Revised August 1985
Option Price Estimates for Water Quality Improvements:
A Contingent Valuation Study for the Monongahela River1
William H. Desvousges
Center for Economics Research
Research Triangle Institute
Research Triangle Park, NC 27709
V. Kerry Smith
Department of Economics and Business Administration
Vanderbilt University
Nashville, TN 37235
Ann Fisher
Office of Policy Analysis
U.S. Environmental Protection Agency
Washington, DC 20460
Running Head: Option Price for Water Quality
Direct correspondence to: William H. Desvousges
Center for Economics Research
Research Triangle Institute
P.O. Box 12194
Research Triangle Park, NC 27709
1
-------
ABSTRACT
This paper presents the findings from a contingent valuation survey
designed to estimate the option price bids for the improved recreation
resulting from enhanced water quality in the Pennsylvania portion -of the
Monongahela River. The findings are based on a survey design that used
professional interviewers to conduct personal interviews determined from a
representative sample of 393 households. In addition, the research suggests
that protest bids and outliers be viewed similarly. Accordingly, a new
technique for identifying outlying responses is proposed. The findings
suggest that the question format affects the option price estimates and that
criteria for determining the final sample of responses have an important
i nfluence on contingent valuation results.
2
-------
1. INTRODUCTION
Recent Federal and State policies that require benefit-cost analyses
of major regulations have helped to focus attention on measuring nonuse
benefits.2 While the available empirical evidence suggests that these
:an be a significant share of the total benefits provided by envi-
resources, important conceptual and empirical issues remain to be
resolved before nonusevcan become a standard component of most benefit-cost
studies.3
Perhaps the most important conceptual issues stem from our changing
understanding of benefit concepts under uncertainty. For example, the en-
vironmental 1 iterature initially focused on option value—i.e., the differ-
ence between option price and the expected consumer surplus—as an "omitted
component" of the benefits provided by changes in unique environmental re-
sources. However, option value now is considered to indicate the impor-
tance of the selection of an ex ante versus an ex post perspective in the
definition of benefits resulting from a change in some dimension of envi-
ronmental quality under uncertainty.4 If an individual's utility function
uMccs'ia.'t >v ecc u
is state dependent (i.e., it may differ depending on outcome^- of-unce^
then the perspective for welfare measurement will affect the role
of the marginal utilities of income for each state in defining benefits
(see Smith [31]).
In this paper, we use option price as our ex ante measure of benefits
under uncertainty. In our water quality application, option price is the
maximum annual payment that an individual is willing to make now for access
to the river with improved water quality, regardless of use. Although op-
3
-------
icantly affected by question format. Finally, there was little evidence of
interviewer bias.
The remainder of this paper is organized as follows. Section 2 high-
lights the definition of option price and how our survey and questionnaire
design implemented 1t. Section 3 examines the implication of both protest
*
and outlying bids as sample screening rules that can affect the analysis of
contingent valuation responses. Equally important, it proposes a new pro-
cedure for determining outlying bids. Section 4 presents the option price
results, along with examining the effect of outliers on the shape of the
distribution of option prices. Section 5 describes the findings of our
tests for framing effects, including the effects of both different question
formats—bidding games, payment cards, and direct questions—and interview-
ers. Section 6 summarizes our main findings and discusses their implica-
tions.
5
-------
. 2. SURVEY AND QUESTIONNAIRE DESIGN
The survey questionnaire elicited an individual's annual option price
bid for water quality changes^j-n—o port4^-^f^he-Monongafre4a--
Ww. Our definition of option price is based on the "timeless" framework.1,
S4-Ree~dsta-H ed-T5\rtews'_af--fches«-j-ecm€-e^ua4--dev€Ji-o|>fflefrt^—av^4-l -
TsWe7-rrr-ttrts--setrtien -we *-def4ne—ept+on-^M-e-e--and-de tw-be-ttre -rfes+gn-
of-"1^*l^lYTonnaire~Ui»eU Lu elicit »t.fi—3fiv-a<}
-------
These functions can be regarded as indirect utility functions, with the
prices of all other goods and services held constant.7 Our specification
assumes that if the option price is paid, then consumption of the environ-
mental service can be at any level desired. What is important for our pur-
poses is the role of income in this relationship. The responsiveness of
consumer surplus to income, as well as the nature of the demand uncertainty,
help determine the bounds for option price.8 This implies that our analysis
of option price bids should consider their sensitivity to respondents' in-
come levels.
In our survey design using option price, trained professional inter-
viewers conducted a household survey based on a single-stage, stratified-
cluster sample of 393 households from the five-county Pennsylvania portion
of the Monongahela River basin, including the Pittsburgh SMSA. With the
household defined as the unit of analysis, the interviewers randomly selec-
ted a respondent 18 years or. older from a roster of individuals in the
household.9 The sample design ensured a representative sample of the target
population. The interviews were conducted in November and December 1981;
301 were usable, resulting in an 80-percent response rate.
Before the option price concept was introduced to the respondent, the
questionnaire established the general framing of the conti ngent market.
This orientation started by eliciting recreation information about the use
of the Monongahela and other water-based recreation areas. These questions
also helped to establish rapport with the respondent. The framing process
continued by introducing the market setting: improved water quality for
the Monongahela River.
7
-------
Following this, introduction, the interviewer handed the respondent our
vehicle for relating water quality to the feasible recreation activities —
the Resources for the Future (RFF) Water Quality Ladder developed by
Mitchell and Vaughan (in. Mitchell and Carson [21]). The interviewer des-
cribed the ladder and used it to link water quality levels to recreation
acti vi ties.
A second important element in the hypothetical market involved identi-
fying the reasons why an individual might value water quality changes. The
interviewer used a second visual aid--the value card—to describe user,
option, and existence values. elicited coethe importance of actual
Pffyx J
use, potential use, and no use in each valuation of water qual ity.
A-
These attitudinal questions reinforced the concepts, provided a break in
the discussion, and yielded an additional check for consistency in re-
sponses.
The payment vehicle was introduced to the respondent as follows:
Now, we would 1 ike for you to think about the relationship
between improving the quality of water in the Monongahela River
and what we al1 have to pay each year as taxpayers and as consum-
ers. We al 1 pay riirectiy through our tax dollars each year for
cleaning up all rivers. We also pay indirectly each year through
higher prices for the products we buy because it costs companies
money to clean up water they use in making their products. Thus,
each year, we are paying directly and indirectly for improvements
in the water quality of the Monongahela River.
I want to ask you a few questions about what amount of money
you would be wi 11 ing to pay each year for different levels of
8
-------
water quality in the Monongahela River. Please keep in mind that
the amounts you would pay each year would be paid in the form of
taxes or in the form of higher prices for the products that com-
panies sell.
