You are on page 1of 6

Environ. Sci. Technol.

2000, 34, 1407-1412

intrinsic value. Values based on preservation for potential


Using Conjoint Analysis To Value future use or to avoid irreversible losses are referred to as
Ecosystem Change† option values. Total value is the sum of user, option, and
nonuse values (3). Both RP and SP techniques can be used
to assess use values; but SP techniques may be more
STEPHEN FARBER* AND BRIAN GRINER appropriate for nonuse values.
Graduate School of Public and International Affairs, Resource and environmental valuation is important in a
University of Pittsburgh, Pittsburgh, Pennsylvania 15260 variety of contexts, including damage compensation and
rational management. The above techniques have all been
developed to deal with the pervasive inability to observe
market prices for all of nature’s goods and services (4). Where
Economists have developed a variety of methods to market prices are unobservable or cannot reflect the full value
measure the values of environmental goods and services. of goods and services, economists seek to determine what
Conjoint analysis (CJ) is a technique developed by they call “shadow prices”, which represent the difference
mathematical psychologists to establish the structure of that a unit of the good or service makes to social welfare (1).
preferences across multi-attribute alternatives. It is a type CVM has an extensive literature (5) and will not be
of stated-preference method that has captured the summarized here. This paper presents an overview of the
attention of economists for purposes of analyzing preferences history and applicability of CJ to environmental valuation.
Section II briefly explains CJ and how it can be used in a
toward environmental goods and services. This paper
manner consistent with utility theory to measure values of
outlines the applicability of CJ to environmental valuation, environmental changes. Section III illustrates the variety of
illustrates applications and valuation issues in the applications to environmental valuation and discusses some
literature, and presents an application of CJ to valuing a of the leading issues in this application of CJ. Section IV
Pennsylvania watershed quality improvement. presents results of a recent CJ valuation analysis of watershed
quality improvements in western Pennsylvania.

