the Constant Volume Sampling - 75 test) as to require a
technological revolution. The catalytic converter was introduced. Since the
catalytic converter was downstream from the engine operation it freed the
engine from earlier modifications. This meant better engine response and fuel
economy from a given engine controlled with a catalyst rather than controlled
without a catalyst.
The first catalytic converters controlled HC and CO leaving NO to be
controlled by conventional means. However, 3-way catalytic converters in
98
-------
which two separate catalyst beds control HC, CO and NO now exist. This
latter innovation, together with advances in electronic monitoring have
allowed control of vehicular emissions while minimizing the effects on
driveability and fuel economy. However, this is sophisticated and costly
technology. It is the costs we turn to next.
The Costs of Emissions Reductions
The estimation of control costs is characterized by controversy and a
large degree of uncertainty. The difficulty in estimation is concentrated
around two central problems. The first is determining the actual cost of any
particular control technique. In many instances, with very little con-
struction experience, the cost of specific control devises is unknown. Also,
marketing strategies affect the direct cost to the consumer. For example, the
cost of California systems which requires larger emission reductions than
their federal counterparts may be spread out among all automobile consumers
rather than those located in California. Further, control techniques
generally imply associated secondary costs and savings which often escape
quantification. These secondary implications can have a significant impact on
the cost of any proposed control option.
The second problem is the determination of actual, rather than alleged,
emissions reductions corresponding to any particular control strategy. For
instance, control strategies may cause synergistic reductions or may negate
each other. In addition, control strategies may be credited with either
overstated or understated emission reductions. The former problem, phantom
decreases in emissions, seems to occur more often in practice.
Therefore, any cost analysis which is not fortified by detailed
engineering cost evaluation and experience is subject to significant error.
This problem is further exacerbated in that estimation errors are generally
not randomly distributed.
Given the background of controversy and substantial uncertainty, the
objective here is to provide a range of cost estimates to be used for com-
parison to the benefit calculations presented in the previous section. The
cost estimates contained herein were derived from a variety of sources,
primarily from Environmental Protection Agency (EPA) publications and auto-
mobile manufacturer statements. All costs are stated in 1978 dollars per ton
removed.
Due to the automobile's substantial role in the South Coast Air Basin air
quality problem (see Tables 1 and 2) mobile source control must be the central
element in any attainment plan. However, control cost figures associated with
99
-------
any level of control vary widely, dependent upon one's assumptions regarding
initial capital cost, size of any mile per gallon benefit or penalty, unleaded
fuel cost differential, etc. In order to offset some of the variation we
standardized the cost estimates by assuming the following: (1) control cost
devices have a lifetime of 50,000 miles; (2) the cost of gasoline is
$1.00/gallon; (3) the unleaded fuel cost differential $.04/gallon [Lloyd,
(1979)]; (4) baseline'mileage is 20 miles per gallon ; (5) maintenance
savings are $25 over the life of the emissions system [Lloyd, (1979)]; (6)
evaporative emissions and altitude control add $15 to initial capital costs;
and (7) the capital costs of going from totally uncontrolled vehicles to the
1977-79 standard of 1.5/15/2.0 of HC/CO/NO is $140 [Lloyd, (1979)], where
uncontrolled vehicles correspond the 19^3 federal standard (Constant Volume
Sampling-72 test) of 3.4/39/3.0. Further, we examined the total cost of
moving from this 1973 level of control to the 1981 federal standard of
.41/3.4/1.0.
Even with this degree of standardization there exists significant vari-
ation in mobile control costs dependent upon the source of information, the
assumed fuel mileage savings or penalty and the assumed allocation of total
costs to hydrocarbon, carbon monoxide and nitrogen oxide control. Tables 5,
6, and 7 present this range of cost estimates for light duty vehicles.
Each of the tables uses the same references to generate total control
costs. The General Motors estimate is based upon initial capital costs of
$460 and a three percent mileage penalty, whereas the EPA estimate is $415 for
initial capital cost with a seven percent mileage improvement. The American
Motors and manufacturer average sticker price estimates for first cost are
$557 and $475 respectively. These are combined with the General Motors and
EPA mileage penalty or saving estimates to obtain two of the estimates
presented. The third estimate assumes an eight percent mileage penalty
[California Air Resources Board, (1979)]. Cost effectiveness is then
determined by dividing the total cost per car by the emission reduction over
50,000 miles.
In Table 5 the lower bound figure of the range for each
cost-effectiveness estimate is based upon an allocation of 30.4 percent of the
total control cost to hydrocarbon measures [Schwing, et al., (1980)]. The
upper bound figures assume one-third of total cost is allocated to hydrocarbon
control. In Table 7, the upper bound figures for each estimate correspond to
an allocation of .33 and .362 to nitrogen oxides control [Schwing, et al. ,
(1980)]. The figures in Table 6 assume one-third of total cost is allocated
to carbon monoxide control.
As is illustrated in the tables, the cost effectiveness range ($/ton
100
-------
Table 5.5
Hydrocarbon Cost Estimates for Mobile Source Control
($1978/ton removed)
Control Category
cost
Reference
Light Duty Vehicles
$ 160- 257
620- 680
880- 965
1340- 470
1610- 770
730- 800
1190-1310
1460-1600
General Motors (GM)'
EPA2
3
American Motors , EPA
Mileage
American Motors, GM
M i 1 cage
American Motors, Eight
Percent Mileage
Penalty
Manufacturer Average4,
EPA Mileage
Manufacturer Average,
GM Mileage
Manufacturer Average,
Eight Percent
Mileage Penalty
Heavy Duty Vehicles
3400-3450
3720-3770
AQMP*
EPA
Inspection and Maintenance
1410-1590
AQMP
References: 1. General Motors Corporation, "Estimated Effects of Exhaust
Emission Standards on Potential Hardware, Fuel Economy,
Fuel Consumption and Additional First Cost to Consumer,"
May 1979.
2. Lloyd, Kenneth H., Cost and Economic Input Assessment for
Alternative Levels of the National Ambient Air Quality
Standard for Ozone, USEPA, February 1979.
101
-------
3. American Motors Corporation Cost Information contained in
"Automobile Emission Control - The Development, Status,
Trends and Outlook as of December 1976, " USEPA, April,
1977.
A. California Air Resources Board, "Status Report on the Need
for Land Feasibility of a 0.4N0 standard for Light Duty
Motor Vehicles, December 1979. X
5, Southern California Association of Governments and South
Coast Air Quality Management District, Draft Air Quality
Management Plan, January, 1979.
102
-------
Table 5.6
Carbon Monoxide Cost Estimates for Mobile Source Control
($1978/ton removed)
Control Category cost Reference
Light Duty Vehicles
160
General Motors (GM)'
85
EPA*
120
American Motors^, EPA
M i 1 cage
184
American Motors, GM
Mi 1eage
220
American Motors, Eight
Percent Mileage
Penalty
100
Manufacturer Average4,
EPA Mileage
163
Manufacturer Average,
GM Mileage
200
Manufacturer Average,
Eight Percent
Mileage Penalty
Heavy Duty Vehicles
290-310
AQ.MP5
320-340
EPA
Inspection and Maintenance
175-195
AQMP (Revised)
References: 1. General Motors Corporation, "Estimated Effects of Exhaust
Emission Standards on Potential Hardware, Fuel Economy,
Fuel Consumption and Additional First Cost to Consumer,"
May 1979.
2. Lloyd, Kenneth H., Cost and Economic Input Assessment for
Alternative Levels of the National Ambient Air Quality
Standard for Ozone, USEPA, February 1979-
103
-------
3, American Motors Corporation Cost Information contained in
"Automobile Emission Control - The Development, Status,
Trends and Outlook as of December 1976," USEPA, April,
1977.
4, California Air Resources Board, "Status Report on the Need
for Land Feasibility of a 0.4 NO standard for Light Duty
Vehicles, December 1979. x
5, Southern California Association of Governments and South
Coast Air Quality Management District, Draft Air Quality
Management Plan, January, 1979.
104
-------
Table 5.7
Oxides of Nitrogen Cost Estimates for Mobile Source Control
($1978/ton removed)
Control Category
cost
Reference
Light Duty Vehicles
1910-2070
1010-1100
1440-1570
2200-2390
2640-2870
1195-1300
1950-2120
2390-2600
General Motors (GM)'
EPA2
3
American Motors , EPA
Mileage
American Motors, GM
Mileage
American Motors, Eight
Percent Mileage
Pena 1 ty
Manufacturer Average4,
EPA Mileage
Manufacturer Average,
GM
Manufacturer Average,
Eight Percent
Mileage Penalty
Heavy Duty Vehicles
2020-2120
2210-2320
AQMP"
EPA
Inspection and Maintenance
1310-1600
AQMP
References: 1. General Motors Corporation, "Estimated Effects of Exhaust
Emission Standards on Potential Hardware, Fuel Economy,
Fuel Consumption and Additional First Cost to Consumer,"
May 1979.
105
-------
2. Lloyd, Kenneth H., Cost and Economic Input Assessment for
Alternative Levels of the National Ambient Air Quality
Standard for Ozone, USEPA, February 1979.
3, American Motors Corporation Cost Information contained in
"Automobile Emission Control - The Development, Status,
Trends and Outlook as of December 1976," USEPA, April,
'1977.
4. California Air Resources Board, "Status Report on the Need
for Land Feasibility of a 0.4 NO standard for Light Duty
Motor Vehicles, December 1979. x
5, Southern California Association of Governments and South
Coast Air Quality Management District, Draft Air Quality
Management Plan, January, 1979.
106
-------
removed) for hydrocarbon control is approximately $600 - $1800 while carbon
monoxide control is $80-$200 and NO control is $1000-$2600. The predominant
source of this wide variation is the assumption concerning fuel use over the
50, 000 mile life of the control device.
The cost effectiveness of heavy duty vehicle emissions control were
calculated in a manner similar to that described above for light duty
vehicles. In this instance total cost per vehicle figures published in the
Air Quality Management Plan (January, 1979) and an EPA report [Lloyd, (1979)],
were utilized. The former reference was also the source for the corresponding
emissions reductions. Total vehicle cost includes all capital costs and costs
for the associated inspection and maintenance program. Cost effectiveness
estimates for heavy duty vehicles are presented in Tables 5, 6, and 7 for the
pollutants HC, CO, NO , respectively. Again, the range of costs is dependent
upon the allocation method used; either one-third to each pollutant or .329 to
HC , .354 to CO and .317 to NO , [Schwing, et al., (1980)] .
X
The final component of on-road mobile source control is the light duty
vehicle inspection and maintenance program. The importance of this program
cannot be understated for without it, auto owners have no incentive to
maintain the performance of their emission control systems. Furthermore, the
lack of performance invalidates the cost-effectiveness figures presented above
which assume that the control devices work as designed. For example, if
control mechanisms on light duty vehicles deteriorate linearly over 50,000
miles from their designed operations levels then the cost effectiveness of
such mechanisms doubles. This situation is worsened if systems deteriorate
more quickly or are tampered with. The success of any control system is
therefore inherently dependent on an effective inspection and maintenance
program.
The annual cost of the program is the sum of the inspection fee
multiplied by number of automobiles plus the cost of repairing failed auto-
mobiles. The air quality management plan assumes a $9 inspection fee, and a
35 percent failure rate with associated $23 repair cost. However^recent
evidence shows that the failure rate may be closer to 42 perceratt:.. The cost
calculations contained in Tables 5, 6, and 7 assume this latter figure with a
corresponding repair cost range of $20 - $25. Emissions reductions associated
with the inspection and maintenance program were obtained from the Air Quality
Management Plan (January, 1979). Allocation of total cost among the
pollutants was based on either one-third to each pollutant or the proportions
used for light duty vehicles [Schwing, et al., (1980)].