This payment vehicle is not problem free because the economic activi-
ties involving water are simplified. For example, it ignores the possibil-
ity that some companies could experience lower production costs if the water
were cleaner. In addition, it does not develop explicitly the share of
costs companies pass on to consumers relative to the share borne by stock-
holders. The points could have been clearer with better wording and a
visual aid. One wording problem stemmed from inadvertently mixing "and"
and "or" in the description of the vehicle. This could have confused some
respondents, although the interviewers did not mention it in a debriefing
session.
Despite these qualifications, the payment vehicle does have its ad-
vantages. It avoids the problem of the implicit starting point that ham-
pers increased water bills or sewage fees, a point made by Mitchell and
Carson [21] in their critique of Greenley, Walsh, and Young [17]. This same
problem seems to appear in Daubert and Young [12] when they used these two
alternatives. Moreover, the payment vehicle is credible--it corresponds
reasonably well to how people actually pay for improved water quality.
The final task in the framing of the contingent market was to elicit
the option price bids for specific levels of water quality. The central
methodological feature of the questionnaire is the comparison of alterna-
tive question formats. For this comparison, we divided the sample into
fourths and gave a different survey instrument to each group. The four
9
-------
questioning modes were: the direct question method both with and without a
payment card and the iterative bidding games with $25 and $125 starting
points. Thus, the survey design provided the information necessary for ex-
plicit tests of starting-point bias and for differences between the direct
question and the iterative bidding formats.
The payment card used in the direct question method was simply an array
of numbers representing annual amounts from $0 to $775 per year in $25
increments. This format contrasts with the Mitchell and Carson [21] pay-
ment card, where the amounts were adjusted based on the income level of the
respondent.
Although the wording was the same both for the direct question and pay-
ment card formats, the payment card was shown to respondents in the latter
case. The process was very simple, with the interviewer asking the respond-
ent for an amount for each water quality level and stressing that additional
amounts are being requested. The water quality ladder and the value card
were available to the respondent during these questions. The principal dif-
ference between the two bidding games was their starting points. In each
bidding game, the interviewer initiated the market process at the starting
point and increased or decreased the requested amount until the respondent's
maximum value was obtained. This was repeated for each of the water qual ity
levels, emphasizing the additional nature of the amounts for higher levels
of water quality.
In the hypothetical market, each respondent was asked to provide an
option price for three water quality levels:
10
-------
Avoiding 9 decrease in water quality in the Monongahela River
from (D) to (E), or from boatable to unsuitable for any
water-based activities (including boating).
Raising the water qual ity from (0) to (C), or from boatable
to a level where gamefish would survive.
Raising the water qual ity from (C) to (B)» or from fishable
to a level where individuals could use the river for swim-
ming.
The option price amounts are based on the Hicksian surplus measures,
using the equivalent surplus measure for the loss of the recreation services
of the Monongahela River (Level D to Level E) and the compensating surplus
measures for the improvements to the fishable and swimmable water quality
levels. These measures correspond roughly to the existing property rights
for the overall level of Monongahela recreation services. Although several
sections of the Monongahela are capable of supporting sport fishing because
of the influence of tributaries, the overal 1 water qual ity of the river cor-
responds to the boatable level. Moreover, most of the survey respondents
lived in or near Pittsburgh, which suggests that their experience with the
river was most likely to be consistent with the boatable designation.
11
-------
3. PROTEST AND OUTLYING BIDS
Estimates from contingent valuation surveys may also be affected by
the procedures used-to determine the final sample used in the analysis of
responses (Randall, Hoehn, and Tolley [24]). Using largely informal pro-
cedures, analysts have screened contingent valuation data sets to eliminate
protest bids and to identify/delete influential observations. In our view,
these procedures should stem from a common objective to detect individuals
who fall into one or more of four categories:
Category 1:
Category 2:
Category 3:
Category 4:
In screening out these individuals, we are imposing, at least implicitly, a
model of how individuals respond to contingent valuation questions. When
protest bids are identified within the survey questionnaire, we are
assuming that these responses are inconsistent with an implicit model of
behavior. Outliers can be identified only by imposing some type of model,
even the informal ones largely used In the past, on the responses.
The main objective of these selection rules is to remove observations
that would most likely lead to biased estimates of a model's parameters.
respondents who reject the framing of the con-
tingent commodity (e.g., the whole notion of \
placing values on fefro¦¦ payjaaMr-vghi c 1 o-)iktLtcuTrnrtj J
respondents who fail to take the valuation ex-
ercise seriously, thereby putting less effort
into searching their preferences
respondents who are systematically affected by
the framing of the commodity (e.g., the start-
ing point in the bidding game)
respondents who misunderstand or are incapable
of processing the information required to par-
ticipate effectively in the contingent market.
12
-------
Yet, characterizing the respondents giving protest zero bids or those class-
ified as outliers may provide information that reflects the effectiveness
of the questionnaire.
In this section, we examine the implications of both protest and out-
lying bids on the effectiveness of contingent valuation. In addition, we
suggest a new procedure for identifying outliers.
A. Protest Bids
Table I presents summary characteristics for zero bidders, protest bid-
ders, the ful1 sample, and the characteristics of the survey area popula-
tion. We constructed t-tests to examine the prospects for differences in
means between zero and nonzero bidders. In addition, we used a logit anal-
ysis to examine the potential determinants of zero bids.10 The analysis of
means indicates that nonzero bidders were on average younger than zero bid-
ders, earned higher annual family Incomes, and were more likely to have
rated the Monongahela at a particular site and to have participated in out-
door recreation during the last year. No significant differences existed
between the groups 1n terms of sex, education, their water quality rating
of the river, boat ownership, and length of residence in the area. Based
on the logit analysis, respondents were more 1ikely to bid zero if they were
older, or if they considered themselves unwilling to pay the cost of improv-
ing water quality. Those respondents receiving the bidding game with the
$25 starting point were less likely to bid zero when compared to those who
received the direct question version of the questionnaire. Generally, these
findings are consistent with those of Mitchel1 and Carson [21].
The questionnaire also elicited the respondent's reason for giving a
zero bid. which enabled us to separate protest zeros from valid zero bids.