I. Introduction II. What is CJ and How Can It Be Used To Estimate


Economists have developed a full tool kit for the economic
Environmental Valuations?
valuation of natural resources, the environment, and its CJ was developed by mathematical psychologists (6) in the
services (1, 2). Valuation methods are divided into revealed- early 1960s to transform subjective choice responses into
preference (RP) techniques, where valuations are inferred estimated parameters. It has been defined as (7):
from actual observations of choice behavior, and stated- “any decompositional method that estimates the structure
preference (SP) techniques, where valuations are directly of a consumer’s preferences (i.e., estimates preference
obtained from hypothetical statements of choice. The RP parameters such as part-worths, importance weights, ideal
methods include the following: points), given his or her overall evaluations of a set of
Hedonic pricing method, whereby premiums individuals alternatives that are prespecified in terms of levels of different
will pay to have environmental amenities (e.g., air quality or attributes.”
beach access) are observed CJs were adopted by marketing research scholars in the
Travel cost methods, whereby costs individuals will incur early 1970s to analyze consumer choice (8). Since that time,
to experience spatially distributed natural systems (e.g., forest CJ has been used in a large number of consumer applications
recreation) are observed (9), transportation (10, 11), and health care (12, 13).
Averting cost methods, where costs that people will incur The application of CJ can be illustrated with a simple
to avoid environmental damages (e.g., unhealthful air) are example. Suppose a decision-maker is faced with considering
observed goods possessing various levels of several attributes, (X1-
The SP techniques include the following: X3), where the goods may be autos and X1 may be
Contingent valuation method (CVM), by which people horsepower, X2 may be color, and X3 may be price. Suppose
are presented with a hypothetical contingency scenario and the decision-maker has a value or utility function represented
are asked explicitly what that scenario, such as improved air by
quality, is worth to them
Conjoint analysis (CJ) method, by which people are U(X) ) f(X1, X2, X3) (1)
presented with a set of hypothetical scenarios involving
various levels of two or more attributes and are asked to For simplicity, suppose this utility function takes the following
choose, rate, or rank among them; and the structure of their linear form:
preferences toward these attributes are inferred from their
choices U(X) ) w1‚X1 + w2‚X2 + w3‚X3 (2)
Economists typically distinguish between values to society
based on use of a resource and values not based on use. Use where w1 represents the weight attached to attribute X1, etc.
values may include consumptive uses, such as resource These weights are termed “part-worths” and reflect the
extraction, or nonconsumptive uses, such as hiking and relative importance of each attribute in the decision-maker’s
viewing. Nonuse values would include knowing that a choice (13). CJ provides a variety of procedures for deter-
resource is in a certain condition when there is no intention mining the part-worth weighting factors, i.e., the structure
of ever using it; this has been referred to as existence or of preferences.
Using the auto example, suppose horsepower has two
†Part of special issue on Economic Valuation. levels, 120 and 150 hp; color has three levels, blue, green, or
* Corresponding author phone: (412)648-7602; fax: (412)648-2605; red; and price has four possible levels, $10 000, $15 000,
e-mail: eofarb@birch.gspia.pitt.edu. $20 000, or $30 000. There can then be 2 × 3 × 4 ) 24 different
10.1021/es990727r CCC: $19.00  2000 American Chemical Society VOL. 34, NO. 8, 2000 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 1407
Published on Web 03/03/2000
cars to choose from. Suppose we list all possible autos as or scenario has some price attached to it, and the subject is
follows: asked the binary choice question whether they would accept
the scenario at that price. CVM is frequently referred to as
alternative horsepower (X1) color (X2) price (X3) a “referendum” method, since the option to the subject is
select or reject the scenario, as in voting. The voting analogy,
1 120 blue $10 000 in fact, is often claimed as an advantage of CVM for valuing
2 150 blue $10 000 public goods, whose delivery is likely subject to some direct
3 120 green $10 000
. . . .
or indirect voting process.
. . . . The primary purpose of CJ is to determine the structure
24 150 red $30 000 of preferences that underlie the judgments of goods or
scenarios. This can be accomplished by either presenting
the individual with a set of alternative goods or by presenting
Next, we ask the individual to make judgments of choices
attributes of goods. In the auto example, preferences are
among the alternatives. We could ask the person to rate or
comprised of the attributes money, horsepower, and color.
rank these alternatives from most to least preferred, called
CJ would seek to establish the relative importance of these
“full-profile” evaluation. Or we could ask the individual to
independent attributes. It would do so by designing prefer-
compare all alternatives against one another (276 compari-