Any control strategy devised to meet the ambient standard would use a
variety of control options, each with an associated cost effectiveness.
107
-------
Therefore, the cost effectiveness figures for light duty vehicles, heavy duty
vehicles and the inspection and maintenance program form the basis for the
derivation of aggregate cost estimates to achieve the federal ambient
standards.
It is not the objective of this exercise to cost out a specific air
quality improvement program, but rather to develop a range of costs for
comparison to benefits. This can be accomplished by an examination of an
upper and lower bound for costs. In either case, a weighted average of light
and heavy duty vehicle costs and the inspection and maintenance costs is
utilized, where light duty vehicle costs are the dominant component.
As was stated in the previous section on-road mobile controls account for
approximately 85 percent of the required emissions reductions. This
translates into 728 tons/day of HC, 6023 tons/day of CO and 392 tons/day of
NO . In order to estimate total control costs these emissions reductions are
further apportioned into the light duty vehicle, heavy duty vehicle and
inspection and maintenance categories. Reductions associated with inspection
and maintenance are determined directly from the Air Quality Management Plan
(January, 1979). The shares corresponding to light duty vehicles and heavy
duty vehicles are determined by their relative shares in annual vehicle sales.
Therefore, three percent of the required reductions minus the effect of
inspection and maintenance are allocated to heavy duty vehicles with the
remainder to light duty vehicles.
A lower bound total cost estimate would correspond to the EPA capital
cost, a seven percent mileage improvement and one-third allocation to each
pollutant. In this case total cleanup costs for on-road mobile controls would
be approximately .61 billion dollars. Conversely, an upper bound estimate
would be 1.32 billion dollars. This latter estimate would utilize American
Motors capital costs, an eight percent mileage penalty and one-third
allocation to each pollutant. Thus, the total cost of using on-road mobile
controls to achieve the above stated pollution reductions range from approx-
imately .61 to 1.32 billion dollars. A best estimate (manufacturer average
capital cost, three percent mileage penalty) would be 1.02 billion dollars.
Before proceeding to the concluding section it should be re-emphasized
that these cost figures are subject to a great deal of uncertainty. There
could be significant error in the estimates. It should also be noted that, as
in the case of the benefit estimate, this is an essentially static analysis.
In a dynamic context, one would expect the costs to increase significantly as
a result of larger emission reductions necessitated by expanded population and
economic activity. The costs would likely increase non-linearly as more
costly control measures were employed to achieve the required reductions.
108
-------
This latter aspect exists because many of the easy technological fixes have
already been made.
COMPARISON OF BENEFITS TO COSTS - CONCLUDING REMARKS
There has been much discussion of the desirability of achieving the
federal air quality standards. This study constitutes an attempt to evaluate
a portion of these standards in the South Coast Air Basin of Southern
California from an economic or benefit-cost perspective. Based upon modeling
contained in the Air Quality Management Plan, achievement of the ambient
standards in 1979 would require emission reductions of the 974 tons/day, 5963
tons/day and 503 tons/day of reactive hydrocarbons, carbon monoxide and
nitrogen oxides. It is the share of these emission reductions attributable to
on-road mobile source control which was evaluated using benefit-cost analysis.
Benefits were calculated through an examination of housing value dif-
ferentials attributed to air quality. Achieving the ambient air quality
standards was consistent with improving the "fair" and "poor" air quality
regions to the "good" category as specified in the previous chapter. In
effect, this constituted an approximate 30 percent improvement in the fair
areas and a 45 percent improvement in the poor air quality areas. Correspond-
ing benefits were estimated to fall between 1.6 and 3.0 billion dollars per
year, independent of any benefits accruing to agriculture and ecosystems. The
share of these benefits associated with on-road "mobile source control was
estimated to be 1.36-2.55 billion dollars.
Cost estimates were developed from existing data sources, primarily from
manufacturer statements and government publications. Given the variation in
control cost options and the uncertain nature of the cost figures, it was
found that on-road mobile source control consistent with a policy to achieve
the ambient standards in 1979 would involve a cost of between ,61 and 1.32
billion dollars, with a best estimate of 1.02 billion dollars.
It seems then, that the benefits from on-road mobile emissions reductions
consistent with satisfying the ambient standards are of the same order of
magnitude as the cost estimates. This implies that the ambient air quality
standards are not without some economic justification, though the uncertainty
concerning the benefit and cost calculations prevents one from accepting the
controls outright. However, on-road mobile controls consistent with the air
quality standards cannot be rejected as economically inefficient either.
Therefore, although the mid-range benefit estimate exceeds the mid-range
cost estimate, the situation is best characterized as highly uncertain.
Further, the static analysis performed herein does not answer significant
109
-------
questions concerning the behavior of the benefit and cost functions over time.
Stronger statements could only be made in the context of a much more detailed
analysis supported by a solid cost data base.
110
-------
Appendix 1
Air Quality Modelling in the
Air Quality Management Plan
The principal modeling procedure utilized in the Air Quality Management
Plan is proportional rollback. This method is based on the assumption that
atmospheric concentrations of the contaminant are in direct proportion to
emissions. Mathematically, the proportional rollback method can be expressed
as:
Baseline Emissions Baseline Air Quality
Emissions Objective Air Quality Objective .
The emissions level consistent with the federal standards (objective) can then
be determined with knowledge of the other three components. The procedure was
employed to calculate the required reductions of carbon monoxide. a somewhat
more sophisticated rollback method was used for total suspended particulate
[Southern California Association of Governments and South Coast Air Quality
Management District, (1979)]. The rollback method provides an accurate
assessment of emissions reductions required in cases where the contaminant is
emitted uniformly over the region and there are only limited atmospheric
reactions among pollutants. Accuracy is severely curtailed when these
conditions are not satisfied. In the South Coast Air Basin, where pollutants
are emitted nonuniformly with nonuniform distribution and photochemical
oxidants are the primary problem, the linear rollback method is of limited
usefulness. Therefore, ozone production was modeled in the AQMP using the
Empirical Kinetic Modeling Approach (EKMA).
The EKMA Method is a mathematical model which generates a set of atmo-
spheric ozone concentration isopleths as a function of early morning concen-
trations of hydrocarbons and nitrogen oxides [Mikolowsky, et al, (1974) and
Southern California Association of Governments and South Coast Air Quality
Management District, (1979)]. Figure Al illustrates the inherent nature of
the ozone isopleths (curves of equal concentration) . The curvature of the
isopleths indicates that a control strategy which reduced only one of the
pollutants -- reactive hydrocarbons or nitrogen oxides -- could conceivably
worsen rather than improve the situation. The proper control strategy would,
therefore, require that both pollutants be reduced simultaneously.
Ill
-------
Figure 5.A1
OX
Ox
ox
Initial Reactive Hydrocarbon Concentration
OZONE ISOPLETHS
-------
REFERENCES
1 Note, for example, that the average daily maximum concentration of NO in
"good" air quality communities is .069 ppm where the ambient standard2
required an average concentration of .05 ppm.
2 In addition to the difficulty in obtaining accurate cost data on
stationary and institutional controls the decision to focus on on-road
mobile source control was a function of its relative share of both
existing pollution and the future clean-up as envisioned in the
Air Quality Management Plan (January, 1979) .
3 The area description follows closely the Air Quality Management Plan
(January, 1979) .
4 A brief discussion of air quality modelling is contained in the appendix
to this section.
5 Air Quality Management Plan (January, 1979) .
6 The share of the reduction in CO attributable to on-road mobile sources
estimated to be in excess of 1.0 indicated an increase in CO emissions
from off-road mobile sources and neither an improvement nor deterioration
from stationary sources.
7 See Harrison and Rubinfeld (1978) for a detailed description of the
methodology.
8 Although the static approach is somewhat unrealistic it was chosen since
there was insufficient data on costs and the dynamics of pollution
emissions, population, etc. to support analyzing a particular attainment
plan.
9 Blowby is the collection of combustion gases that slip past the piston
rings from the combustion chamber into the crankcase. These fumes were
vented to the atmosphere to prevent contamination and thinning of
113
-------
crankcase oil [Mills, et al. (1978) and Wakefield (1980)].
10 Personal Communication with Dr. Richard Perrine, UCLA.
11 The 20 miles per gallon assumption corresponds to the CAFE mileage
standards for- 1980 on a sales weighed basis. Further, these standards
are front loaded up to 27.5 MPG in 1985. Thus, using 20 MPG overstates
the lifetime fuel cost differential and the mileage penalty.
12 Even though the 1973 federal standard was chosen as the level of un-
controlled emissions, the 1973 levels represent approximately 61%, 55%
and 25% control over truely uncontrolled emissions of HC, CO and NO
respectively. The 1973 level was chosen to be conservative (overstate)
in the cost effectiveness of emission control devices.
13 Personal communication with Dr. Richard Perrine, UCLA.
114
-------
BIBLIOGRAPHY
Brookshire, David S., d'Arge, Ralph C., Schulze, William D., and Thayer,
Mark A., "Experiments in Valuing Non-Market Goods: a Case Study
of Alternative Benefit Measures of Air Pollution Control in the South
Coast Air Basin of Southern California," Methods Development for
Assessing Tradeoffs in Environmental Management, Volume II, USEPA
(October, 1978) .
California Air Resources Board, "Status Report on the Need for/and Feasibility
of a 0.4 NO Standard for Light Duty Motor Vehicles," (December, 1979).
X
Emission Control Technology Division, "Automobile Emission Control - The
Development Status and Outlook as of December 1976," Office of Air
and Waste Management, USEPA (April 1977).
General Motors Corporation, "Estimated Effects of Exhaust Emission Standards
on Potential Hardware, Fuel Economy, Fuel Consumption, and Additional
First Cost to Consumer," Unpublished, (May 1979).
Harrision, D., Jr., and Rubinfeld, D.L., "Hedonic Housing Prices and the Demand
for Clean Air," Journal of Environmental Economics and Management, (May
1978) .
Herfindahl, O.C., and Kneese, A.V. , Economic Theory of Natural Resources
Resources for the Future, Inc., Charles E. Merrill Publishing Company,
(1974) .
Lloyd, Kenneth H., Cost and Economic Impact Assessment for Alternative Levels
of the National Ambient Air Quality Standard for Ozone, USEPA, (February
1979) .
Mikolowsky, W.T., Gebran, J.R., Stanley, W.L., and Buckholz, G.M., The Regional
Impacts of Near-Term Transportation Alternatives: A Case Study of
Los Angeles, prepared for the Southern California Association of
Governments, Rand Corporation, R-1524-SCAG (June 1974) .
115
-------
Hills, E.S., and White, L., "Auto Emissions: Why Regulation Hasn't Worked,"
Technology Review, (May/April, 1978).
Southern California Association of Governments and South Coast Air Quality
Management District, Air Quality Management Plan, (January, 1979) .
Schwing, R.C., Bra'dford, W.S., Von Buseck, C.R., and Jackson, C.J., "Benefit-
Cost Analysis of Automotive Emission Reductions," Journal of Environmental
Economics and Management, (March, 1980).
Wakefield, R., "The Regulated Automobile, Part 1: Emission and Noise Reg-
ulations," Road and Track, (April, 1980).