13
-------
Table I. Mean Characteristics by Zero tod Protest Bids froa ttoi»itgabe!a Survey Versus Census Information
(1) (2) (3) (4)
Zero Nonaero Zero protest bids Total sanple (5)
Standard Standard Standard Standard Population
devi- devi- devt- devi- . - cc"»"*
Mean ation N Mean ation N Nean ation N Htan at ion N Mean
1 ivf yes, 0 if no
for ownership or
use of a boat
0.11
0.32
100
0.10
0.39
193
0.15
0.37
50
0.16
0.36
301
1 if yes, 0 if no for
participation in any
outdoor recreation
in the last year
0.30
0.49
100
0.66
0.40
193
0.50
0.50
50
0.56
0.50
301
ftuaerical rating
of the Moiranyahela
River 0 lowest - 10
highest
3.51
1.76
61
3.92
2.07
160
3.63
1.60
30
3.01
1.99
221
1 if yes, 0 if no
if rating is for a
particular site
0.07
0.26
100
0.21
0.41
193
0.10
0.31
50
0.16
0.37
301
length of residence
6.82
0.95
100
6.00
1.02
193
6.74
1.10
50
6.01
1.00
301
Vears of education
12.30
2.20
06
12.93
1.99
177
12.77
1.73
47
12.75
2.07
263
10.96
Race, 1 If white
0.94
0.23
107
0.00
0. 33
193
0.93
0.26
57
0.90
0.30
300
.92
Iik:om>
17,577
11,500
07
20,534
13,079
173
19,095
11,404
40 .
19,530
13,104
260
19,907
Aye
54.55
16.91
100
44.06
10.07
193
52.60
17.27
50
47.02
10.34
301
45. GO
Sex, 1 if Mle
0.35
0.40
100
0 37
0.40
193
0.44
- • —
0.50
rr- •=*•-_=== sss==sskss
88
=._=-!=
0.36
0.40
301
--
.47
"V S. Bureau of the Census 136], 5-county area in Pennsylvania that Includes the Monongahela River,
-------
Valid zeros—respondents who indicated that the water quality change was
not worth anything or that was all they could afford—are about half of the
total number of zeros. The 10 respondents who bid zero because that is all
they could afford tended-to be elderly persons living on limited incomes.
The protest bidders—58 out of 30i respondents--either rejected the idea of
^uaIi'Vu
putting a dollar value or some aspect of the payment vehicle. There is
S
little systematic relationship between protest bids and question format.
Hcwwe#. there are . or no socioeconomic differences between the
A
protest response* and the target population. screening
out these responses should not affect the representativeness of our sample.
@0U)
Generally, the plausible reasons for zero bids and theA»veral> rate of pro-
test bids suggest that the questionnaire was reasonably effective.
B. Identifying Outliers
Nearly all analyses of contingent valuation surveys have used some
judgmental procedure to eliminate some bids from the ful 1 sample of re-
sponses. For example, Brooks hi re, Ives, and Schulze [8] noted that very
high and low bids relative to the mean may indicate false bids. Alterna-
tively, this same phenomenon has been interpreted to imply a rejection of
the contingent market. Generally, analysts have used one or the other rea-
son to reject responses outside 10 standard deviations of the mean (see
Rowe, d'Arge, and Brookshire [27] and Brookshire et al. [6] as examples).11
However, all of these approaches implicitly describe the process gen-
erating a number of large bids. (See Randall, Hoehn, and Tolley [24].)
For example, one might assume that a diffuse distribution of bids may re-
flect that contingent valuation surveys are "imperfect estimators" of the
representative individual1 s value of an environmental service. That is, a
15
-------
bid could be viewed*as having a nonstochastic component based on a person's
socioeconomic characteristics and a random error. This error may have a
high variance, leading to a wide variation in bids. Thus, size of the bid
alone is a poor basis for judging when an individual has rejected the mar-
ket or has given a strategic response.
As noted earlier, under ideal conditions, we would specify a behavioral
model that would explain how respondents answer valuation questions (see
Hanemann [20], Smith [34], and Carson, Casterline, and Mitchell [9] for fur-
ther discussion) and use it to interpret responses. As a first step, our
approach is a sample selection rule that combines judgment with one of the
Belsley, Kuh, and Welsch (BKW) [2] regression diagnostics, which is calcu-
lated for the estimated parameters of economic variables relevant to the
option price responses. To describe it, we .first review the BKW diagnostic
index that we selected and then explain how we used it.
Regression diagnostics are procedures designed to identify influential
observations. Using these methods in the context of an economic model im-
plicitly acknowledges that some observations may be inconsistent with the
model hypothesized to explain behavior. These observations may reflect one
or more categories of inconsistent behavior noted earlier. Our selected
diagnostic judges the effects of each observation on estimates of the param-
eter for income. This is the only economic variable that can be unambigu-
ously specified a priori as important to the option price responses.12
Our approach begins with a 1 inear-in-parameters model for option price
in Equation (2):
Y = Xp ~ e , (2)
16
-------
where
Y = Txl vector of T observations for the option price
X = TxK matrix of T observations for each of K determinants of
*
the option pricO (including a column of ones for an inter-
cept)
p = Kxl parameter vector
e = Txl vector of stochastic errors.
The ordinary least-squares (OLS) estimate of p is given as b and defined in
Equation (3):
b = (X'K)"1 X! Y . (3)
The BKW diagnostic, DFBETA, is the change in each estimated coefficient as
a result of deleting a single observation. It can be calculated without
repeated regression estimates on all possible samples (size T-l) as defined
in Equation (4) for the deletion of the i observation:
(XV1 X- 8.
DFBETA = b - b(i) = T \ 1 T , (4)
(1 - x.(X X) ^ ^j)
where:
b(i) = the OLS estimate of p with the ith observation deleted
x. * the 1th row of X
e^ = the OLS residual for the ith observation (i.e.,
e = Y[I-X(XTX)_1 XTj).
DFBETA measures the influence of each observation. We normalize this
index by the estimated parameter from the full sample and use the result
(which is analogous to an elasticity) to identify outlying observations.
Subsequently, we ranked the sample by the absolute magnitude of this per-
17
-------
centage change and .examined the characteristics of the observations (i.e.,
the respondents) having the largest effect to see if they had any common
characteristics.
The specific model used in this analysis was linear with the option
price specified to be a function of the respondent's income and a variety
of other individual and survey format variables.13 We based our initial
screening of the sample on the option price combined for all levels of water
quality (i.e., column 4 in Table V [discussed later in more detail]).14 We
used a 30 percent (±) change in the estimated parameter for income as the
threshold for identifying the 32 influential observations shown in Table II.
Most of the index values for the remaining observations were much less than
the ±30 percent value used to classify a response as an outlier with only a
few responses around ±20 percent. Thus, while our selection relied on an
observed empirical threshold in our calculated index values, based on judg-
ment, we evaluated the relative importance of each of the judgments under-
lying this selection and found that they had little effect on the group of
responses considered outliers. That is, the 32 observations shown in
Table II proved to have substantially large effects on the income coeffi-
cient across individual levels of option price and alternative model specif-
ications.
Table II summarizes the characteristics of these respondents. These
results show a striking consistency in the characterization of the outl iers.