ence-type experiments, such as rating or ranking scenarios
sons), called “discrete choice” evaluation (14), perhaps asking
with varying levels of the attributes, whereby subjects state
him/her to note the strength of preference. This information
their preferences across all alternatives or by using choice-
can then be used to establish the utility function using a
type experiments in which subjects are asked to choose a
variety of statistical procedures (7).
most preferred scenario among two or more alternatives
Suppose the estimated utility function is
when levels of attributes differ across scenarios (10).
The economist’s interest in CVM and CJ stems from the
U(X) ) 500X1 + 1000X2 - 10X3 (3)
ability to use these techniques to determine the value that
This function not only establishes the relative weights individuals or groups place on nonmarketed goods and
attached to each attribute but also establishes the tradeoffs services, such and environmental and resource services.
between attributes. For example, when horsepower (X1) Suppose that under scenario 1 the individual or group can
increases by 1 unit, utility remains the same if the price (X3) obtain marketed goods at a price of P with an income of Y,
of the car increases by $50. This implies that the value of a and it has a fixed quantity (quality) of public good, Q1; under
unit of horsepower to the individual is $50; i.e., this is the scenario 2, the individual or group faces the same market
maximum amount of money the person would give up to prices and income but has Q2 of the public good. Let the
obtain a 1 unit increase in horsepower. This $50 is what individual or group utility function U be
economists refer to as the shadow value, willingness to pay
(WTP), or marginal benefits of the attribute horsepower. U ) U(P, Q, Y) (4)
This oversimplified example is used to explain the essence
of CJ. There are considerable complexities in selecting The economic valuation of scenario 2 relative to scenario 1
experimental designs for choice situations. These include is an amount of money, compensating surplus (CS), such
defining the attributes and the levels in ways that ensure a that
comprehensible number of alternatives and attributes (7)
and choosing between judgment (ranking or rating) versus U(P, Q2, Y - CS) ) U(P, Q1, Y) (5)
discrete choice analyses. In any case, CJ uses individual or
group choice among multi-attribute alternatives to provide CS is the maximum amount of money that would be given
evidence of relative valuations of attributes. This is the up to make the utilities of the two scenarios equal. Of course,
usefulness of the procedure to economists in valuing scenario 2 may be worse than scenario 1 and CS would be
environmental goods and services. The environment and negative, requiring compensation.
ecosystems provide complex multi-attribute values to people; Economists are concerned that attempts to determine
e.g., forests provide timber, habitat, soil protection, hydrologic CS employ techniques and estimating methodologies that
management, climate moderation, etc. Each attribute has a are consistent with developed utility theory, i.e, are utility
unique shadow value. In addition, hypothetical multi- theoretic (10). Choice-type experiments using CVM or CJ
attribute scenarios of CJ enable the determination of prefer- techniques meet this requirement and are consistent with
ence structures for scenarios that subjects may not be familiar recently developed random utility models (RUM) of choice.
with or for nonuse values. They permit the determination of Under RUM, choice is a probabalistic behavior relying on
part-worths of attributes that are impossible to identify in both systematic and random elements. The utility function
real world situations, perhaps because those attributes may takes the form
be highly correlated, such as clean air and panoramic views
or because we cannot infer value from use in the case of U ) V(P, Q, Y) +  (6)
nonuse values. CJ allows more generic determinations of
preference structures independent of specific contexts. where V(.) is the systematic element and  is a random
The distinctions between CVM and CJ are subtle but element in choice. The probability that an individual or
important. The primary purpose of CVM is to “elicit people’s group will chose scenario 2 over scenario 1 is then given by
preferences for public goods by finding out what they would
be willing to pay for specified improvements in them.” (5). Pr{V(P, Q2, Y) + 2 > V(P, Q1, Y) + 1} or
It is called “contingent” because the WTP is contingent upon Pr{V2 - V1 > 1 - 2} (7)
the hypothetical good or scenario described to the subject.
The method presents the subject with a characterization of where Vi refers to systematic utility under scenario i. If the
the good being valued and asks directly, “What would you error terms are distributed in a particular way (identically
pay?” or “Would you pay $X?” for the good. In the automobile and independently distributed with a type I extreme value
example, CVM would ask “What would you pay for a 150 hp, distribution), the probability of choosing scenario 2 over
green car?” The good or scenario itself is directly valued, scenario 1 is given by the conditional logit probability
either by asking what a person would pay for it or the good function:

1408 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 34, NO. 8, 2000


Pr{2 chosen over 1} ) exp(V2)/Σ expVj (8) asked to select the most preferred in a paired choice. Specific
sites could be scored using the estimated utility function
where Vj is the nonrandom element of utility for scenario j weightings of attributes. Estimates of individual utilities could
() 1, 2 in this simple case) (1). After specifying the functional be used to determine the percentage of the voters who would
form of V, maximum likelihood econometric techniques can favor a given site location. Although they do not suggest it,
be used to estimate the parameters in V. Once V is known, the estimated utility function could also be used to determine
CS can be determined. For example, when V ) β1 + β2P + compensation necessary to offset site location decisions.
β3Q + RY, CS for a representative individual in a sample can CJ responses to choices can be expressed as ratings (e.g.,
be determined from eq 5: scale 0-10), rankings (e.g., most preferred to least preferred),
and simple binary choices (e.g., prefer scenario 1 to scenario
2) with or without intensity of preferences. Johnson and
CS ) -β3(Q2 - Q1)/R (9)
Desvousges (22) used a binary choice CJ, with intensity of
preferences, to estimate the structure of preferences for
The aggregation of CS measures across choice options has various energy programs. Using nine attributes, such as fish
been worked out by Hanemann (15). Establishing a choice consumption bans, respiratory cases, electric bill, sugar maple
experiment, using either CVM or CJ, in conjunction with damage, etc., they presented respondents with two scenarios
estimation of a utility-theoretic utility function permits the and asked them to rate their choice; e.g., strongly prefer
measurement of the monetary valuation of environmental scenario 1 to scenario 2, moderately prefer scenario 1 to
quality changes. scenario 2, etc. The economic value of policy alternatives
could be evaluated since electric bills differed across
III. How Have Economists Applied CJ to Environmental alternatives. For example, if an energy policy requires a fish
Valuation? consumption ban on a lake, people would be willing to incur
The ability to quantify individual or group preferences toward a 1.2% increase in their electricity bill per month to avoid the
attributes of the environment or ecosystems provides a useful ban.
avenue for the determination of monetary compensation or Economists are beginning to test the applicability of
in-kind compensation for damages as well as useful measures different types of CJ specifications to valuation problems.
of benefits and costs of environmental enhancements or One concern is whether ratings, rankings, or binary choice
degradation. CJ is particularly useful in this regard as it provides the most useful valuation information. Mackenzie
permits the design of experiments resulting in measurements (23) used each of the three CJ choice methods to determine
of acceptable tradeoffs between ecosystem services and the structure of preferences for waterfowl hunting. He found
between services and money. CJ has received approval by that the ratings model yielded the most efficient estimates
NOAA as an acceptable method for the determination of of valuations, meaning lower variances in estimates. Roe et
in-kind compensation for resource damages (16). al. (24) also tested for differences in using CJ rankings versus
One of the first uses of CJ in quantifying environmental ratings in valuing salmon management options. Their
preferences was a study of water quality (17). This study experimental design required individuals to rate four man-
used a ranking of water quality and antipollution tax agement scenarios on a scale of 1-10. These ratings were
alternatives to estimate a utility function. The majority of transformed into ratings differences from the status quo, full
respondents preferred the swimmable water quality ac- rankings (highest ratings ranked first, etc.) and binary
companied by an annual $240 tax alternative. Researchers rankings (ratings higher than status quo versus ratings lower
have used the ranking procedure to value visibility improve- than or equal to status quo). They used tobit procedures to
ments in national parks (18). Lareau and Rae (19) used CJ estimate ratings and ratings differences functions, ordered
to estimate the WTP for diesel odor reductions in Philadel- logit to estimate full rankings functions, and binary logit
phia. They presented respondents with a set of alternatives, procedures to estimate binary rankings functions. Ratings
varying in the number of weekly odor contacts for two differences and binary rankings yielded significant and
different odors (inhaled by respondents prior to the choice utility-theoretic correct coefficient estimates on the prices
experiment) and in the annual costs associated with odor of alternatives, while ratings and full rankings did not. The
reductions. Respondents were asked to rank 12 alternatives. derived measures of CS differed considerably across mea-
Smith and Desvousges (20) used CJ to estimate the surement scales. The conclusion is that measurement scales
valuation of water quality improvements for a river in make a difference in determining the form of utility functions
Pennsylvania. They used a water quality “ladder” developed and valuation measures. However, the experimental design
by Resources for the Future, where water quality is defined was to derive other measurement scales from the initial
on the basis of safe activities, including boating, fishing, ratings data rather than to test whether there would be
swimming, and drinking. They presented respondents with differences if the initial measurement scales were ratings,
four alternatives, each representing a different water quality rankings, or binary rankings.
and accompanying annual payment. Respondents were asked While there remains controversy over whether SP tech-
to rank the alternatives. Estimates ranged from roughly $40 niques provide useful valuation information due to their
to $80 per year per household for improving water quality hypothetical context (25), concerns have been expressed that
from boatable to fishable. CJ may suffer from the same problem relative to CVM. For
Opaluch et al. (21) used a discrete choice CJ to determine example, Roe et al. (24) note that while CVM asks people
the preference structure for siting a municipal landfill. Focus whether they would “buy” a good or scenario, CJ only
group sessions suggested the decision process be broken determines the structure of underlying preferences and not
into two parts: one for the general location and another for whether the person would behave on the basis of those
the specific site position. There were six location attributes: preferences. They suggest combining CJ with contingent
number of homes, presence of park land, presence of farm behavior questions.
land, presence of an elementary school, use of local roads, Adamowicz et al. (26) used a procedure to combine
and annual cost to each household. There were seven site information from CJ experiments and RP for recreation site
attributes: acres of marsh, acres of woods, acres of farmland, choice. They applied the procedure to derive preference
wildlife habitat, pond acreage, groundwater quality, and structures for recreational activities in western Canada. The
annual household payment. Respondents were presented CJ choice experiment asked respondents to choose one of
with abstract pictures of the alternative scenarios and were three scenarios: using one of two recreational sites or staying