116
-------
CHAPTER 6
EFFECTS OF AIR POLLUTION AND OTHER ENVIRONMENTAL
VARIABLES ON OFFERED WAGES
INTRODUCTION
Much of the recent interest in the econometric estimation of labor supply
models using individual or micro data has been stimulated by important policy
questions such as the role of women in the labor force and the advisability of
negative income tax programs. Frequently, these models have consisted of two
interrelated equations that explain: (1) how an individual's offered wage
rate is determined and (2) how this wage rate together with other factors
affects the amount of time an individual chooses to work. Effects on wages
and hours in response to changes in exogenous variables including the actual
negative income tax rate faced or the number of pre-school children in the
home can then be estimated through this framework. This general approach can
be easily extended to make parallel estimates of the labor market effects of
changes in environmental amenity levels. Such extensions would have obvious
policy relevance in that the extent of reduced productivity due, for example,
to air pollution could then be assessed.
The purpose of this report is to construct some exploratory estimates of
the effect of changes in air pollution levels on offered wage rates.
Repercussions on the work time choice are not explicitly considered.
Specifically, hedonic equations are estimated that allow for an individual's
offered wage rate to be determined by his own labor supply characteristics
together with measures of amenity levels in the community in which he lives.
In this type of analysis, supply characteristic's such as education, work
experience, and health status are frequently used exclusively to explain the
variation in the offered wage. This specification carries the restrictive
implicit assumption that the demand schedule for classes of individuals
possessing identical values of these independent variables is infinitely
elastic. That is, observed differences in individual wage rates are
attributed only to supply characteristics. In order to circumvent this
limitation, Nakamura, Nakamura, and Cullen (NNK) (1979), have suggested the
inclusion of work environment variables such as the local unemployment rate
117
-------
and a local job opportunities index as additional regressors. These work
environment variables, obviously) capture the fact that local labor demand
conditions may influence offered wages after adjusting for the effect of
individual labor supply characteristics. However, as recognized by other
investigators, variables measuring working conditions and job related hazards
(Lucas 1977, Hamermesh 197 7, Thaler and Rosen 197 5, Viscusi 197 8, and Brown
1980), social infrastructure (Nordhaus and Tobin 1972, and Meyer and Leone
1977), as well as environmental amenities (Hoch 1977, Rosen 1979, and Cropper
1979) can also play an important role, in explaining the behavior of wage
rates. For example, in the case of environmental amenities, 1 if a community is
located in an area that is subject to extreme temperatures or unusually high
air pollution levels, employers may find it necessary to pay their workers a
premium in order to induce them to remain there.
SPECIFICATION AND THE DATA USED IN ESTIMATION
The general form of the offered wage rate equation to be considered here
is then
WAGE = f(P,W) (1)
where WAGE denotes the offered wage rate paid, P denotes a vector of personal
labor supply characteristics, and W denotes a vector of work environment
characteristics. Moreover, the vector P is assumed to contain measures of:
(1) whether the individual is a union member (uNON), to an individual working
400 hours or less had that individual have chosen to work, for example, full
time. An excellent survey of the sample selection problem as it relates
hedonic wage and labor supply estimates is contained in the recent paper by
Wales and Woodland (1980) .
The exact specification of the wage equation used in the present study is
shown in Equation (2).
Ln(RWGH) = f(UNON, HVET, FMSZ, HLTH, EDC2, EDC3, TOJ2, WARM,
JACR, COLD, HUMD, SOXM, TSPM, NOXM, P**2, S0XM**2, (2)
N**2, CONSTANT).
In Equation (2), the function f is linear in the parameters and RWGH denotes
the real wage. Also , note that the squares of the levels of the three
pollution variables are included as regressors in order to allow for possible
nonlinearities in the way that air pollution affects the real wage. This
equation was estimated by ordinary least squares for both the complete sample
of 1395 observations and for selected partitions of this sample constructed on
the basis of age (AGEH), race (RACE), sex (SEXH), and occupation (OCCP). In
particular, there were three age categories (1729, 3049, 5069), two race
118
-------
categories, (white, nonwhite), two sex categories (male, female), and two
occupation categories (white collar, blue collar) . The total number of
possible partitioned regressions was therefore 24(3x2x2x2). However, not all
of these possible regressions were actually estimated because fjsr certain
partitions the number of available observations was insufficient.
Before turning to a discussion of the results of these regressions, two
additional points should be made regarding the pollution variables. First, as
previously indicated, observations on these variables were not available for
each of the 669 counties of residence for families (2) whether the individual
is a veteran (HVET), (3) the size of the individual's family (FMSZ), (4) the
individual's health status (HLTH), (5) the individual's prior educational
achievement (EDC2,EDC3), and (6) the length of time the individual has spent
on his present job (T0J2), Next, W contains measures of: (1) mean January
and July temperature in the individual's area of residence (COLD, WARM), (2)
the job accident rate in the industry where the individual works (JACR), (3)
average rainfall in the individual's area of residence, and (4) levels of the
air pollutants sulfur dioxide (SOXM), total suspended particulate (TSPM), and
nitrogen dioxide (NOXM).
Unfortunately, this formulation may be subject to a specification error
of unknown severity resulting from the omission of relevant explanatory
variables. While the personal labor supply characteristics are fairly
standard for analyses of this type, biased coefficient estimates may result
from the exclusion of still other relevant work environment variables. That
is, climate, job hazards, and air pollution do not exhaust the list of
potential amenities that may affect the offered wage rate. (For good surveys
of the role other variables may play, see Brown (1980) and Rosen (1977).)
Proximity to recreational opportunities and the amount of local social
infrastructure are but two examples of work environment variables that could
in principle be measured and included. Also, the more labor market specific
variables used by NNK have been excluded from consideration here. Due to
budgetary and time constraints, no efforts were made to collect observations
on these potentially relevant variables. The variables used to explain vari-
ations in the offered wage rate were simply chosen from those that had been
collected previously by the Resource and Environmental Economics Laboratory at
the University of Wyoming for use on other research projects.
More specifically, the basic data set used to estimate the wage equation
consisted of observations drawn from the Panel Study of Income Dynamics (PSID)
for the 1971 interview year. In total, there are observations for household
heads on variables that can be used to construct a measure of their real
wages, together with measures of the variables in the P vector defined
previously in Equation (1). The exact definitions of all of these variables
as well as their numerical codes used on the PSID tapes are provided in Table
119
-------
1 entitled Variable Definitions. Table 1 also gives definitions of the vari-
ables appearing in the vector W. For the 1971 interview year, the PSID data
gives the household's state and county of residence and two digit SIC industry
of employment. Consequently, data were collected on COLD, WARM, KUMD, SOXM,
NOXM, and TSPM by county and then were matched to the individual observations
obtained from the PSID.
For the variables COLD, WARM, AND HUMD, this matching process was quite
simple and requires no further elaboration. However, the matching of the air
pollution variables to counties should be explained in greater detail. The
matching process was begun by listing each of the 669 counties in the 50
states where PSID families lived during 1970. Outdoor air pollution monitoring
data existed for at least one of the three measures SOXM, NOXM, AND TSPM for
247 of these counties. In cases, where data from only one monitoring station
in the county were available, those data were automatically assigned to all
PSID families residing there. On the other hand, where data were available
from multiple monitoring stations in the county, data from the single station
that had operated for the greatest portion of the nine year period 19671975
were selected. The monitoring stations selected using this rule tended to be
at central city locations. Finally, since no pollution data were available
for 422 counties (699247), values were assigned to the air quality variables
for these counties using one of two procedures for handling missing
observations that will be described momentarily.
For the purpose of estimating the hedonic wage equation, the data set was
reduced from the roughly 3300 possible observations to 1395 observations after
excluding all housholds where: (1) any family member received transfer
income, (2) the head's annual hours of working for money was less than 400
hours. The first of these exclusions was made in order to reduce the
statistical problem created by families that may be facing nonconvex budget
constraints while the second was made in order to eliminate casual workers,
who may be out of equilibrium because their asking wage may exceed offered
wage, from the sample. Curiously, after making these two exclusions, there
were jio families remaining in the sample where the head: (1) received income
from overtime, bonuses or commissions, or (2) was self employed.
The restricted sample used here is quite similar to that used by Wales
and Woodland (1976, 1977, 1978) in their numerous papers on the empirical
determinants of labor supply using PSID data. However, by excluding household
heads who worked less than 400 hours, the estimates reported in the next
section are subject to sample selection bias, a problem dicussed at length by
Heckoan (1976, 1979). Essentially, Heckman contends that the estimates
resulting from such a sample do not apply to the general population. Instead,
they apply only to those in the population having the same characteristics of
those in the sample. In short, the estimates say little about the wage rate
120
-------
DISPLAY 1
VARIABLE DEFINITIONS*
A. PECUNIARY VARIABLES
HOURS = (1839) (head's annual hours working for money)
AWGH = (1897) (head's money income from labor)
WAGH = 0 if HOURS = 0, otherwise WAGH = AWGH/HOURS
BDAL = Index of comparative living costs for a four person family for various
areas as published by Bureau of Labor Statistics in the Spring 1967
issue of Three Standards of Living for an Urban Family of Four Persons.
The lowest living standard was used. This index is published for the
39 largest SMSAS and by region for other SMSAS.
RWGH = WAGH/BDAL
B. SUPPLY CHARACTERISTIC VARIABLES
HLTH = 1 if (2121) = 1 or 3 or if (2122) = 1 or 3 or both.
= 0 otherwise (If HLTH = 1, there are limitations on amount
or kind of work that the head can do)
UNON = 1 if (2145) ¦ 1, zero otherwise (Head belongs to a labor union
if UNON = 1)
EDC1 = 1 if (2197) * 0, 2, 3, or 9 zero otherwise (If EDC1 = 1, head"
has completed grades
08 or has trouble
reading.)
(If EDC2 = 1, head has
completed grades 912 +
possible nonacademic
training.)
(If EDC3 = 1, head has
completed at least some
college.)
= 1, head is a veteran.)
(head's length of time
on present job is 3 years
or less if TOJI = 1)
(head's length of time
on present job is longer
than 3 years if T0J2 = 1)
EDC2 = 1 if (2197) = 3, 4, or 5 zero otherwise
EDC3 = 1 if (2197) = 6, 7, or 8 zero otherwise
HVET = 1 if (2199) = 1 zero otherwise (If HVET
FMSZ = (1868) (Family size in 1971)
TOJI = I if (1987) = 1, 2, or 3 zero otherwise
toj2 = I if (1987) =4, 5, or 6 zero otherwise
121
-------
Variable Definitions (Continued)
C. WORK ENVIRONMENT VARIABLES
WARM = Mean annual July temperature in the county of residence in 1970 in
F" xlO.O. These data are from the U.S. Bureau of Census, County and
City Data-Book, 1971.
COLD = Mean annual January temperature in the county of residence in 1970 in
F" xlO.O. These data are from U.S. Bureau of Census, County and City
Data Book, 1971.
JACR = Number of disabling work injuries in 1970 for each million employee
hours worked by 2 and 3 digit SIC code. The data were obtained from
Table 163 of Bureau of Labor Statistics, Handbook of Labor Statistics,
1973, Bulletin 1735, U.S. Department of Labor, Washington, DC., tISGPO,
1972.
SOXM = Annual 24 hour geometric mean sulfur dioxide micrograms per cubic meter
as measured by the Gas Bubbler Pararosaniline Sulfuric Acid Method.