Sixty-three percent earned annual incomes of $2,500 a year or less, and 78
percent of them earned less than $7,500 a year. Thirteen of the respondents
are 60 years of age or older. Female respondents comprised 80 percent
of the outliers, while only four respondents had more than a high school
18
-------
Table II, Profile
of Outliers
*
BW 0F8FTA
elasticity
Vert ion
Option price: avoid
loss of site (D-t)
<*/yr)
Option price:
Improve water
quality to SwInaMfale
(Vyr)
' Incwse
J/yr
Age
Sex
Education
(yr)
User of
Konongahela
site
Boat
ownership
-213.1?
A
125
260
2,500
25
K
12
Ho
Ho
-155.99
A
125
200
2,500
20
F
12
Yes
Ho
-100.04
8
200
200
7,500
67
K
12
No
Ho
-79.83
A
500
500
22.500
39
K
14
Ho
. Yes
-66.19
A
125
220
7,500
43
F
10
Yes
No
-63.25
£
25
5
2,500
70
F
10
Ho
No
-62.95
0
450
200
17,500
37
F
12
Yes
• No
-56.70
€
60
85
2,500
23
F
12
Ho
Ho
-54.96
B
0
10
2,500
82
F
10
Ho
Ho
-49.68
0
50
25®
7,500
40
F
14
Yes
No
-44.62
A
155
250
12.500
57
F
12
No
No
-43,30
C
5
5
2,500
69
F
10
No
No
-4). 16
A
155
250
12,500
44
f
10
No
No
-37.34
C
S
5
2.500
62
F
10
No
No
-36.46
c
25
0
2,500
46
F
10
No
No
-36.03
c
0
0
2,500
76
F
16
No
No
-31.40
B
200
300
27,500
21
F
12
Yes
No
-30.43
A
200
285
22,500
66
F
12
Yes
No
31.24
B
5
3
7,500
34
K
12
Ho
No
33.98
A
0
0
12,500
38
F
12
No
No
35.39
A
0
0
2,500
78
F
0
No
Ko
3?.??
0
75
10
2,500
59
F
12
Yes
No
41.78
0
25
10
2,500
72
F
12
Ho
No
47 15
A
5
130
2,500
61
F
12
Yes
No
52.23
A
0
30
7. tOO
50
F
12
Yes
No
52.86
B
0
0
2,500
43
F
10
No
No
58. 18
A
0
0
2,500
79
F
10
No
No
65 70
A
o
10
2,500
66
f
12
No
No
69 15
a
10
20
2,500
33
f
12
Yes
No
79.58
s
55
0
2,500
71
f
10
No
Nn
82.52
D
0
0
2.500
51
f
12
No
No
11? 04
D
0
?'!
2.5110
26
r
1?
Yes
Yes
NQIE; A = $12S biiMifrg C = $?!> hidriinQ
B - direct q«r^5tton D - pdymenL caul.
-------
Xf- IS
education. Hw=335*A'Interesting *3»n5ift^4*- that 14 of the 32 outliers had
received the $125 starting point bidding game—twice as many as the next
most frequently occurring version (the payment card). Since these influen-
tial observations were eliminated from the subsequent analysis of the sam-
ple , the analysis of starting point bias discussed later cannot be distin-
guished from this decision.
While definitive implications from this simple characterization of the
features of the outlying responses are impossible, one reason seems to ex-
plain a 1arge fraction of the responses. Low income bidders are seen in
these groups to be from two extremes in their responses—very low bids and
fairly high bids, relative to their specified income. When we examine the
ages of these corresponding individuals, several of these low income re-
spondents (e.g., students) seem to have based their bid on anticipated
permanent income levels rather than on reported current income. Of course,
this does not explain all of the discrepancies. For example, the prevalence
of the $125 starting point may suggest an unanticipated problem with bidding
games. That is, the starting point may affect the likelihood of someone
being judged an outlying response.
4. Option Price Results
In this section, we consider the features of the distribution of option
price bids and discuss the estimated mean values for the option price re-
sponses. We evaluate the means grouped by questionnaire versions with users
distinguished from nonusers.
A. Distribution of Responses
The option price responses are available for the loss of the recreation
services of the site (avoiding a decrease from Level D to Level E on the
20
-------
WlLr
water quality^ and for improvements in water quality first from boatable to
fishable (D to C) and then from fishable to swimmable (C to B). The indi-
vidual 1s aggregated option prices also are presented for both water quality
To characterize the distribution of option price estimates we report
the kurtosis (K) and Uthoff's U-statistic.15 Smith's [33] small sample
experiments suggest that these two statistics, used together, provide a
robust approach for detecting heavy tailed distributions. Table III pre-
sents these results for the full sample and the sample ultimately selected
for analysis—that is, by eliminating protest zeros and the outlying obser-
vations.
Using the full sample, the null hypothesis of a symmetric distribution
(a close approximation of the normal) would be rejected in nearly all
cases—that is, across al 1 water quality changes and question formats.
Moreover, both test statistics support this conclusion. The iterative bid-
ding with a $25 starting point for an improvement from boatable to fishable
water quality conditions is the only exception. Eliminating the protest
and outlying observations clearly affects the responses from questionnaires
having the iterative bidding with the $125 starting point. This case is
€-
consistent with ©urinalysis of the features of the outlying observations.
Eliminat-
ing the outlying bids clearly affected judgments on the shape of the dis-
tribution. For 3 of 4 water quality changes, the results change from strong
evidence of non-normal distributions to reasonably strong support for sym-
metric distributions. Moreover, deleting only the protest zero bids did
not lead to reversals in the conclusions formed based on the full sample.
21
-------
Table III. Results for Thick-Tailed Tests3
All responses Oh3ot) Selected responses
Version/water
quality change K U K U
1. Iterative bidding - $25 starting point (Version C)
0 to E
19.05
1.58
18.35
1.80
D to C
3.54*
1.28*
2.71*
1.17*
C to B
5.07
1.70
3.96
1.45
Total (E to B)
8.41
1.40
8.86
1.45
Iterative bidding $125
starting point
(Version D)
0 to E
12.50
1.41
2.09*
1.15*
D to C
4.11
1.42
2.89*
1.35
C to B
6.81
1.96
6.15
2.09
Total (E to B)
9.70
1.44
2.57*
1.24*
Direct question (Version B)
D to E 11.62
D to C 8.91
C to B 16.42
2.17
2.05
2.62
13.36
8.72
9.38
1.86
1.85
2.19
Total (E to B) (9.44
1.97
5.05
1.61
Direct question payment card (Version A)
D to E 13.39
D to C 16.30
C to B 23.66
1.78
1.90
2.72
6.40
17.92
23.20
1.56
1.68
2.56
Total (E to B) 9.57
1.61
6.41
1.41
aFor 10 percent sigm"ficance level and sample size of 50, Smith [33]
estimated empirical critical values of K = 3.543 and U = 1.314.
*Not significantly different from normality at 10 percent level.