VOL. 34, NO. 8, 2000 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 1409


home. Their estimation model allowed determination of the
additional distance a user will travel to a site if recreational
opportunities at the site are enhanced. Using the monetary
cost of travel, they valued recreational improvements from
these additional distance measures. This is interesting, since
the payment for the scenario is not directly monetary but
can be inferred from distances persons would be willing to
travel for improved site conditions. Combining CJ and RP
information to jointly estimate preference structures and
welfare measures shows promise, combining the advantages
of each and offsetting their individual weaknesses
As researchers have claimed that choices are based on
FIGURE 1. Typical choice question.
people’s perceptions rather than purely objective measures,
Adamowicz et al. (27) further explored whether objective or are severely polluted, moderately polluted, and unpolluted.
perceptual measures of attributes make a difference in A severely polluted stream has been determined by biologists
estimates of preference structures and welfare measures. By to be incapable of supporting fish and other organisms.
focusing on increasing moose populations, they found that Fishing conditions would be poor to nonexistent. Wildlife
welfare measures differed substantially. They suggested that could not rely on these streams for food and habitat. However,
perceptions of moose populations were higher than was human health is not typically affected. Moderately polluted
objectively the case. This resulted in lower perceived marginal streams will support only some fish and other organisms.
utility of moose population increases than objective marginal Reproductive conditions for fish are poor, and fishing is
utility measures, leading to much lower welfare estimates supported but catch would be limited. Unpolluted streams
using perception as compared to objective data. are those where fish and other organisms can thrive. By these
Like CVM, CJ can be used to assess values of environmental designations, Loyalhanna Creek is currently moderately
enhancements or degradation for both use and nonuse polluted, and the Conemaugh River is severely polluted.
(passive) values of the environment. Nonuse value is not The survey was accompanied by a brief letter presenting
reflected in revealed behaviors, such as purchases, trips, etc. commonly asked questions and answers about water quality
Adamowicz et al. (28) used CJ to estimate nonuse preferences in the study region. A map of the streams in the watershed,
for the threatened woodland caribou population in Alberta, shaded according to the degree of degradation, followed the
Canada. The choice experiment included seven attributes, introductory letter. A major difficulty was defining the “good”
including wilderness area, recreation restrictions, moose for the choice alternatives. After discussions with fish and
populations, etc. Income taxes were the payment vehicle for wildlife managers, we decided to use a mixed habitat-
financing the caribou management scenarios. Using a RUM, fisheries characterization. The resulting characterization of
they estimated both CVM and CJ choice models. Interestingly, severely polluted, moderately polluted and unpolluted, based
a comparison of the two models showed similar parameter on survivability of fish and other organisms, was described
estimates for the choice attributes common to the two models to survey recipients. The areas of stream valuation were
(income) but very different welfare measures. illustrated on the map provided. Instructions to recipients
Boxall et al. (29) also found that CJ and CVM welfare noted that stream cleanup would be costly, resulting in higher
measures differed substantially, by an order of magnitude, prices or taxes. A typical choice question is illustrated in
in their recreation application. The CJ valuations were similar Figure 1.
to RP (travel cost) estimates obtained in a related study. This As Figure 1 illustrates, the choice involved three at-
suggests that CJ and CVM may yield very different welfare tributes: stream A quality, stream B quality, and a price.
estimates. This may be because CJ sets up scenarios in which Stream A had two levels: moderately polluted and unpolluted.
individuals explicitly see alternatives, while a CVM scenario Stream B had three levels: severely polluted, moderately
does not present a variety of alternatives. Substitution may polluted, and unpolluted. In addition, five price levels were
be an important factor in some resource valuations, making selected. There were five possible improvement combina-
CJ a more realistic method in those contexts. tions, and each respondent was given all five as choice options
to the status quo. Prices were varied across respondents.
IV. A Conjoint Application to Watershed Quality in Sampling of respondents came from zip codes in grids
Western Pennsylvania constructed around the stream sites with higher sampling
Stream quality is a severe problem in western Pennsylvania, rates for grids closest to the sites. A sample of 3958 households
with over 580 of 1100 assessed stream miles not meeting was drawn, and first class postage surveys were mailed to
state water quality standards, and 430 of those nonattainment them. Because of a limited budget, only one mailing was
miles due to acid mine drainage. This study used CJ to value sent. We did not follow the Dilman (30) mailing procedure
stream quality improvements to users and nonusers for two of repeat contacts. A total of 510 households responded, while
streams in western Pennsylvania. The subbasins in which 372 surveys were returned as undeliverable. This was only
the two streams of interest are located, Loyalhanna Creek a 14% response rate, excluding undeliverables, which is low.
and Conemaugh River, had 60% and 30% of their stream However, this rate is not far below that of other general
miles degraded, respectively. A mail survey was administered population CJ samples (3, 2) when considering the fact there
during 1996 to a random sample of residents living within were no attempts to predetermine who was willing to
90 mi of the evaluated streams. Individuals were presented respond. Only 367 of the 510 returned surveys were usable
with several choice scenarios and were asked to choose as some respondents did not answer all valuation related
between the status quo and various combinations of stream questions, particularly income.
quality improvements for the two streams. Each alternative Suppose the individual has an interactive RUM utility
had a price attached. The marginal value of each type of function of the form
stream quality improvement was estimated using binary
choice (BC) and intensity of preference (IP) responses in a U ) β1 + β2P + β3Q + β4QY + R(Y - w) +  (10)
RUM framework.
Habitat evaluation by the U.S. EPA and Pennsylvania State where P is the price of goods and services, Q is the
fisheries biologists distinguishes between stream reaches that environmental quality condition of the streams, I is income,