These data were obtained from the annual USEPA publication, Air Quality
Data Annual Statistics, and refer to a monitoring station in the
county of residence for 1970.
HUMD = Mean annual precipitation in inches x 100.0. These data are taken
from the U.S. Bureau of Census, County and City Data Book, 1971.
NOXM = Annual 24 hour geometric mean nitrogen dioxide in micrograms per cubic
meter as measured by the Salznan Method. These data were obtained from
the annual USEPA publication, Air Quality Data Annual Statistics and
refer to a monitoring station in the county for residence for 1975.
TSPM = Annual 24 hour geometric mean total suspended particulate in micrograms
per cubic meter as measured by the HiVol Gravimetric Method. These data
were obtained from the annual USEPA publication, Air Quality Data
Annual Statistics and refer to a monitoring station in the county for
residence for 1975.
SOXM** = S0XM2
P**2 « TSPM2
N**2 = NOXM2
D. PARTITIONING VARIABLES
AGE = (1972) (head's age in years)
OCCP = 1 if (1984) = 1, 2, 4, or 5 otherwise = 0 (head is a white collar
worker if OCCP = 1 and,
a blue collar worker if
OCCP = o)
SEX = 1 if (1943) = 1 otherwise = 0 (head is male if SEX = 1)
RACE a 1 if (2202) = 1 zero otherwise (If RACE = 1, head is white.)
122
-------
Variable Definitions (Continued)
E. AUXILIARY VARIABLES
REG1 = 1 if (2284) = 1 otherwise = O (head lives in a northeastern
state if REG1 = 1)
REG2 3 1 if (.2284) = 2 otherwise = 0 (head lives in a tiorthcentral
state if REG2 = 1)
REG3 ¦ 1 if (2284) = 3 otherwise = 0 (head lives in a southern
state if REG3 = 1)
REG4 = 1 if (2284) = 4 otherwise = 0 (head lives in a western
state if REG4 = 1)
PRX1 * 1 if (2210) * 1 zero otherwise (If PRX1 = 1, head's dwelling
unit is within 5 miles of center
of city of 50,000 or more.)
PRX2 = 1 if (2210) = 2 zero otherwise (If PRX2 - 1, head's dwelling
unit is between 514.9 miles of
city center.)
PKX3 = 1 if (2210) = 3 zero otherwise (If PRX3 = 1, head's dwelling
unit is between 1529.9 miles of
city center.)
PRX4 » 1 if (2210) = 4 zero otherwise (If PRX4 = 1, head's dwelling
unit is between 3049.9 miles
from city center.)
PRX5 = 1 if (2210) = 5 zero otherwise (If FRX5 = 1, head's dwelling
is greater than 50 miles from
city center.)
AVGT * Average annual temperature for counties in degrees centigrade for
1970.
^Variable numbers from the psiD tape code book are given for the data collected
from the PSID interviews. For the remaining data, no variable numbers are given.
123
-------
that would be paid in. the PSID data set. In these cases, the missing
observations were either replaced by the means of the observed values for the
pollutants or estimated using atechnique suggested by Dagenais (1973). A
brief discussion of the replacement with means method is outlined in Maddala
(1977) . The Dagenais procedure involves running a regression of each
pollution variable on: (1) all remaining (nonpollution) explanatory variables
in Equation (2) , and-. (2) relevant auxiliary variables that may be selected
and then pr|dieting the values of the missing observations from these
regressions. Predicting equations for each of the three pollutants are shown
in Tables 21, 22, and 23. As shown in these tables, the auxiliary variables
used are dummies relating to the distance of a family's residence from a city
center (PRX1, PRX2, PRX3, PEX4, PRX5), the region of the country where the
family lives (REG1, REG2, REG3, REG4) and a measure of the averag^ temperature
in the family's county of residence (AVGT). Unfortunately, the R 1 s for these
regressions ranged from .33 for NOXM to .37 for TSPM to .54 for SOXM
indicating that their forecasting power may not be particularly high. An
alternative to either the replacement with means or the Dagenais' procedures
would be to restrict the sample to only those observations where actual
measurements were available on all variables, including the pollutants. Even
though this restriction reduces the available data settoll2 observation g. it
was employed in the estimation of one equation for illustrative purposes.
A further problem with the SOXM data is that they were obtained using the
Gas Bubbler PararosanilineSulfuric Acid Method. This method has been shown to
result in estimates of SO levels that are biased downward. Mathtecti,
2
however, has supplied a conversion equation that corrects for the bias m the
original data. That conversion equation is given below.
CSOX = 10.625 + 1.97269(SOXM) -0.10891 [SOXM .AVGT] (3)
where CSOX is the converted sulfur dioxide measure. In estimating Equation
(2), CSOX was substituted in place of SOXM, and its square, 'CSOX = S**2 was
used in place of S0XM**2.
EMPIRICAL RESULTS
As previously indicated, three basic versions of Equation (2) were
estimated where: (1) the restricted sample of 112 observations was employed,
(2) the Dagenais procedure was used to construct the pollutants, and (3) the
replacement with means procedure was used. All regressions were estimated by
OLS.
Table 2 reports the results from estimation with the restricted data set.
In this equation, all of the supply characteristic variables are significant
at the 1 percent level except HLTH and T0J2. However, the work environment
124
-------
variables are all insignificant at conventional levels. in fact, the t-
statistics on the pollution variables in no case exceed 1.1 in absolute value.
Using the replacement with means procedure, the quality of the estimated
coefficients improves considerably. These results are shown in Table 3. With
the increase in the number of observations employed from 112 to 1395, all of
the supply characteristic variables turn out to be significant at the 1
percent level and.have the correct sign. Differences in data sets and in
equation specifications make it difficult to directly compare these results to
those obtained in previous studies. Nevertheless, their general pattern of
the estimates presented in Table 3 corresponds closely to those obtained by-
other investigators.
The estimates of the coefficients on the work environment variables also
tend to be more highly significant and are more plausibly signed than in the
case where the restricted sample of 112 observations is used. Also, they are
generally consistent with the findings of other investigators. As indicated
in Table 3, the variables WARM and COLD enter with a significant negative
sign. In the case of WARM, the negative sign indicates that the individuals
in the sample are willing to accept a lower wage in order to live in an area
with hot summers. That same qualitative result has been obtained by Rosen
(1979) using individual data from the Current Population Survey together with
SMSA specific attributes and by Hoch (1977) and Cropper (1979) using aggregate
SMSA data exclusively. On the other hand, the negative sign on COLD suggests
that individuals must be paid a premium to live in areas where mean January
temperatures are low and winter weather is probably severe. Of the three
studies just mentioned, only the one by Hoch employs a similar variable. The
coefficient on "winter temperature" is positive in his regressions on Samples
I and II and negative in his regression on Sample III (see Hoch's Table 5, p.
39) .
Next, the coefficient on JACR is positive and significant supporting
Viscusi's (1978) result that employers must pay a premium in order to induce
workers to accept jobs where the probability of accidents is higher. Also ,
this result is consistent with the findings of other investigators who
measured other dimensions of working conditions. For example, Lucas (1977),
Hammermesh (1977) , and Thaler and Rosen (1975) consider the effect of wages of
variables including: (1) a generalized measure of poor working conditions,
(2) the presence of hazardous materials and/or equipment, and (3) deaths per
1,000 man years of work. All three of these variables have been found to be
positively and significantly related to similar dependent variables to the one
used in the present study.
With respect to the HUMD variable, Table 3 shows that its coefficient is
negative but 1 statistically insignificant at the 5 percent level. Although
this negative sign is intuitively implausible, that same result was obtained
125
-------
in Hoch's regressions on each of his three samples. Rosen, however, obtains
the more appealing result that increases in precipitation are positively
associated with real wages. The precipitation variable that Rosen uses, which
is defined as number of rainy days, was always positive and usually
statistically significant in each of 29 different equation specifications (see
Rosen's Table 3.3, p. 94).
/
The pollution variables do not perform quite as well as the other
variables in the equation. Both the linear and quadratic terms for CSOX and
for NOXM are statistically insignificant at the 5 percent level. The result
for CSOX conflicts with those of Cropper (1979) . In her regression for all
earners and in four of her eight occupation specific regressions, a measure of
S02 turned out to be positively and significantly related to median earnings
of males who were employed full time. However, in the Cropper study S02was
the only pollution measure used and, therefore, this variable could also be
proxying the effects of other pollutants. Rosen's results show that this
conjecture is a real possibility. His SO measure occasionally has the right
sign, but is more frequently negative an<^ significant. Particulate, on the
other hand, exhibit superior performance in Rosen's equation. This variable
was positive in each of the 32 cases where it was used and had a tstatistic
exceeding 2 in 27 cases (again, see Rosen's Table 3.3, p,94). The results on
the TSPM variable used in the present study compares favorably with the
findings of Rosen. As Table 3 shows, the linear TSPM term has a positive and
statistically significant coefficient and the quadratic TSPM term has a
smaller negative but significant coefficient.
The elasticity of the real wage with respect to a change in TSPM can be
computed from the estimates presented in Table 3 according to
3RWGH TSPM o ,, 2
_ _ _ - a TSPM + 23 TSPM (4)
TSP M3TSPM RWGH
where e denotes the elasticity, a denotes the estimated coefficient on the
TSPM
linear term and |5 denotes the coefficient on the quadratic term. Evaluated at
the mean of the observed values for TSPM, e = 0.0367, evaluated at the
national primary standard, ® = .1322, ancPcl^evaluated at the national
secondary standard, e The mean of the actually observed values
of TSPM = 96.56 and tlie^tiational primary and secondary standards for TSPM are
shown in Table 24. The comparatively high value for the mean of TSPM can be
attributed to a relatively small number of counties in the data set where
total suspended particulate was considerably in excess of 100. In any case,
these results suggest that in the neighborhood of the national air quality
standards benefits from reducing TSP concentrations are likely to exist.
Illustrative calculations of benefits of national pollution abatement
126
-------
programs are presented for two SMSAs, Denver and Cleveland. These
calculations are derived from the pooled regression estimates in Table
In particular:
(i) SMSA specific means for the variables EDC2, EDC3, HVET,
and FMSZ were obtained from the 1970 U.S. Census 1 in
100 public use sample tapes and substituted into the
equation reported.
(ii) SMSA specific averages for the variables WARM, COLD,
and HUMD were obtained from other sources and substi-
tuted into the equation reported.
(iii) For the remaining nonpollution variables, UNON, HLTH,
TOJ2, and JACR, the sample means reported in Table 22 were
substituted into the equation reported. This
procedure was used because of the difficulties in
obtaining meaningful SMSA specific means for these vari-
ables.
These means, which are reported in Table 26, were then multiplied by their
respective coefficients in order to obtain a predicted wage exclusive of
pollution effects.
For the pollution variables, it was assumed that neither community would
have air pollution levels higher than the primary standards for SO , N02 and
TSP by 1985 and that the secondary standards for all three pollutants would be
met by 1987, In cases where current (1978) pollution concentrations are lower
than the secondary standards, those current concentrations were assumed to
prevail throughout the forseeable future. As previously indicated, Table 27
reports the national primary and secondary standards legislated to take effect
in 1985 and Table 28 reports 1978 pollution concentrations for Denver and
Cleveland.