22
-------
Thus, these findings offer some evidence of starting point effects on both
the mean and the shape of the distribution of bids. This analysis may also
suggest one reason why there has been such divergent evidence on this issue
*
in the past 1iterature. > That is, indirect methods of detecting outlying
observations may not be able to distinguish between responses that imply
rejection or misunderstanding of the terms of the contingent market and
those influenced by the starting points in iterative bidding. As Table III
indicates, the deletion of protest and outlying responses had no effect on
the results of hypothesis tests for the remaining question formats.
Figure 1 reports the actual frequency distribution of option price re-
sponses by version for an improvement in water quality from D to C. These
results provide further support to the statistics used to detect thick-
tailed distributions. However, neither the thick-tailed tests nor the fre-
quency distribution take account of the potential role of the respondents'
characteristics and income for their option price bids. Nevertheless, they
do suggest that using test statistics for means based on normality should
be done cautiously, with special attention given to sample size for the
relevance of critical values.16
B. Mean Option Price Responses
Table IV reports the estimated means for the various water quality
changes.17 The means also are grouped by question format and by user/non-
user. Generally, the estimated means are sizable for the Monongahela River
and are of the same order of magnitude, regardless of the method used to
elicit the amount. Grouping users with nonusers, option price bids aggre-
gated for all water quality levels range from a mean of $54 per household
per year for the bidding game with a $25 starting point to $118 for the bid-
23
-------
Iterative Bidding
Framework—$25
Starting Point
Iterative Bidding
Framework—$125
Starting Point
Direct Queition
Pramawork
Direct Queition
Framework—
Payment Card
30
| 20
£ io
o
30
> 20
2
s
ST
£ 10
• o
30
> 20
|
£ 10
20
19
3
i_L
1
J ' ' ¦, » f 4 1 '
1 .1 .1
25
3 4 3
111. I
2 2
- - * i 11 " |
JU ,1 ,i„„ * ju-j i. ,, s *_J i, —I—; _J—L
20
14
1 i2
-jJj.
1
; I-
£ 20
10
29
8
m
10
6
z\
1 i
2
1,1. 1 1,.
J,—J,
0 10 20 30 40 50 60 70 80 90 100 110 120130 140 150160
$ Bids
Fig, I
24
-------
Table IV. Estimated Option Price for Changes in Water Quality:
Effects of Instrument and Type of Respondent
Change in
water quality
User
Nonuser
Combined
1. Iterative bidding--$25 starting point fVgyni-i-on ¦€)
(avoid)
D to E
D to C
C to B
D to B
All levels
27.4
18.9
11.8
32.1
59.5
16.7
16.3
14.5
27.1
38.1
19
29.7
14.5
7.2
21.7
51.4
35.7
15.2
11.6
24.0
53.1
39
29.0
15.9
8.7
25.1
54.1
30.6
15.5
12.7
25.3
48.51
2. Iterative bidding--$125 starting point (Ver&ioo-4>)
4. Direct question: payment card ¦fVuvs'ian ft-)
0 to E (avoid) 46.8 42.2
D to C 45.3 71.4
C to B 22.9 48.7
D to B 71.2 117.7
All levels 117.9 117.0
17
53.0
21.9
7.7
29.9
82.8
76.3
33.8
20.0
47.5
104.7
37
51.0
29.3
12.5
42.9
93.9
NOTE: X = sample mean.
s = sample standard deviation,
n = number of observations.
Estimates are reported in 1981 dollars, the year of the survey.
58
D to E (avoid)
94.7
66.0
38.8
51.3
57.4
62.0
D to C
58.1
51.9
26.3
45.4
36.9
49.5
C to B
33.1
48.4
16
11.6
33.1
32
18.8
39.7
48
D to B
99.7
87.9
40.5
69.0
60.2
•80.0
All levels
194.4
136.5
79.2
102.5
117.6
126.0
Direct question
D to E (avoid)
45.3
65.2
14.2
27.1
24.5
45.4
D to C
31.3
44.2
10.8
21.6
17.6
32.1
C to B
20.2
35.5
17
8.5
21.9
34
12.4
27.4
• 51
0 to B
52.9
72.5
i
20.3
41.4
31.2
55.2
All levels
98.2
103.5
34.5
66.4
55.7
85.2
67.1
49.3
32.2
78.1
108.9
54
25
-------
ding game with a $125 starting point. Means for the aggregated bids for
the payment card and direct question formats equaled $94 and $56, respec-
tively. The range of mean option price amounts is even narrower when only
the bids for improvements are considered, varying from $25 to $60 per year,
with the two bidding games again indicating the widest differences.
These results can be compared with the estimates used in EPA's regula-
tory impact analysis (RIA) for valuing water quality improvements associa-
ted with the effluent guidelines for the iron and steel firms along this
river. Since this analysis was conducted without access to the results from
this study, such a comparison provides an informal plausibility check for
our estimates. Based on existing 1 iterature from other sites, their aggre-
gate benefit estimates implied a range of $7.50 to $17.00 in annual benefits
per household for improving the water from boatable to fishable water qual-
ity level. This range was derived using al1 three benefit estimation meth-
ods reported in the RIA—participation, indirect, and survey.18 For the
improvement from D to C shown in Table IV, our survey estimates are uniform-
ly larger for users. This outcome is consistent with our a priori expecta-
tions. Option price is an ex ante benefit measure, while those in the RIA
are ex post measures. Option price wi 11 reflect both anticipated use and
the individual's desire to adjust to water quality conditions in presence
of uncertainty over the prospects for that future use.
5. Test Findings
The results of the tests for differences in means between question for-
mats using pairwise comparisons for users, nonusers, and the combined groups
are reported in Table V. Given our earlier findings, these conventional
tests should be interpreted cautiously. Nonetheless, the results suggest
-------
Table V. Student t-Test Results for Option Price
Means combined Users Nonusers Combined
Payment card v. direct question
0 to E ' — 2.806 2.353
E to B ¦ — 2.300 1.991
Payment card v. $25 iterative bidding
D to E — — 2.263
D to C ~ — 1.954
E to B 2.061 -- 2.530
•
Payment card v. $125 iterative bidding
D to E -2.499
Direct question v. $25 iterative bidding
D to E — -2.074
Direct question v. $125 iterative bidding
0 to E -2.161 -2.453 -3.020
D to C — — -2.308
D to B — -- -2.109
E to B " -2.289 -2.117 -2.8786
$25 iterative bidding v. $125 iterative bidding
D to E -4.294 — -3.072
D to C -3.119 — -3.046
D to B -3.183 — -3.159
E to B -4.131 — -3.539
Note: Only the cases where statistically significant differences in the means
were found at the 0.05 significance level are reported in this table.