1410 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 34, NO. 8, 2000


and w is the fixed payment (price) for a choice alternative.
It is expected that β3 and R are positive and that β4 is positive TABLE 1. Estimation Results for Binary Choice (BC) and
if increased incomes enhance the utility of stream quality Intensity of Preferences (IP) Logit Random Utility Modelsa
improvements. In the choice experiment, P and Y are constant coeff for variable BC (1) IP (2)
across alternatives, so when comparing the utility of two
alternative scenarios constant -0.3959 1.1864
(0.03) (0.00)
∆q1, A: MfUb 0.0953 0.2463
y ) ∆U ) β1∆Q + β2 ∆QY - R(∆w) + (2 - 1) (11) (0.60) (0.08)
∆q2, B: SfMb 0.0311 0.2080
where ∆Q ) Q2 - Q1, etc. Under binary choice (BC), if we (0.89) (0.22)
let ∆q3, B: SfUb 0.2436 0.3011
(0.26) (0.07)
y ) 1 if ∆U > 0; select scenario 2 over 1 ∆w -0.0077 -0.0056
(0.00) (0.00)
) 0 if ∆U e 0 ∆q1I 0.0437 0.0101
(0.17) (0.67)
and i is distributed in a certain manner (Weibull), McFadden ∆q2I 0.0854 0.0418
(31) has shown that a logit model can be used to estimate (0.02) (0.12)
the coefficients in eq 11. ∆q3I 0.1043 0.0803
(0.00) (0.00)
The binary choice RUM outlined above can be extended
µ2 nac 0.9020
to incorporate the IP for a choice using ordinal response (0.00)
categories. An IP model has been used by Bergstrom et al. µ3 na 1.7465
(32) to estimate demand functions for local school expen- (0.00)
ditures, by Magat et al. (33) to value morbidity risk reductions µ4 na 2.8520
associated with chemical products, by Kamakura and Bala- (0.00)
sundaram (34) to estimate the demand for electric utility log (L) -1001.1 -2709.9
products and services, and by Colias and Salazar-Velasquez a Probability values in parentheses, N ) 367. b The symbolism A:
(35) to estimate purchase intent for new products. [Like the MfU represents the hypothetical qualitative change in stream A
binary choice RUM, ∆U defined in eq 11 is not directly (Loyalhanna) from moderately polluted to unpolluted; S for stream B
represents severely polluted. c na, not applicable.
observable. However, we can assign the following values to
y:

yij ) 0 if ∆U < 0; TABLE 2. Estimated Marginal Valuations of Compensating


the person chooses scenario 1 over scenario 2 Surplus (CS) for Stream Improvements Using Binary Choice
(BC) and Intensity of Preferences (IP) RUMsa
) 1 if 0 e ∆U < µ1; all samples (N ) 367) nonusers (N ) 70)
the person moderately prefers scenario 2 to scenario 1
stream improvement BC (1) IP (2) BC (3) IP (4)
) 2 if µ1 e ∆U < µ2; A: MfU $35.76 $51.35 $3.05 $12.55
the person strongly prefers scenario 2 to scenario 1 (15.89) (17.20) (35.65) (48.02)
B: SfM $49.62 $67.64 $1.39 $47.61
) P if µP e ∆U; (15.55) (22.82) (64.33) (64.33)
B: SfU $87.43 $112.44 $38.87 $54.22
the person very strongly prefers scenario 2 to scenario 1
(21.78) (23.13) (49.59) (63.42)
where there are P+1 ordinal response categories and µp a Values represent household WTP for Each of 5 years; standard