In Denver, for example, the change in the predicted RWGH associated with
a reduction in total suspended particulate concentrations was obtained holding
constant the values of the other pollution and nonpollution variables. The
values for the remaining pollution variables were held constant because Denver
is already meeting the national secondary standards for them. Also, the
values of the nonpollution variables were assumed to remain unchanged over
time. Projected benefits were then obtained by multiplying the change in the
hourly real wage by annual hours of full time work and then multiplying this
result by an estimate of the number of affected household heads in each SMSA.
127
-------
Annual hours of full time work were assumed to be 2000 and the 1 in 100 Census
Bureau public use sample indicated that there were approximately 382,700
household heads in Cleveland and 218,100 household heads in Denver with the
hours of work and employment characteristics required for inclusion in the
sample used to make the pooled regression estimates.
Annual benefit estimates from pollution abatement in the two cities are
positive according to the calculations made here. For Denver, meeting the
national secondary standards for TSP results in a reduction in the offered
real wage, from $4.1758/hr. to $3.9626/hr. Multiplying this difference of
$.2136/hr. by the number of persons affected times 2000 hours yields an
estimated annual benefit for Denver of $92,968,935. A similar calculation for
Cleveland reveals that meeting the national secondary air quality standards
causes the real wage to fall from $3.8756/hr. to $3.7693/hr. implying a
benefit of $81,360,489. Note that benefits per household head in the two
cities are $426.35 for Denver and $212.60 for Cleveland. Simple calculations
using the estimates in Table 3 and the mean values in Table 26 show that
reductions in TSP levels would be responsible for all of these estimated
benefits. The larger value for benefits for all of these estimated benefits
per person in Denver arises because greater reductions must be achieved as
compared with Cleveland, in order to achieve the national secondary standards.
Finally, the results from estimating Equation (2) using the Dagenais
procedure to construct the missing^observations on the pollution variables are
reported in Tables 4 through 20. Tables 4 through 19 contain various
partitions of Equation (2) based upon age, race, and sex and Table 20 contains
the pooled sample regression. The coefficients on the supply characteristic
variables reported in Table 20 are very similar to those reported in Table 3.
However, both the linear and quadratic terms for all three pollutants enter
the pooled regression insignificantly at the 5 percent level using a twotailed
test. In the partitioned regression equations, the8air pollution variables
are seldom significantly different from zero either. More specifically, there
are five of these regressions where one of the pollution variables entered
significantly. These are: (1) the Male, White, White Collar Worker, Age 5069
partition (TSPM), (2) the Male, White, Blue Collar Worker, Age 3049 partition
(TSPM) , (3) the Male, White, Blue Collar Worker, Age 1769 partition (CSOX) ,
(4) the Male, NonWhite, Blue Collar Worker, Age 3049 partition (CSOX),(5) the
Female, White, White Collar Worker, Age 1769 partition (TSPM). Neither the
linear nor the quadratic term on NOXM was ever significantly different from
zero at the 5 percent level. In the five cases where a pollution variable was
significant, the elasticity of the real wage with respect to a change in the
pollution was computed using the method shown in Equation (4). All of these
elasticities were evaluated at the grand mean (computed over all 1395
observations) of the pollution variables. These means, together with the
means and standard deviations of all variables used in this analysis are shown
128
-------
in Table 25. Finally, the results of the elasticity calculations are
presented beneath the coefficient estimates for the equations to which they
pertain. As indicated there, three of the calculated elasticities are
positive while two are negative.
The relatively weaker performance of the pollution variables in the
equations estimated using the Dagenais procedure can perhaps be attributed to
several factors. First, although Dagenais shows that his method produces
consistent prediction of the missing observations, this asymptotic property
may say little about the finite sample properties of such a procedure,
particularly when a large fraction of the observations are missing. Table 29
shows how this missing observations problem relates to each of the 16
partitional equations estimated. In particular, this table presents the
number of observations for each partition for which actual pollution data were
available. As can be seen, four of these partitions had no observations where
data on all three pollutants were available. Second, the consistency of
Dagenais' method depends upon the use of a generalized least squares procedure
to estimate the hedonic wage relation that requires the solution of a set of
simultaneous, nonlinear equations. Because of computational difficulties, OLS
was used instead. In this setting, it is not clear what statistical
properties can be claimed for the Dagenais approach. Two other reasons for
weak performance, which are common to the replacement with means procedure can
also be offered: (1) observations that do exist on the air pollutants may be
measured with so much error that they provide a great deal of misinformation,
(2) after adjusting for the other factors included in each regression, air
pollution, even if measured perfectly, may not be an important determinant of
wages paid.
Illustrative benefit calculations were also made for Denver and Cleveland
using the estimtaes presented in Table 20. The procedure for making these
calculations was the same as that described previously. For Denver, meeting
the national secondary standards for TSP results in a reduction in the offered
wage from $4.3545/hour to $4.0490/hour implying that annual benefits per
household head are $611 and total benefits are $133,198,000. For Cleveland,
on the other hand, meeting the national secondary air quality standards causes
the real wage to fall from $3.3251/hour to $3.2336/hour so that annual
benefits per household head are $183 and total benefits are $70,034,100.
129
-------
Table 2
Restricted Sample Regression
Est i mates
VAR B T
UNON .313 2.920
HVET .265 2.991
FMSZ .0302 2.074
HLTH -.202 -1.324
EDC2 .205 2.136
EDC3 .495 4.477
TOJ 2 .0801 .957
WARM .942 1.050
JACR .0000594 .0433
COLD -.291 -1.357
HUMD .0102 1.388
Csox .532 .895
TSPM -.832 -1.060
NOXM .0394 .117
P**2 .00000334 1.066
S**2 -.00000305 - .538
N**2 .000000526 . 39&
CONSTANT -30.473 - .818
R2 " .59 DF = 94
130
-------
Table 3
Pooled Sample Regress ion-
Replacement with means
VAR B T
UNON
.127
4.576
HVET
.187
7.179
FMSZ
.0218
3.969
HLTH
-.107
-2.873
EDC2
.0726
2.153
EDC3
.491
12.747
TO J 2
.133
4.929
WARM
-.00977
-2.865
JACR
.00145
3.561
COLD
-.00807
-3.148
HUMD
-.00192
-1.589
Csox
-.00298
- .609
TSPM
.00945
2.045
NOXM
.00206
.268
p a j* 2
-.0000509
-2.203
S:';"2
-.00000548
- .0805
N**2
-.0000252
- .294
CONSTANT
1 .505
3.237
R2 = .30 DF = 1377
131
-------
Table k
Male, White,
Whi te Collar
Worker,
xouseho d
Heads Aged
7-29
MAR
B
T
X10-UN0N
,17476
1 ,4327
X12--HVET
* 6679bIJ""01
- ,60372
X1 9 F H 3 Z
~ 117S3
2,7474
X26-HLTH
>362711-01
-,27360
X23--DC2
. 3: =504
-,97654
X 2 9 -1=; D C 3
c 32479D-01
8985611-01
X 4 0 - "f 0 J 2
~ 4 7!" 91
- »813 0 8
X43-WARM
,3 49 /'4D-02
»3 613 2
X 4 4"" J A C k
31325D--02
1,9076
X45-CGLH
. / 137CD™02
-1,0911
X46-HUMH
~ 213 94.0-02
,40559
X41-C30X
~ 33 633D-04
,25623D-02
X4/-TSPM
.40137D-02
-.31493
X49~'N0XH
.90972D-02
-1,2930
X PK%2
19633D-04
, o J. o
X 2-;8**2
,313 ;> 3 D - 0 4
, 11465
X 3~.N**2
,33521D-04
1,7349
CONSTANT
1 - 3133
1,0201
R-SQUARE-
S 3 R::::
0,2104
21 . 60
D !r ~
95
132
-------
Tab Ie 5
Male, White, White Collar Worker,
Household Heads Aged 30-49
'JAR B "T
XIO-IJNON
•> 26432D-0 :l.
,50824
X12-HUET
,10866
2,3615
X19-FMSZ
-.15115D-01
;L >3827
X26--MLTH
--,21725
-3,5129
X2S-EDC2
~16370
• 2,33 6 /'
X29-EHC3
,46052
6,1510
X40-T0J2
,11154
2,4769
X43-WARM
-,99376D-02
-1.6573
X44-JACR
~33032D-03
1,0669
X45--C0LD
~ 4 2 6 8 9 D - 0 4
.11083D
X46-HUMD
~30551D-02
1,2979
X4:!. -C30X
-.26112D-02
-,39982
X47-T3PM
~ :L 6092D-01
1,4245
X49-NOXM
*26707D-02
~ 6 9 y 6 0
X :L --p:i;S2
¦- , 35353D-04
-,91941
X 2-S:Kt:2
•••• ~ 10503D-03
-,81162
X 3-ri:fwK2
- > 1609211-04
-,56085
CONSTANT
,96252
1 > 4981
ARE> 0,2632
53,12 DF 346
133
-------
Table 6
Ma e, White, White Collar Worker,
Househo d Heads Aged 5o~69
XI (•-•UN Or!