27
-------
that major differences do occur between the means in the bidding games, sug-
gesting some influence from the difference in the starting points. The
means would be judged under conventional criteria to be significantly dif-
¦
ferent at least at the 5-percent level for users and for the combined
e
groups. However, the estimates do not permit the null hypothesis of equal
»
means to be rejected for nonusers. There is also some indication that the
mean option price for users of the Monongahela is significantly higher when
the bidding game with the $125 starting point is used to elicit option price
compared to the direct question technique.
Our estimated option price equations shown in Table VI, which control
for differences in individuals socioeconomic characteristics and a set of
qualitative variables to account for the interviewer, provide additional
insights into the effect of question format on option price. Based on a
dummy variable that was defined to compare the payment card with the other
three versions, option price would be judged to be significantly higher for
some water quality changes for respondents with the payment card relative
to the direct question and the $25 bidding game, finally, as noted earlier
in Section 4, the influence of the starting point cannot be separated from
the effects of omitting influential observations.
Overal 1, our results on the effects of starting points would seem to
fall between the results of Rowe, d'Arge, and Brookshi re [27], who found
evidence of a starting point bias, and Thayer [35], who found none. Some
of the differences in the results may be due to the "commodity" sold in the
hypothetical market. In Thayer's case, the respondents were al1 users who
had a very clear conception of the commodity and of the costs of substitute
recreation sites. Both our results and those of Rowe, d'Arge, and Brook-
28
-------
Tiblt vl. Rtgrtsiion Results for Option Price Estimates®
Water quality changes
Total improve'
mints only
D to E (avoid)
D to C
C to 8
Total all levels
Intercept
-34.512
-29.307
~5.430
-56.653
-22.141
(-.973)
(-1.098)
(-.257)
(-.916)
(-.517)
$«x, 1 if u1t
8.4S1
-.672
-1.657
6.484
1. 967
(-916)
(-.097)
(-302)
( 403)
(-.177)
Age
-.292
.290
-.365
(-854)
-.562
(-1.094)
(-1.440)
(1.668)
(-1834)
(1 743)
Education
5. 294
2.I01
-5.17
8.066
2 773
(2-071)
(1.508)
( 347)
(1.810)
( 699)
InCCNM
.0008
.0003
.0003
. 0012
.0006
(1.652)
(1151)
(1.260)
(1-832)
(1278)
Oirect question
-32.311
-14.372
-3.500
-50.734
-18 423
(-2.771)
(-1.638)
(.505)
(-2.495)
(-1. 309)
Iterative bidding gewe C$21)
•20.623
-12.572
-5.657
-39.566
-18.943
(-1 852)
(-1.500)
(-.854)
(-2.037)
(1.409)
Iterative bidding g««* ($115)
1.7522
6.639
.739
31.089
13.568
(1.421)
(.716)
(.101)
(1446)
(.912)
User, 1 if user
8.640
8.083
6.839
26.026
17.187
(.919)
(1.117)
(1.96)
(1 552)
(1.481)
Killing to pay cost of water
17.001
21.960
10.023
51.325
34. 326
pollution, 1 if vary Mucl>
(1.7M)
(3.068)
(1.772)
(3.095)
(2.990)
or *c*t#wt>at
26. 509
Interviewer #1
14. 211
7.090
11.334
12. 298
(.750)
(.497)
(1.006)
(.802)
( 538)
Interviewer §2
1.723
12.242
16.849
24.719
22. 996
(.099)
( 938)
(1634)
(.817)
(1.099)
Interviewer #3
-22.833
21.141
17. 578
9.292
32.125
(-1.3*4)
(1.653)
(1.740)
(314)
(1567)
Interviewer #4
-2B.125
3.050
20.605
-12.334
15. 791
(-.660)
(. 124)
(1.059)
(-216)
( 400)
Interviewer #5
6.932
4.996
2.191
11.435
4.503
(.404)
(.387)
(.215)
(-382)
(217)
Interviewer #6
47.012
95.513
66.288
198.450
151.439
(.887)
(2.394)
(2.102)
(2.146)
(2.366)
Interviewer #7
27.670
2.470
4.130
39.645
11.975
(1 425)
(.169)
(.357)
(1170)
(.511)
Interviewer #6
14. 022
29.961
19.871
58. 063
• 44.041
(.801)
(2.274)
(1.908)
(1.902)
(2.08)
Interviewer #9
17.874
39.586
-7.935
37.330
19.456
( 454)
(1.336)
(-.339)
(.54®)
( 409)
Payment card plus direct
question, 1 if eitner
*
i:
.334
.284
.166
. 36 6
.269
F
3.78
3.00
1.50
4.36
.278
Degrees of freedo®
136.0
136.0
136.0
136.0
136.0
®Tht nuaoers in ptrtnthtws below the estiaated coefficient* are t-ftitistics for the null nypotnesn of m
•iiociation.
29
-------
shire [27] used samples of households who may be more sensitive to the for-
mat used if they did not have a clear conception of the commodity.
The regression results in Table VI also provide some evidence on the
effects attributable to .differences in interviewers.19 Using dummy vari-
ables, the results indicate that the interviewer effects are limited. Only
two interviewer variables appear to have a significant effect on bids at
the 5-percent level. Even in these cases, the effects are only present for
some of the water quality changes. One of the cases involved an interviewer
who conducted only two interviews before being removed from the interviewing
team. This interviewer did not take part in the training session and also
conducted interviews only in the Latrobe area, which is a considerable dis-
tance from the Monongahela River. The second interviewer also conducted
interviews in the Latrobe area and in one area very close to the river. We
cannot unambiguously attribute these differences to the interviewers in-
volved, since the model cannot differentiate between an interviewer effect
and omitted area specific variables.
30
-------
I
6. SUMMARY AND CONCLUSIONS
Our findings indicate that contingent valuation approach can be used
to elicit individuals' valuations of changes in water quality. Our esti-
mates of option price with uncertain use were related to income as economic
theory would imply. Overal1, the prognosis from the Monongahela River case
study for the continued use of the contingent valuation approach is posi-
tive. The empirical models performed reasonably well in explaining vari-
ations in option price, with 1ittle indication that individua1 interviewers
influenced results. In addition, respondents apparently perceived enough
realism in the survey that they did not have problems with its hypothetical
nature.
Generally, the results confirm the recent state-of-the-art assessment
by Cummings, Brookshire, and Schulze [11] and with the earlier summary judg-
ments of Randal 1, Hoehn, and Tolley [24]: contingent valuation surveys seem
capable of providing order-of-magnitude estimates of the benefits realized
from enhancing one or more aspects of environmental quality. In addition,
our findings further suggest that these benefit estimates are not confined
to user-related values. Individuals can understand and incorporate values
derived from uncertain future use into their bids for environmental improve-
ment.