represents the unobserved threshold for response category errors of estimates in parentheses. Evaluated at sample mean household
income.
p. The single 0 threshold of the BC model is replaced by P
thresholds in the IP model. A presumed advantage of
permitting choice response intensities is that it may more moderately polluted to unpolluted, depending upon which
accurately mimic mental heuristics in hypothetical choice estimation model is used. The IP model shows higher
situations as opposed to real choice situations where a 0,1 valuations than the BC model for all quality changes; offering
choice is actually made. This may lead to less rejection of the respondents the ability to express degrees of preference
hypothetical choice experiment, i.e., protest responses.] appears to reduce their rejection of more costly choice
The estimation results for eq 11 using both BC and IP are alternatives. The WTP for cleanup of the more severely
shown for the complete sample in Table 1, columns 1 and degraded stream B is greater than for stream A, which we
2. Quality changes, ∆qi, are discrete with ∆q1 representing would expect. Table 2 shows that WTP ranges from roughly
the improvement of stream A from moderately polluted to $50 to $68 annually for 5 years to improve stream B from the
unpolluted, etc. All coefficients have the expected sign. The current severely polluted status to moderately polluted and
IP model appears to improve the significance of the quality from $87 to $112 annually for 5 years for an improvement
change variables, suggesting that allowing individuals to to an unpolluted condition.
respond to choices with degrees of preference might yield Equation 11 was also estimated for persons who desig-
more meaningful evaluations of willingness to accept quality nated that they had not used either stream during the past
improvements than forcing them into a binary “accept/reject” year. The resulting CS estimates are shown in columns 3 and
choice. 4 of Table 2. Estimates of WTP for nonusers are clearly less
We can use eq 9 to determine the compensating surplus than for the sample as a whole, as might be expected. For
(CS) or WTP for the three types of stream quality improve- example, the typical nonuser household CS for improving
ments used in the study. Table 2 shows estimated CSs, using stream A ranges from only $3 to $13 per year for 5 years.
mean household incomes in the sample, for each quality However, like users, nonusers are willing to pay more for
change. These results suggest that households would be greater stream quality improvement, as the CSs for stream
willing to pay between $35 and $51 per year for a 5-year B show. The standard errors of the estimated CSs are also
period in order to have stream A quality improved from larger than for the sample as a whole, suggesting less