* 75762D--01
. .85593
X12--HVET
-.90 :L 92D--01
~1 , 1334
X19-FMSZ
-.95895D-02
-•>44268
X26-HLTH
¦> 68573D--01
•-,81402
X28-ELC2
v .C ••
2,5788
X29--EBC3
~ 51817
4,8639
X 40 TO J2
--, 32128D-01
-,43872
X43--WARM
- . 16985D--01
-1,9505
X 4 4 ¦¦¦• J A C R
-• •• I4722D-02
-•1 ,2219
X45-COLD
, 3851 41?-02
,56162
X46-HUMD
»50936D-03
. 11387
X41 ¦••• C 3 OX
,67178D-02
,53131
X 4 7 ¦••• T S P ri
,10101
3,3601
X49-N0XM
--, 66079D--02
-1,0324
x 1 ••••?:;;•:< 2
•-,521371!-03
-3,1102
X 2 •••• S >K £ 2
¦••• v 50256D~04
-,20280
X 3-N**2
431952D-04
, 72943
CONSTANT
- J. i 9 4 3 6
-1,5904
R--SGUARE" 0.37;
SSR- 14 • 0:
DP -- 1
eTSPM =
134
-------
Tab e 7
Ha e. White, 3 ue Co ar Worker,
Household Heads Aged 17-29
'v A R B T
X .LO-UMON .22126 1.5232
X12-HVET ,11507 .77295
XI::? FMih-Z • 67060D—01 1,0862
X 2 6 - H I... IH - . 1 J. 19 4 D ¦¦- 01 -,29234 If - 01
X2S EDC2 ~ '50214 1,4339
EDC3 v 37717 1,0469
X40-T0J2 -.85560D-01 -,13650
X43-WARM ,26415D-01 1,3533
X44-JACR .14590D-02 ,65136
X45-C0LB - , 21366D— :) 1 -2.0895
X46-HUMD -.1234711-01 -1 .3453
X I1-C30X - . 26390D-01 -1.7609
M7-SPM - <¦ 54343D—01 --1,7497
< !•-• MGXii : 59654D-02 -<.56895
X 1 P •!' . 3 0 4 '{/ .0—0 3 1 , 8 0 b 6
2-S**2 * 55753D-03 1 .8201
X 3~N**2 ,57881 •< 4 .79457
ONSTfiiN 7 2.. 1767 1,1070
¦¦ Q U f-'i K I::.::: O -¦ 3 2 3 9
i!' 9,739 Dl-- 45
135
-------
Table 8
Male, White, Blue Collar Worker,
Household Heads Aged 30-49
0 A R B T
X10-UN0N .14275 1»3?55
XI2- H'v'ET ~ 11510 1 * 4074
X :L 9 F H S Z - ,1113 8 H - 0 J. ~ , 4 8 8 91
X 2 6 — H I... f I"! , 2 16 .(J ~ 0 2 — ~ 1 8 -J 8 31.1
X 2 8 £ I' 02 -,64656 D - 01 ,667 7 2
X2Q-EDC3 ,6601811 ~01 ,40793
X40-TGJ2 *28272 3,2842
X43--WARM ~»73356D-02 '•-,49400
X 4 4 J A C R > 17985D-02 1,1995
X45~COLD ,50207D-02 ,58515
X 4 6 — H U M D -,43683 D - 0 3 — , /' 7 /' 5 7 D
X41--CS0X -, 13004D-01 -1,0804
X4--TSPN ,25542D-01 2,1776
X49-N0Xr) ,4854211-02 ,79327
X 1 ~!-' :-:K2 - ~ 1 1023D-03 -, 9741
X 2-3**2 ,17094D-03 ,75189
X 3 — N >K >K 2 - '• 4 y /' 5 6 .U—0 4 — 1 , 3613
CONSTANT ,19701 ,13496
SQUARE^ 0,4280
. R -••• 4,471 DF- 56
= 1 4204
TSPM '
136
-------
Table 9
Male,
fl/h i te, B1 ue Co 1 1 ar
Worker,
Household Heads Aged
17-69
VAR
B
T
X10-UNGM
~ 16 463
2,3087
X12-HVET
»13318
2,1021
X19-FMSZ
»19745D-01
:L . 0386
X26--HLTH
-•> 18033D--01
-.18505
X23-EDC2
v 6945ID-02
.83603D
X29-EDC3
. 26313D-01
»25971
X 4 0 T 0 J 2
.35903
5*1153
X43-WARH
-. 12433D-01
-1.6865
X 4 4 J m C K
»29094D-03
.28674
X45-COLD
- . 39463D~02
-1.7421
X46--HUMD
-~99736D-03
-.27515
X 41 ¦¦•¦'CSOX
- . 24192D-01
— 2 . 8 5 4 4
X47-XSPM
~85630D-02
.83961
X49--N0XM
, 18851D-02
.35982
X 1 -P ;•!<* 2
-.290 4 4 D-0 4
-.55213
X 2 - 8 :'< ••!< 2
* 46 75 4 Ei ~ 0 3
2 > 7 '7 0 2
X. 3"-NWS
- ¦> 24129li~04
-.69613
CONSTANT
:L . 9040
2 .4022
R -SQUARE- 0 , 3457
88R~ 23.65
"CSOX
= -.3125
137
-------
Table 10
Male, Non-White, White Collar Worker,
Househo 1 d
Heads Aged 30-^9
UAR
B
T
X10-UN0N
, 23558
2,39 9 2
X12-HVET
~ 10339
,96532
X19-FMSZ
. 93703D--02
.43587
X2A-HLTH
>15968
,89901
X28-E0C2
-~19323D-02
-.133980-
X29--E0C3
,33745
1,9824
X 40-TO-J 2
- ~ 84307.0-01
—,88253
X43-WARM
.39743H-02
,29485
X44-JACR
-~257330-02
-1,7306
X45-COLD
.887100-03
,15121
X46-HUMO
.12822H-02
.2.1563
X41. -CSOX
.141840-01
,69880
X47-T3PM
,435850—01
1,5422
X49-N0XH
- ~ 435090-03
-,528200
X 1-P*>K2
-»208060-03
-1,3906
X 2-S**2
,571660-04
,18097
X 3-N**2
,216930-04
,34107
CONSTANT
-1,6930
-1>1465
R-~SQUARE- 0,373?
SSR-- I I. ,37 DF-
138
-------
Table 11
Male, Non-White, White Collar Worker,
Household Heads Aged 17~69
v Aft B T
X10-lJNGN
,136 0 4
2,5040
X12-HVET
~10293
1,4493
X19-FriSZ
--.30211D-02
-~22040
X26-HLTH
,75763D-01
,49945
X28--EDC2
, 77325D--01
,80932
X29-EDC3
, 2 Cj Ct o o
2,2519
X40-TOJ2
~ 66240D-01
~ 93804
X43-WARM
- , 41372D-02
-~34556
X44-JAGR
->1392111-02
-1,2404
X45-C0LD
~25076D-02
~47563
X46-HUMD
~ 17020D-02
~ 33027
X41-CSGX
~ 13301D-01
» 88935
X 4 7 - T S P i'i
»44063D-01
1,3041
X49-N0XM
¦- ~ 8 6 6 6 0.0 — 0 2
-1,3342
X 1-P:{:*2
-,19905D-03
-1,5454
X 2--S:('^2
-,20679D-03
-,92235
X 3-N*#2
,64510D--04
1,2350
CONSTANT
-,96137
-,76367
-SQUARE™ 0,266?
3 R- 13,915 DF- 130
139
-------
Table 12
Male, Non-White, Blue Col
HousehoId Heads Aged
lar Worker,
17-29
.2^234
•: r ¦— - •»
1 x ^
X'2-HME7
.18335
1=4333
XiS-FfSZ
,3234l3-01
1.4233
X2S-HL7H
,203333-01
.54SS3
X23-E322
i.1371
2.0347
XZ3-Z3CZ
1,0S33
2.1051
X43-:aJA?M
*33153D-0.
1 , 741;;
X44-_A-ZR
> 327722-02
1,3113
X'~,5-CDLD
, 244122-01
: , 0 3 0 1
X43-:-:U!*D
-.353343-02
-.24330
X4i-csox
.:1230
1.7170
X 47-^5?'*!
- . 3220 "-D-01
-.75043
X4S-NGX.''!
i S i 0 4 3- 0 i
-.33437
X 2
.144202-02
~
X 2-S*-*Z
— - ~ 2 0 ~ 3 2 — 0 "~
-¦.4337
X Z-:!lf--«-2
,230*72-03
- -T ^
_ S 4. . O w'
CC\S7AN7
-7,3120
-1 » 5332
r-3GUAPE= 0.4325
33rf = B.2SS " DF-
140
-------
Table S
Male, Non-White, Slue Go a-r Worker
Househo d Heads Aged 30-kS
WAR
T
X10-UNON
XI2-!-! VET
X19-FMSZ
X26-HLTH
X28 EDC2
2 9 £ D C 3
X40-TQJ2
X 4 3 •• • W .-i K M
X44-JACR
X4o ;: 3LD
X46-HUM0
X41 CSC !
X 4 7 -¦ T 3 P M
X 4 9 - N ~ X ri
X 1--P $:¦¦!£ 2
X 2--8'n-K2
X : - N 5K ;M 2
CONSTANT
,34653
¦ ,598780-01
,256300-01
• . l' S| A '*!* A
•J w ^ »' .v.. w
¦,493950-01
¦,324360—01
,621050-02
•,791030-02
,226910-02
,894630-02
, 687900-02
,195980-01
,109880-01
-,199170-04
•,282700-04
-,313170-03
,743860-05
-,40612
5,5150
¦, 9 3 8 4 6
2,3752
-2,1214
-.77176
-»28477
,957390
-,96984
2,3888
:L , 9652
2,2345
1,9061
-,463440
-,56956
-2,1435
,20174
-,44776
-SQUARE- 0,5342
T - 9,055
0 F- 109
CSox
= .2959
141
-------
Tab e ^
Male, Non-White, Blue Collar Worker,
Househo d Heads Aged 50-69
VAR
XI0
-UNON
•r r.r •".» ¦"> •••>
» J. -.j / .v.: .c.
1
t / A» 3 V.'
X12
-HUET
i •..> -.'.i o --"-1 / Jj 01
¦— *
38795
XI?
-FMSZ
-»56542D-02
— +
28441
'¦¦¦ A
-HLTH
-~47119
'"5
»9557
•\ wl O
-EBC2
¦> 71372D-0 J.
87034
• \ A. '
-EDC3
-•,2102£
70450
:(40
- TO J2
-.1316311-01
- ~
12876
X 4 3
-WARM
- •> 24253D-02
™ »
91059
X 4 4
-JACK
-.37103D-03
"* ~
29262
X45
-COLD
.1.073D-02
— ~
10988
X 4 6
-HUMD
15479D-01
-1
* 7 323
X4:L
-csox
» 7 6148D-02
43807
X47
-TSF'i'i
.12736H-01
35004
X ! 9
- N 0 X M
¦••• ~ 22636D-02
- f
21454
X 1
~P>5c*2
11332D-03
r.r n • £. •(
A- .L
X 2
- S-K#2
,19079D-03
*
95534
V
- N >!< >K 2
* 4894913-04
43614
CON
ST ANT
1,5146
*
6 o 8 5 b
142
-------
Table 15
Male, Non-White, Blue Collar Worker,
HousehoId Heads Aged
17-69
A R
B
T
X10--UN0N
, 29899
5,7490
X12-HVET
~ 18808D-01
,37703
X19-FMSZ
~25626D-01
2,3439
X26-HLTH
-,25763
-3,2402
X23--EBC2
-v 25374D-01
-,50980
X29-EDC3
.10341.
1,2425
X40-TOJ2
-.70433D-01
-1,3048
X43-WARM
* 2S275D-03
— ,35284 D
X44—J A CI';
~203771)¦•¦¦02
2,9009
X45-COLD
.32852D-02
,37020
X46-MUMD
~31924D-02
1,0331
X41-C30X
~ :L 234730-01
:L ,4095
X47 — TSPM
¦> 29400D-02
,40219
X49--N0XH
,2732211-03
.71365D
X I. •P:i<*2
, 70330D-05
,19892
X 2~S-!-!-'2
13161D-03
-1,4782
X 3-N&&2
, 18 6 6 6 .(J •¦¦ 0 4
,52094
CONSTANT
-,16171
-,191.44
K"" Li U A K I::.— 0 •> 3 6 2 5
3 SIR-- 26,2? D F::::
143
-------
Table 16
Fema'le, Non-White, White Collar Worker,
Household Heads Aged 17-69
.9-
><43-
X44-
X43-
X^3-
}! -11 ¦
\.' IT
/\ V- / •
CON:
UMC-:
F '*32
1332
¦£ 333
_ 2
¦w AC!?
¦C3L3
-3S3X
¦~3FM
-h:0X*
-B*-*2
:7AM7
„ 232'*-33- '-I 1
1 3 £ 21 D - 0 3
34333
322373-0!
' *» -? r.r ~r
—. / w
333333-01
' 2 3 2 0 D -
:
2081-3-01
i SS23D-01
33G33D-01
23303D-0:
2347D -01
73^333-04
1£0243-03
, .737:3-03
3,3250
• 3343D
- . 3 0 4 3 7
-¦23711
; , 3
. 23232
-1 .3014
- : 15 «¦ 0 3
- 1 3 1 3 3
1,3233
-1.4 S 0 3
< 2041
3 33~
— - n ;r
a w' w w
1 .4333
:?-s"uar;
333 =
0,313,
1 .432
20
144
-------
Tab e 17
Fema e, White, White Co a.- Worker,
Household Heads Aged 17~69
v a <
v -!
O-.NGIv
> / •
j.
2--:"=Z
\ ;• <-?
/ .