However, our results do suggest that the question format can be impor-
tant to the estimates of option price. The bidding game with a $125 start-
ing point and the payment card approach appear to have led to higher re-
sponses than the other two formats. There is some evidence of a starting
point bias in the bidding game, but the resul ts are not conclusive. Com-
31
-------
pared to the other.formats, the higher starting point resulted in a higher
percentage of outlying responses. However, the combined comparison of bid-
ding games with nonbidding approaches showed no significant differences.
Nonetheless, we caution against routinely using bidding games until more is
known about their effect on how a respondent processes contingent valuation
information.
Finally, we have argued that sample selection rules to identify protest
and outlying observations should be viewed in an empirical framework that
recognizes these rules as the result of each analyst's implicit model of
how individuals respond to contingent valuation surveys. Interpreted in
this way, regression diagnostics, which focus on the estimated parameters
of economically relevant variables, provide signals of response patterns
inconsistent with the sample norm for these variables. In effect, the
individual is not responding in the same way as his peers in that income
category (after using simple methods to hold other socioeconomic and ques-
tionnaire-related effects constant). Clearly, a model of the individual
decision process leading to all responses would be preferrable. Nonethe-
less, this is a step toward developing such a model.
32
-------
FOOTNOTES
Mwo anonymous reviewers contributed constructive comments on an earlier
*
draft of this paper. Matthew McGivney provided research assistance while
Hal 1 Ashmore 1 ent editorial guidance. This research was supported by the
U.S. Environmental Protection Agency under Contract No. 68-01-5838. The
views expressed in the paper are those of the authors, not their respective
institutions.
2Confusion abounds in the terms used to represent these benefits. In this
paper, use benefits are those directly 1 inked to use of a resource. Nonuse
benefits do not require use of the resource. Existence values, which are
a type of nonuse benefit, are not included in our empirical analysis.
3For a review of the empirical evidence on the relationship between nonuse
and use values for water quality changes see Fisher and Raucher [14].
4This argument has been offered by a number of analysts in recent papers.
See Bishop [3,4], Hanemann [19], and Smith [31,32] as examples.
sThe reference operating conditions defined by Cummings, Brookshire, and
Schulze [11] can be summarized as follows:
• Participants must understand and be fami1iar with the commodity to
be valued.
• Participants must have had or be allowed to obtain prior valuation
and choice experience with respect to consumption levels of the com-
modity.
• There.must be 1 ittle uncertainty.
• Wil1ingness-to-pay and not willingness-to-accept valuation measures
should be elicited.
33
-------
These conditions are in contrast to the earlier optimism of Schulze,
d'Arge, and Brookshire [30] and seem to be more in line with the view of
Rowe and Chestnut [26].
6For a detailed discussion of the theory underlying the definition of option
price in the timeless framework, see Bishop [3]Bohm [5], Graham [16],
Schmalensee [28,29], and Smith [32]. For the time-sequenced analysis,
see Arrow and Fisher [1] and Hanemann [19].
7Plummer and Hartman [23] have investigated the implications of relating
uncertainty in various ways to the features of the indirect utility func-
tion. Freeman [15] has al so used this 1ine of argument to develop a set
of empirical bounds on the size of the difference between option price and
the expected consumer surplus £££&}.
8These general comments are consistent with Freeman's [15] conclusions.
9For more details, see Chapter 3 in Desvousges, Smith, and McGivney [13].
The interviewers participated in a 2-day training session devoted to meas-
uring benefits, water pollution issues, and mock interviews using all four
question formats.
10The details of the test results and logit analysis are presented in
Desvousges, Smith, and McGivney [13].
"Randall^ Hoehn, and Tolley [24] describe two general types of procedures
for dealing with the outlying bids. The first, and most popular, uses some
threshold for either the bid or the bid as a fraction of income and elimi-
nates responses with values exceeding that threshold. The second censors
the bids by altering these large bids to correspond to a set maximum
threshold value. While these authors do not compare the approaches, there
is no basis for recommending the second approach. Indeed, if one is con-
34
-------
cerned with deleting observations, iteratively re-weighting observations
*
with robust regression techniques would seem to provide a better alterna-
tive than a procedure that deliberately introduces a censoring problem in
the sample of bids.
12A variety of other techniques could be used to detect influential observa-
tions, including using the fitted values for the option price based on the
estimated models. See Belsley, Kuh, and Welsch [2] and Cook and Weisberg
[103 for further discussion.
13These s-ets—of variables refer to the measures used to describe the charac-
teristics of each -jwdWidoa) respondent, such as age, sex, income, and ed-
ucation, and the measures used to •teke account,^* the specific survey
questionnaire received by the individual, such as payment card, direct
arc
question, type of bidding format.
14The estimates reported in this table are based on the sample with the
influential observations deleted. These equations describe the functional
Move. Is
yyec-ificatrrons used.— The spciific~y*^4matasreported in Desvousges,
A
Smith, and McGivney [13], Appendix C.
lsThe kurtosis statistic is defined as
K * (Z(X. - X)4/n)/(Z(X. - X)2/n)2 .
i 1 i 1
The Uthoff U is the ratio of the standard deviation to the mean deviation
from the sample median.
U = (Z(X. - X)2/n)V(Z X. - Xm /n)
-------
where
*
n - sample size
X. =* value for ith observation
• X = sample mean
*
= sample median.
For more details on the performance of these and other tests for thick-
tailed distributions see Smith [33].
l6The central 1imit. theorem assures that as long as response can be
assumed to arise from distributions with finite first and second
moments, the mean will have a normal distribution. What is relevant
for practical purposes is the sample size at which the smal1 sample
distribution for test statistics approaches the hypothesized form used
in defining the critical values for hypothesis tests. This is the
reason for caution in interpreting test results using conventional cri-
teria.
17Appendix C in Desvousges, Smi th, and McGivney [13] presents the esti-
mated means for both the ful 1 sample and the sample with only the pro-
test bids excluded. Calculated t-statisties revealed no statistically
significant differences between the means estimated from the ful1 sam-
ple and those estimated with the protest bids excluded.
18These estimates are derived from Raucher and Fisher [25] by dividing
the aggregate estimates reported in their Table 1 by the number of
households in the region to derive an implied "average" willingness to
pay. Therefore, they do not take account of the geographic dispersion
of the households in the region.
36
-------
19This test is limited because interviewers were assigned geographic seg-
ments, designed to minimize travel costs. The high cost of randomly
assigning interviewers makes a complete test impractical for a house-
hold' survey.
37
-------
REFERENCES
1. K. J. Arrow and A. C. fisher, "Environmental Preservation, Uncertainty
and Irreversibility," Quarterly Journal of Economics 88, 313-319 (May
1974). —* — —
2. David A. Belsley, Edwin Kuh, and Roy E. Welsch, Regression Diagnostics,
New York: John Wiley and Sons (1980).
3. Richard C. Bishop, "Option Value: An Exposition and Extension," Land
Economics 58, 1-15 (February 1982).