VOL. 34, NO. 8, 2000 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 1411


confidence in the CS estimates for nonusers than for the (7) Green P. E.; Srinivasan, V. J. Marketing 1990, 54 (4), 3-19.
sample as a whole. (8) Green P. E.; Rao, V. R. J. Marketing Res. 1971, 8, 355-363.
The purpose of this CJ study was to obtain estimates of (9) Wittink, D. R.; Cattin, P. J. Marketing 1989, 53, 91-96.
(10) Hensher, D. A. Transportation 1994, 21, 107-33.
quality improvements for only two streams. Water quality (11) Louviere, J. J.; Hensher, D. A. Transportation Res. Rec. 1982,
degradation was not the policy issue, so valuing WTP to avoid 890, 11-17.
or willingness to accept compensation for degradation was (12) Ryan, M.; McIntosh, E.; Shackley, P. Health Econ. 1998, 7 (4),
not an appropriate valuation concept, although CJ could be 373-378.
used for valuing degradation. The intention of the study was (13) Ryan, M.; Hughes, J. Health Econ. 1997, 6 (3), 261-273.
(14) Louviere, J. J. J. Transport Econ. Policy 1988, 22 (1), 93-119.
not to establish the underlying preference structure of
(15) Hanemann, W. M. Applied Welfare Analysis with Qualitative
respondents toward the quality of streams, in general. If it Response Models; Working Paper 241; University of California:
were, a different design for alternatives would have to be Berkeley, CA, 1984.
used. For example, preferences toward streams may have (16) Fed. Regist. 1995, 60 (149), 39804-39834.
included use dimensions, such as fishability and swimma- (17) Whitmore, G. A.; Cavadias, G. S. Decis. Sci. 1974, 5 (4), 614-631.
bility; aesthetic dimensions, such as clarity and color; and (18) Rae, D. A. The Value to Visitors of Improving Visibility at Mesa
Verde and Great Smoky Mountain National Parks. In Managing
health dimensions, such as bacterial condition. Application Air Quality and Scenic Resources at National Parks and
of CJ to determine general preference structures would then Wilderness Areas; Rowe, R. D., Chestnut, L. G., Eds.; Westview
require determination of appropriate dimensions of utility Press: Boulder, CO, 1983.
and salient measures to be used in representations of the (19) Lareau, T. J.; Rae, D. A. South. Econ. J. 1989, 55 (3), 728-742.
alternatives. Focus group sessions could determine these (20) Smith, V. K.; Desvousges, W. H. Measuring Water Quality Benefits;
dimensions and measurement. Kluwer-Nijhoff: Boston, 1986.
(21) Opaluch, J. J.; Swallow, S. K.; Weaver, T.; Wessells, C. W.;
Wichelns, D. J. Environ. Econ. Manage. 1993, 24 (1), 41-59.
IV. Conclusion (22) Johnson, F. R.; Desvousges, W. H. J. Environ. Econ. Manage.
CJ is one of many tools economists can use to value resources 1997, 34 (1), 79-99.
and environmental conditions. While suffering from the same (23) Mackenzie, J. Am. J. Agric. Econ. 1993, 75 (3), 593-603.
(24) Roe, B.; Boyle, K. J.; Teisl, M. F. J. Environ. Econ. Manage. 1996,
concerns as CVM regarding the hypothetical nature of the 31 (2), 145-159.
valuation, it offers some advantages. For example, it offers (25) Hausman, J. A. Contingent Valuation: A Critical Assessment;
the opportunity to explicitly determine tradeoffs in envi- Contributions to Economic Analysis Vol. 220; Elsevier: New
ronmental conditions through its emphasis on discovering York, 1993.
preference structures and not just monetary valuation. This (26) Adamowicz, W. L.; Louviere, J.; Williams, M. J. Environ. Econ.
may be especially important when valuing ecosystems, which Manage. 1994, 26 (3), 271-292.
(27) Adamowicz, W. L.; Swait, J.; Boxall, P.; Louviere, J.; Williams, M.
provide a multitude of joint goods and services; it more J. Environ. Econ. Manage. 1997, 32 (1), 65-84.
reasonably reflects multi-attribute choice contexts than the (28) Adamowicz, W. L.; Boxall, P. C.; Williams, M.; Louviere, J. Am.
typical unidimensional CVM. It may also be important when J. Agric. Econ. 1998, 80 (1), 64-75.
the objective is to determine preference structures in a general (29) Boxall, P. C.; Adamowicz, W. L.; Swait, J.; Williams, M.; Louviere,
manner independent of particular contexts. It offers ranking J. Ecol. Econ. 1996, 18 (3), 243-253.
and rating measurement scales rather than just binary choice. (30) Dilman, D. A. Mail and Telephone Surveys-The Total Design
Method; Wiley: New York, 1978.
On the other hand, it may be more complicated to administer (31) McFadden, D. Conditional Logit Analysis of Qualitative Choice
and more difficult for respondents to understand. Behavior. In Frontiers in Econometrics; Zarembka, P., Ed.;
Academic Press: New York, 1974, pp 105-142.
Literature Cited (32) Bergstrom, T. C.; Rubinfeld, D. L.; Shapiro, P. Econometrica
1982, 50 (5), 1183-1205.
(1) Freeman, A. M. The Measurement of Environmental and Resource
(33) Magat, W. A.; Viscusi, W. K.; Huber, J. J. Environ. Econ. Manage.
Values: Theory and Methods; Resources for the Future: Wash-
1998, 15, 395-411.
ington, DC, 1993.
(2) Kopp, R. J.; Smith, V. K. Valuing Natural Assets: The Economics (34) Kamakura, W. A.; Balasundaram, A. Designing Innovative
of Natural Resource Damage Assessment; Resources for the Products Or Services With A Reservation-Utility Model; Working
Future: Washington, DC, 1993. Paper; Katz Graduate School of Business, University of Pitts-
(3) Tietenberg, T. Environmental and Natural Resource Economics; burgh: Pittsburgh, 1993.
Addison-Wesley: Reading, MA, 2000. (35) Colias, J. V.; Salazar-Velasquez, C. Can. J. Marketing Res. 1995,
(4) Daily, G. C. Nature’s Services: Societal Dependence on Natural 14, 46-56.
Ecosystems; Island Press: Washington, DC, 1997.
(5) Mitchell, R. C.; Carson, R. T. Using Surveys to Value Public Received for review June 30, 1999. Revised manuscript re-
Goods: The Contingent Valuation Method; Resources for the ceived November 22, 1999. Accepted December 1, 1999.
Future: Washington, DC, 1989.
(6) Luce, R. D.; Tukey, J. W. J. Math. Psychol. 1964, 1, 1-27. ES990727R

1412 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 34, NO. 8, 2000

You might also like