3 --'17'-
••*¦
A
3-E2C2
V **'
2-2222
\»
/ \ • *
C-"CJ2
A
2-UiA^!":
-'--A:"
x43-22_2
X4S-HUMD
X41-CS0X
X'47-73?M
X42-*GXr1
x 2
X I - ?* 2
— *: ;< X
- J
1 3E 52
.3743
^ ^ 3
w W —
.22-'33 — .a
330CS2-02
1 07 022-03
1 2252-0 1
70S042-02
232053-02
3!1S7D-01
'.522SD-01
2CCE72-03
27133D-03
7 i 723D-04
. 72~3
: .1272
¦.17421
¦1=0157
,77133
i « 6 4 31
¦ • 12--22
,•3204
¦.357S00
•1 .34G1
¦, 13102
.17302
•2 ,7307
•: .7352
;> "> ~ n 71
¦ 1 = 02 1 0
2 <1 ~
2.03SS
7-53LAR2= 0,3300
~ SR= '3.0^4
3 0
eTSPM " "2- 58
145
-------
Table 18
Female, White, Blue Collar Worker,
Household Heads Aged 17~69
X10-LJNON
X12-HYET
X19--FMSZ
X26-HLTM
X2S-EDC2
X29-EDC3
X 4 0 - T 0 J 2
X43-WARM
X44-JACR
X45-CQLD
X46-HUMD
X 4 J. - C S G X
X47-TSPM
X 49-MOXM
X 1 -i;;,'r-K2
X 2-3*:!'2
X 3™N**2
CONS Ti-iNT
RSQUARE-" 0,6809
S S R1,513
~ .u- o Z
10974
,26944D-01
, 390301.1-01
,41989
.89085D-01
,21485
,62321D-02
,30756D-02
>10648D-01
,78924D-02
, 1 4 2 4 6 D - 01
, 12046
,20425D--01
, 66602D-03
,12062 D-0 3
,10477D-03
6,3873
~ ^ J / O \J
. 19444
,82491. D •••¦ 01
,92274
, 13941
,55869
,87444D-01
,97336
,27562
¦,21904
,22577
1 , 6 627
,82539
¦1 ,7580
•,94019D-01
•,81019
•,84302
D F - 8
146
-------
Tab Ie 19
Female, Non-White, Blue Collar Worker,
Household Heads Aged 17~69
3
X:. 0-U;\iC;M
X12-C*3Z
;;23-HL7H
X2b-£222
::23-2322
X-0-~2J2
X43->iA!;?:v!
X'^5 -22L3
X G-HUMD
X.4 1-CSOX
xi7--re?v
w O-.N'pVVi
A • '* w , S w / \ :
V — O -> 2
w 2-5
X
C2N:STAM "
7 2 2 0
572713-01
2 2 7 0
232353-0i
212 i 1
330223-01
101 3SD-02
A i 0233-02
401122-02
ri i — b / 3 — 0 2
373143-02
723 1 32-02
237323-02
137222-04
30703D-C4
212373-04
. 4323
c 2 S
r-i —¦ *n -r
, 0333
3352:
5^1^.02
2C0S4
'-<37373-01
»0354
48224
. 3332
3-531
48730
433"4o
~
_ -
i 3 s 1
34323
n - ~ ¦- 2
3-23UA33;
0,4277
3 .= 33 1
147
-------
Table 20
Poo led Sample Reg ress i on
VAR
B
T
X10-UN0N
. .12826
4.5299
XI2-HVET
.18686
7.1260
X19-FMSZ
.22907D-0!
4.1336
X2 6 -HLTH
-.999430-01
-2.6863
X28-EDC2
•74755D-01
2.2167
X29-EDC3
.49221
12.708
X40-T0J2
.12811
4.7644
X43-WARM
-.77862D-02
-2.4942
X44-JACR
. 14097D-02
3.4679
X45-C0LD
-.73928D-02
-4.1044
X46-HUMD
-.783580-04
- .56596
X41-CS0X
. .154620-02
.55284
X47-TSPM
.82340D-02
1.8694
X49-N0XM
. 16475D-02
.84763
x l-p**2
-.373980-04
-1.6590
X 2-S**2
-.97076D-04
-1.5851
x 3-N**2
-,40s460-05
- .28572
CONSTANT
1 .1411
3.5403
R-SQUARE = 0.3065
SSR = 281.1 DF = 1377
148
-------
Table 21
Regression to Construct SOXM
VAR
X10-UN0N
X12-HVET
X19-FMSZ
X26-HLTH
X28-EDC2
X29-EDC3
X40-T0J2
X42-AVGT
X43-WARM
X44-JACR
X45-C0LD
X46-HUMD
X35-REG1
x3 6-reg2
X37-REG3
X30-PRX1
X31-PRX2
X32-PRX3
X33-PRX4
CONSTANT
3.0550
-2.5744
.27915
- .69527
.35916
-1.3497
-1.6145
-2.2513
- .96528D-01
- .329790-01
.46237
- .39650
30.051
9.0446
21.416
-3.6440
-4.0488
-2.9865
-6.1270
40.391
3.5062
-3.0967
1.7229
- .57786
.35429
-1 .1199
-1.9672
-3.3839
- .52877
-2.6856
1 .9103
-5.1329
10.556
3.3579
6.9766
-1
-1
-1
0700
2150
86171
3064
2.9248
R-SQUARE
SSR =
0.5420
.3229D + 05
DF = 482
149
-------
Table 22
Regression to Construct TSPM
B
- .27108
.13148
.20762
.11578
-1.2855
-5.3022
- .62307
-4.2312
.786o1
- .96968D-02
1.5041
-1 .1924
39.444
32.932
29.224
14.834
17.663
13.110
-2.0068
41.964
VAR
X10-UN0N
X12-HVET
X19-FMSZ
X2 6-HLTH
X28-EDC2
X29-EDC3
Xi»0-T0J2
X42-AVGT
X43-WARM
XH-JACR
X45-C0LD
X46-HUMD
X35-REG1
X36-REG2
X37-REG3
X30-PRX1
X31-PRX2
X32-PRX3
X33-PRX4
CONSTANT
R-SQUARE = 0.3727
SSR = .22410 + 06
DF = 691
T
- .17266
.885050-01
.65983
.57^560-01
- .65959
-2.3759
- .41121
-7.9669
2.9667
- .41621
8.91^9
10.784
8.8559
7.7270
5.8749
3.6951
4.4771
3.0617
- .36068
2.2201
150
-------
Table 23
Regression to Construct NOXM
B
4.0234
-3.1084
- .65904
-5.6057
-2.8367
- .42449
.58784
10.479
-3.5136
.6172AD-02
-1.6996
- .29514
7.1282
8.1533
-15.842
-3.4347
- .47142
-7.1744
-31.613
271.03
VAR
X10-UN0N
XI2-HVET
X19-FMSZ
X26-HLTH
X28-EDC2
X29-EDC3
X40-T0J2
x4 2 -avGt
X43-WARM
XM»-JACR
X45-C0LD
X^6-HUHD
X35-REG1
x3 6 -reg2
X37-REG3
X30-PRX1
X31-PRX2
X32-PRX3
X33-PRX4
CONSTANT
R-SQUARE = 0.3337
SSR = .4039D + 05
DF = 236
T
1.8520
1.6117
1.8977
1.9463
1.2861
.16104
.31671
4.5831
3.2760
.22688
2.6106
.74684
.67026
1.0176
1.3853
.59532
.83510D-01
1.2183
2.8677
3.6656
151
-------
Table 2^
Correlation Matrix
1
3
10
12
IS
5
23
2B 2
3
4<»
41
4
3
44
45
4b
4 7
4 G
: <. .< 2
S ~:
-it 2
N -i- ¦* 2
UNCN
HUET
FM
,.J L.
GCCP
R
HWfi
HLTH
GDC2 E
DC 3
f !i J 2
C
SOX
1-i
ARC:
jacf;
¦cc
L.D
,!l. .
TBfVi
KG
1 2
l . 00
0 .
' 0
0. OS-
0. 05
0. 14
0.
U.08-
0
, 03
4 ,J4
0 . >-"'5 0
„ 03-
0 . 03-
0
. 03
0
. 02
0.0 2
— 0 <
02-
0 .03 -
" Li , :i '
¦ 0 .
2'S*»2
O .. 2S
1
0 0
- 0.00 -
0. 0 2
0. 30
0.
oe
0.44
0
.35
- 0 . 0 5
-0.17 0
. A 0
0. 13-
0
, 0 0 -
0
0.02
- 0 .
w. «JJ ' '
• 0. 15
0 0 5
0 _
3vl k-sZ
G , OB-
-- 0 .
0 0
! . 00
0, 04
O . Id
-0 ,
07
0 . 01
0
. 00
- 0 . 0 3
0.2B-0
•-> -n
0 .01
0
, 0 2 -
0
. OS
0 . 04
-¦0
ij 3 ~
• 0 . 0 5
0 0 i -
0 .
iOUKGM-
0 . 0 3-
-0.
02
0. 04
1 . 00
0. 07
0,
08-
0 . 13
0
, i 1
-0.02
0.14-0
.21
0.12
0
. oa-
0
. (.13
0. 27
U-
1.1. '-"'G
0 . 0 7
0 ,
12HULT
0. 1 4
0 .
30
0,16
0.0 7
1 .00
0 .
12
0. i s
0
.28
-0 . 01
0.03 0
.07
0.12-
0
. 02-
0
. 1 s
0. 1 i
0.
0 3 -
0.12
0 . 0 2 -
0.
isrnsz
0 . Oo
0 .
08
-0. 07
0. 0 S
0.12
1.
00-
0.01
0
.03
-0.01
-0.02-0
. 1 1
0- 1 4-
0
. 02
0
. OS
0.14
0.
02
0 . 03
0 . 0 2 -
0 .
22QCCP
0 . 08
0 .
44
0 . 01 -
0. 1 3
0 . 1 S
-0 .
01
1.00
0
y 4S
0.01
-0.14 0
0.06-
0
. 05-
0
¦ 13-
¦ 0 .. 0 7
-0,
15-
0 J i -
0 .. 0 0
' >'
2SRHMG-
0. 03
0.
85
0. 00
0 . 1 1
0. 28
0.
03
0.46
1
. 00
- 0 .05
-0.1G 0
. 3S
0.17
0
. 03-
0
/-» I"?
O.OB
- 0.
25-
< :¦15
0 . 0 7
0 .
i'3HL FH-
0 .04
-0.
05
- 0 . 0 3 --
0. 0 2
- 0 . 0 1
-0.
01
0 . 01 -
¦ 0
.05
1 .00
-0.05-0
. 02
0. OB-
0
. 00-
0
. 12-
0.03
- 0 „
07-
0. OS
0 . 0 4 -
0
28EL 22
0 . 05'
-0.
17
0. 26
0. 1 4
0.03
-0.
02-
0. 14-
0
.IS
- 0. 05
1.00-0
. G5~
0.01
0
. 08
0
. 01
0 . 07
- 0,
05-
0.0 2
0 , 03-
¦0 .
22ET23
0. 03
r.
40
-0.22-
0 ,21
0 .07
-0 .
1 1
0 . 32
0
. 3 G
-0«02
-0.65 1
. 00-
0 . 04-
0
. 04 -
0
. 13-
0 , 20
-0.
OG-
0. 14-
¦0. 04
0.
4 0 T 0 J 2 -
0 .0 3
0 .
13
-0. 01
0. 12
0,12
0.
14
0 . OS
0
. 17
0. 08
-0.Oi-O
. 04
1 . 00
0
.03-
0
. US
0. 0 5
- 0.