4. Richard C. Bishop, "Option Value or Option Price: Principles for
Empirical Resource Valuation Under Uncertainty." Department of Agri-
cultural Economics, University of Wisconsin, Madison (April 12, 1984).
5. Peter Bohm, "Option Demand and Consumer Surplus: Comment," American
Economic Review 65. 733-736 (September 1975).
6. David S. Brookshire, Ronald G. Cummings, Morteza Rahmatian, Willi am D.
Schulze, and Mark A. Thayer, Experimental Approaches for Valuiog Envi-
ronmental Commodities, draft report prepared for U.S. Envifonmental
Protection Agency, University of Wyoming, Laramie, Wyoming (April
1982).
7. David S. Brookshire, Larry S. Eubanks, and Alan Randall, "Estimating
Option Prices and Existence Values for Wi1dlife Resources," Land
Economics 59, 1-15 (February 1983).
8. David S. Brookshire, B. Ives, and Wi 11 iam D. Schulze, "The Valuation
of Aesthetic Preferences," Journal of Environmental Economics and Man-
agement 3, 325-346 (December 1976).
9. Richard T. Carson, Gary I. Casterline, and Robert Cameron Mitchell, "A
Note on Testing and Correcting for Starting Point Bias in Contingent
Valuation Surveys," Discussion Paper D-116, Quality of the Environment
Division Resources for the Future (January 1984).
10. R. Dennis Cook and Sanford Weisberg, Residuals and Influence in Regres-
sion, New York: Chapman and Hal 1 (1982).
11. Ronald G. Cummings, David S. Brookshire, and William D. Schulze, Valu-
ing Public Goods: The Contingent Valuation Method, Totowa, New Jersey:
Rodman and Alanheld (forthcoming).
12. John T. Daubert, and Robert A. Young, "Recreational Demands for Main-
taining Instream Flows: A Contingent Valuation Approach," American
Journal of Agricultural Economics 63, 666-676 (November 1981).
38
-------
13. William H. Oesvousges, V. Kerry Smith, and Matthew P. McGivney, A Corn-
par i son of Alternative Approaches for Estimating Recreation and Rela-
ted Benefits of Water Quality Improvements, report prepared for U.S.
Environmental Protection Agency, Research Triangle Institute, Research
Triangle Park, North Carolina (March 1983).
*
14. Ann Fisher and Robert Raucher, "Intrinsic Benefits of Improved Water
Quality: Conceptual and Empirical Perspectives," Advances in Applled
Microeconomics 4, V. K. Smith and A. D. Witte, eds, Greenwich, Con-
necticut: JAI Press (1984).
15. A. Myrick Freeman III, "The Size and Sign of Option Value," Land Eco-
nomics 60, 1*13 (February 1984).
16. Daniel A. Graham, "Cost-Benefit Analysis Under Uncertainty," American
Economic Review 71, 715-725 (September 1981).
17. D. A. Greenley, Richard G. Walsh, and Robert A. Young, "Option Value:
Empirical Evidence from a Case Study of Recreation and Water Quality,"
Quarterly Journal of Economics 96, 657-674 (November 1981).
18. Douglas A. Greenley, Richard G. Walsh, and Robert A. Young, "Option
Value: Empirical Evidence from a Case Study of Recreation and Water
Quality: Reply," Quarterly Journal of Economics, 100, 295-299 (Febru-
ary 1985).
19. W. Michael Hanemann, "Information and the Concept of Option Value,"
Quarterly Journal of Economics (forthcomi ng).
20. W. Michael Hanemann, "Statistical Issues in Discrete Response Contin-
gent Valuation Studies," Northeastern Journal of Agricultural and
Resource Economics (April 1985).
21. Robert Cameron Mitchell and Richard T. Carson, An Experiment in Deter-
mining Willingness to Pay for National Water Qua)ity Improvements,
draft report prepared for U.S. Invfronmental Protection Agency, Re-
sources for the Future, Inc. , Washington, D.C. (June 1981).
22. Robert Cameron Mitchel1 and Richard T. Carson, "Option Value: Empir-
ical Evidence From a Case Study of Recreation and Water Quality: Com-
ment," Quarterly Journal of Economics, 100, 291-294 (February 1985).
23. Mark L. Plummer and Richard C. Hartman, "Option Value: A New Ap-
proach," Economic Inquiry (forthcoming).
24. Alan Randall, John P. Hoehn, and George S. Tolley, "The Structure of
Contingent Markets: Some Results of a Recent Experiment," paper pre-
sented at the American Economic Association Annual Meeting, Washing-
ton, D.C. (1981).
39
-------
25. Robert L. Raucher and Ann Fisher, "Measuring the Recreation Benefits
of Improved Water Qual ity: A Comparison of Alternative Methodspaper
presented at the Southern Economic Association meetings, Atlanta,
Georgia (November 10-12, 1982).
26. .Robert D. Rowe and Lauraine G. Chestnut, "Valuing Environmental Com-
modities: Revisited," Land Economics 59(4), 404-410 (November 1983).
27. Robert D. Rowe, Ralph C. d'Arge, and David S. Brookshire, "An Experi-
ment on the Economic Value of Visibility," Journal of Environmental
Economics and Management 7, 1-19 (March 1980).
28. Richard Schmalensee, "Option Demand and Consumer Surplus: Reply,"
American Economic Review 65, 737-739 (September 1975).
29. Richard Schmalensee, "Option Demand and Consumer Surplus: Valuing
Price Changes Under Uncertainty," American Economic Review 62, 813-824
(December 1972).
30. W. D. Schulze, R. C. d'Arge, and D. S. Brookshire, "Valuing Envi ron-
mental Commodities: Some Recent ExperimentsLand Economics 57(2),
151-173 (May 1981).
31. V. Kerry Smith, "Intrinsic Values in Benefit Cost Analysis." Unpub-
lished paper, Vanderbilt University (1984).
32. V. Kerry Smith, "Option Value: A Conceptual Overview," Southern Eco-
nomic Journal 49, 654-668 (January 1983).
33. V. Kerry Smith, "A Simulation Analysis of the Power of Several Tests
for Detecting Heavy-Tailed Distributions." Journal of the American
Statistical Association, 70, 662-665 (September 1975).
34. V. Kerry Smith, "Some Issues in Discrete Response Contingent Valuation
Studies." Northeastern Journal of Agricultural and Resource Economics
(April 1985 ^ _
35. Mark A. Thayer, "Contingent Valuation Techniques for Assessing Environ-
mental Impacts: Further Evidence," Journal of Environmental Economics
and Management 8, 27-44 (1981).
36. U.S. Bureau of the Census, 1980 Census of the Population and Housing,
Washington, D.C. (1982).
I
40
-------
Figure 1. Effects of instrument—distribution of option price for a change
in water quality from beatable to fIshable, protest bids
excluded.
------- |