:! 0 -
< j . 0 0
0 . 02"
¦0.
4 1 C L i! A -
0 . 0 3
-0 .
< J 0
0. 02
0 ,03
0 .02
- 0.
02-
0. 0 5
0
.03
- 0 . 0 0
0.08-0
. 04
0 . 03
1
. 0 0 -
0
. 1 0-
¦0, 08
-0.
r- j *-1
0 . OS
0« 28-
0 .
>13WArv.".
0.02
-0.
—
- 0 . 0 £ ~
¦0. OS
- 0 . 1S
0 .
0G~
0. 13-
¦0
-0.12
0.01-0
. 13-
u. (J s -
0
. 10
1
. 0 0
0 .0 3
t.
3G
0 . 42 -
¦0 . 12-
0 ..
4 4 JAC."?
0. 02
0 .
0 2
0 . 04
0.27
0. 1 1
0.
14-
0„ 0 7
0
.03
-0. 03
0.07-0
. 20
0 . 0 5 -
0
. OB
0
. 03
1 . 00
03
0 .07 -
0 . 0 3 -
0 .
45COLD -
0. 0 2
• 0 .
n <--<
.i. O
-0.OS-
0.11
- 0 . 0 8
0 .
02-
0. 1 5-
0
. 25
- 0 . 0 7
-0.05-0
. OS-
0. 10-
0
. 23
0
.36-
•0.03
1,
00
0 . 15-
0 . 2 5
0 .
4GHUI-.D-
0 ,03
0.
15
-0.05-
¦0 . OS
_fi 1 r>
• J w
0.
03-
0. 11-
0
.15
- 0.00
-0.02-0
. 14-
0. 00
0
. i.) 3
0
.42
0.07
0.
15
1 .00-
r n
L- .t ~
0 .
-1 7T3FH-
¦ 0 . 0 3
0 .
OP
0. 0 1
0. 0 7
0. 02
0 .
C2-
¦0 . 00
0
.07
0 . 04
0.03-0
.04
0.02
0
.29-
0
.12-
0.03
-0.
r i L~ _ _
0 . 42
i „ 00-
0 .
IGMGaK-
0.0 4
0 .
*-1
- 0 . 0 4
0. 08
-0 .01
— u .
10
000
o
. 02
-0» 05
—0.OS 0
. 07-
0.02-
0
.07-
0
. 1B-
¦0 .04
0.
37-
¦ >. 2S-
0. 08
1 .
III1E 13.S Sc.CS
-------
Table 25
Means and Standard Deviations
of Variables
• 'ft
AN
SD
X
9261,2
3391,5
x 2~s.r#2
423(37
430,09
x 3-N**2
3 5 4 6 <• 6
2470,7
X 4--QTBC
0 »
0,
X 5-TINC
0 ,
0 ,
X 6-AGEH
41 ,107
5,5200
X 7--SEXM
1»0000
0 ~
X S-HLT1
4,4780
1 ,3287
X 9-HLT2
4,7582
; ,94289
X10—UNOM
,30769
,46217
X1:!.~EDUC
4 , 9588
1,9641
XI2-HVET
»o 3 4 6 2
,48220
XI3-RACE
1,0000
0,
X14-CITY
* 72253
,44337
X15-PRCX
2*3934
1,1538
X1 6-'REG
2. 2 4'i 5
1,0950
X I?-SELF
1 , 0000
0,
X:L8-SIZE
,80495
,39679
X19-FMSZ
4 * 7335
2,0049
X20-LTQJ
3,9588
1»3335
X21-INBX
1,0091
~ 39053]
X22-0CCP
1,0000
" 0.
X23--CIi;«C
2 WO,
¦f ''5 A "> O
1. U' v
X24-AHWG
5,3630
• "7 r ;•=;
X25--RHWG
1 * 6320
,44564
X26-HLTH
,14335
,35594
X27-EDC :l.
~ 13462
,34178
X2S-EDC2
,43407
,49632
X29-EDC3
~ 43132
,49594
;<30--PRX 1
,22527
,41834
X31--PRX2
,33733
,48782
X32-PRX3
,23332
» 4 2 3 6b
X33-PRX4
, 714290-•••01
,25789
X34-PRX5
,82413D™01
,27538
X35--REG :i.
,32143
,46767
X 3 6"" K' I;;. U 2
,29670
.15743
X37-REG3
>19780
,39889
X33-REG4
,18407
, 8 8 0 /
X39-T0J.L
,60989
,48845
X40-T0J2
»39011
,48845
X 4:!. C3 0 X
25,456
11,253
X42-AUGT
11,431
3,9808
¦warm
74.3 i 6
4,85 43
X ¦! 4- JACR
58>464
30,9 79
'X'iS-"C0LD
31,723
1 i. 0:73
X 4 6---HUH.0
35,935
1L,221
X47-TSPM
94,513
13,148
X4S-SQXH
17,771
10 ~ 3 8 6
> 4 9 -• M 0 X H
56,763
18 , 0 4 0
-------
Tab i e 26
MEANS OF NON-POLLUTION VARIABLES
USED IN BENEFIT CALCULATIONS
VARIABLE MEAN
Denver CI eve I and
UNON .307 .307
HVET .402 .556
FMSZ 3.40 3.46
HLTH . 148 .148
EDC2 .456 .567
EDC3 .449 .298
TOJ2 .390 .390
WARM 72.00 71.90
JACR 58.46 58.46
COLD 30.60 18.90
HUMD 13.73 33.66
CONSTANT 1.00 1.00
Table 27
NATIONAL AIR POLLUTION STANDARDS
(In Micrograms Per Cubic Meter)
PRIMARY STANDARD SECONDARY STANDARD
S02 75 60
no2 100 100
TSP 75 60
Table 28
1978 POLLUTION CONCENTRATIONS
IN DENVER AND CLEVELAND
(In Micrograms Per Cubic Meter)
Soz
N02
TSP
DENVER CLEVELAND
16.9 61.49
100 65.0
36 72.2
154
-------
k
5
6
7
8
9
10
11
12
13
\k
15
16
17
18
Table 29
Cross-Tabulation of Incidence of
Actual Pollution Data By Partition
TSPM SOXM NOXM TSPM.SOXM TSPM.NOXM SOXM.NOXM TSPM,N0XM,S0XM
58
30
17
26
5
5
.4
164
94
48
87
25
18
15
59
49
22
42
10
10
28
19
2
16
1
2
28
17
5
11
0
1
0
22
15
10
12
6
5
4
59
39
19
33
13
17
13
78
60
36
46
22
29
19
40
36
23
30
13
17
11
80
61
32
50
17
24
14
25
23
11
19
9
10
8
145
1 20
66
99
39
51
23
24
18
6
18
6
6
6
32
22
7
19
3
4
2
16
6
5
6
0
0
0
57
52
32
43
21
27
18
-------
REFERENCES
1. See, for example, papers by Hall (1973), Heckman (1976), Rosen (1976), and
Wales and Woodland (1980).
2. The variables contained in P and W will be defined more explicitly
momentarily.
3. The procedure used to assign air pollution measures to the individual
observations is similar to that used by Crocker, Schulze, et al. (1979) .
4. Regressions for partitions containing less than 50 observations were not
estimated. For these cases, the observations from two or more partitions were
pooled and one regression was run on the combined data sheet.
5. Dagenais also suggests using a generalized least squares approach to
estimate the hedonic wage equation. However, the approach recommended
required that a system of k simultaneous nonlinear equations be solved in
order to obtain estimates of the slope coefficients where k denotes the number
of regressors. Because of the computational burden involved in using the
procedure, it was abandoned in favor of the simpler OLS approach.
6. Additionally, even if the NOXM variable was eliminated from consideration,
there would still have been only 432 families for whom data on both SOXM and
TSPM could have been matched.
7. Note that in some of these partitioned regressions, variables such as UNON
and HVET are excluded because all observations on them are equal to zero. For
example, HVET has been excluded for this reason in Table 17.
8. The regressions used to construct the missing values for the pollution
variables are shown in Tables 21, 22, and 23.
156
-------
BIBLIOGRAPHY
Brown, C. "Equalizing Differences in the Labor Market," Quarterly
Journal of Economics, 94 (1980) , 11334.
Crocker, T., Schulze W., et al. Methods Development for Assessing Air
Pollution Control Benefits, Volume 1 EPA600/57900/a (February,
1979).
Cropper M. Methods Development for Assessing Air Pollution Control
Benefits, Volume IV EPA 600/579001/d (February, 1979).
Dagenais, M. "The Use of Incomplete Observations in Multiple Regression
Analysis," Journal of Econometrics, 1 (1973), 31728.
Hall, R. "Wages, Income, and Hours of Work in the U.S. Labor Force," in
G. Cain and H. Watts eds., Income Maintenance and Labor Supply.
New York: Rand McNally, 1973.
Hamermesh, D. "Economic Aspects of Job Satisfaction," in Ashenfelter
and W. Oates, eds., Essays in Labor Market and Population Analysis.
New York: John Wiley and Sons, 1977.
Heckman, J. "The Common Structure of Statistical Models of Truncation,
Sample Selection and Limited Dependent Variables and a Simple Esti-
mator for Such Models," Annals of Economic and Social Measurement,
5 (1976), 47592.
. "Sample Selection Bias as a Specification Error," Econometrics,
47 (1979), 15362.
Hoch, I. "Climate, Wages and the Quality of Life," in L. Wingo and
A. Evans, eds., Public Economics and the Quality of Life. Balti-
more: Johns Hopkins Press, 1977.
Lucas, R. "Hedonic Wage Equations and Psychic Wages in the Returns to
Schooling," American Economic Review, 67 (1977), 54958.
Maddola, G. Econometrics. New York: McGrawHill, 197 7.
157
-------
Meyer, J. and Leone, R. "The Urban Disamentiy Revisited," in L. Wingo
and A. Evans, eds., Public Economics and the Quality of Life.
Baltimore: Johns Hopkins Press, 1977.
Nakamura, M., Nakamura, A., and Cullen, D. "Job Opportunities, the
Offered Wage, and the Labor Supply of Married Women," American
Economic Revfew,' 69 (1979), 787805.
Nordhaus, W. and Tobin, J. "IS Growth Obsolete?" in Economic Growth.
New York: National Bureau of Economic Research, 1972.
Rosen, H. "Taxes in a Labor Supply Model with Joint WagesHours Deter-
mination," Econometrics, 32 (1976), 485507.
Rosen, S. "WageBased Indices of Urban Quality of Life," in P.
Mieszkowski and M.Straszheim, eds., Current Issues in Urban
Economics. Baltimore: Johns Hopkins Press, 1979.
Thaler, R. and Rosen, S. "The Value of Saving a Life: Evidence from
the Labor Market," in N.E. Terleckyj, ed. , Household Production
and Consumption. New York: National Bureau of Economic Research,
1975.
Viscusi, W. "Wealth Effects and Earnings Premiums for Job Hazards,"
Review of Economics and Statistics, 60 (1978), 40816.
Wales, T. "Labor Supply and Commuting Time: An Empirical Study,"
Journal of Econometrics, 8 (1978), 215226.
Wales, T. and Woodland, A. "Estimation of Household Utility Functions
and Labor Supply Response," International Economic Review, 17
(1976), 397410.
_ and . "Estimation of the Allocation of Time for Work, Leisure,
and Housework," Econometrics, 45 (1977), 11532.
158
------- |