A meta-analysis of wetland contingent valuation studies
Original article
A meta-analysis of wetland
contingent valuation studies
R. Brouwer á I.H. Langford á I.J. Bateman á R.K. Turner
asked for either their willingness to pay (WTP) or their
willingness to accept (WTA) compensation for the gains or
Abstract There is growing interest in the potential
losses involved (Mitchell and Carson 1989). Of these op-
for producing generally applicable models for
tions the WTP approach has become the most frequently
valuing non-market environmental services which
applied and has been given peer review endorsement
do not rely upon expensive and time-consuming
through a variety of studies (see, for example, Arrow et al.
survey work, but rather extrapolate results from
1993). When aggregated across those who will be affected
previous studies. This paper presents a meta-
by the suggested environmental changes, this stated WTP
analysis for the use and non-use values generated by
amount is used as a socio-economic indicator of the
wetlands across North America and Europe. The
environmental values involved.
study assesses the socio-economic values
Given the substantial indirect, often off-site, use and non-
attributable to the hydrological, biogeochemical and
use values involved, wetlands have been the focus of at-
ecological functions provided by such complex
tention in several CV studies (Crowards and Turner 1996).
environmental assets. The clustering of multiple
Many of these studies try to estimate the total economic
values derived from single studies is examined
value of wetlands. Total economic value, not to be
through the application of multilevel modelling
confused with total ecosystem value, consists of use and
methods allowing for the hierarchical structure of
non-use values (Pearce and Turner 1990). CV is the only
such data.
economic method to date that is able, in principle, to
account for possible non-use motivations underlying
Key words Meta-analysis á Contingent valuation á
people's value statements. Whereas use values refer to the
Wetlands á Ecosystem functions
values associated with the actual use of the various goods
and services wetlands provide, non-use values are unre-
lated to any actual or potential use of these goods and
services.
Introduction Wetlands are complex hydro-ecological systems, whose
structure provides us with goods or products involving
some direct utilisation of one or more wetland character-
This paper addresses the socio-economic values of the
istics, while wetland ecosystem processes provide us with
various functions performed by wetland ecosystems.
hydrological and ecological services, supporting or pro-
Environmental economists have developed a variety of
techniques for measuring such values, of which the con- tecting human activities or human properties without
tingent valuation (CV) method is probably the most widely being used directly. The stock of wetlands is a multi-
applied in contemporary research. CV is a survey method functional resource with signi®cant economic value, as
also has been suggested by Costanza et al. (1997). Fifteen
where individuals are presented with information about
percent of the value of the world's ecosystem services and
speci®c environmental changes, and their perception,
natural capital is generated by wetlands (Costanza et al.
attitudes and preferences regarding these changes are
1997). However, all over the world countries have experi-
elicited. In order to measure the effect of the suggested
enced severe wetland losses (Tolba and El-Kholy 1992;
changes on people's welfare, respondents are typically
Turner 1992). Sustainable management of these assets is
highly relevant. Since this management process is not
costless, they require accurate and meaningful valuation in
order to be able to weigh the costs and bene®ts of their
Received: 24 February 1999 á Accepted: 6 June 1999
conservation.
R. Brouwer (8) á I.H. Langford á I.J. Bateman á R.K. Turner In this paper, the main ®ndings of CV studies of
Centre for Social and Economic Research on the Global
wetlands in temperate climate zones in developed
Environment (CSERGE), School of Environmental Sciences,
economies will be investigated. The main objective is
University of East Anglia, Norwich, UK
to quantify the socio-economic values associated with
Tel.: 01603 593741; fax: 01603 593739,
wetland ecosystem functioning in a meta-analysis of
e-mail: r.brouwer@uea.ac.uk
47
Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
to increase policymaker demand for transferable valuation
wetland CV studies, supplementing qualitative analyses
È results.
provided, for example, by Gren and Soderqvist (1994)
Meta-analysis enables researchers to identify criteria for
or Crowards and Turner (1996). Natural and social
valid environmental value transfer1 or to test the conver-
science are brought together by relating the various
hydrological, biogeochemical and ecological wetland gent validity of value estimates. In the ®rst case the data-
functions to the societal bene®ts derived from these set is entirely used to determine the factors that help to
functions and the socio-economic values attached signi®cantly explain variances in valuation outcomes. In
to these bene®ts. the second case the data-set can be split, for example into
two parts, one of which is used for the ®rst purpose and
the other to test whether the value estimates based on the
signi®cant factors fall within the con®dence interval of the
Approach other half's estimates.
Environmental value transfer is commonly de®ned as the
transposition of monetary environmental values estimated
The results from 30 different CV studies of wetlands in
at one site (study site) through market-based or non-
temperate climate zones in developed economies were
market-based economic valuation techniques to another
compared and synthesised in a meta-analysis. Only very
site (policy site). The most important reason for using
few tropical wetland valuation studies exist (Barbier 1993).
previous research results in new policy contexts is cost
Tropical wetland studies are excluded from the
effectiveness. Applying previous research ®ndings to
meta-analysis presented here because of the enormous
similar decision situations is a very attractive alternative
differences between population samples in developed and
to expensive and time-consuming original research to
developing countries regarding socio-cultural and demo-
quickly inform decision making.
graphic-economic characteristics.
The criteria for selecting studies for environmental value
Since the beginning of the 1990s, meta-analysis has been
transfer suggested in the literature focus on the environ-
playing an increasingly important role in environmental
mental goods involved, the sites in which the goods are
economics research (van den Bergh et al. 1997). Originally
found, the stakeholders and the study quality (Desvousges
a technique used in experimental medical treatment and
et al. 1992). However, very little published evidence exists
psychotherapy, meta-analysis is the statistical evaluation
of studies that test the validity of environmental value
of the summary ®ndings of empirical studies, helping to
transfer. Moreover, in the few studies that have been
extract information from large masses of data in order to
carried out, the transfer errors are substantial (Brouwer
quantify a more comprehensive assessment. It enables
1998).
researchers to explain differences in outcomes found in
As more information about factors in¯uencing environ-
single studies on the basis of differences in underlying
mental valuation outcomes becomes available, for in-
assumptions, standards of design and/or measurement. As
stance through the meta-analysis presented here,
such, meta-analysis is an important extension of quanti-
transfers across populations and sites become more
tative analyses and can be seen as a supplement to quali-
practicable.
tative analysis.
Compared to qualitative analysis, important advantages
of meta-analysis are that on the `input' side it does not
prejudge research ®ndings on the basis of the original
Dataset and study characteristics
study's quality, while it avoids a differential subjective
weighting of studies in the interpretation of a set of
®ndings on the `output' side (Glass et al. 1981). How- The list of wetland CV studies included in the meta-
ever, one drawback is that it may be biased towards analysis is presented in Table 1. Most studies have been
including signi®cant study results only, since it may well published in journals. Half of all studies were carried
be that insigni®cant study results will not be published. out between 1985 and 1989, with most being published in
Furthermore, multiple results from the same study are the ®rst 3 years of the 1990s. One study was carried
often treated as individual, independent observations out in the 1970s, 19 in the 1980s and 10 in the 1990s.
without explicit testing for intra-study correlation (Wolf Besides the inclusion of published signi®cant results,
1986). Table 1 illustrates two other problems in this
In the ®eld of environmental valuation, meta-analyses
have focused on a range of environmental issues from
outdoor recreation to urban air pollution, based on single
1
The term `environmental value transfer' is used here instead of
or multiple valuation techniques. The increase in meta-
the popular term `bene®ts transfer', because CV can also measure
analytical research seems to be triggered principally by (1)
the bene®ts foregone, which makes the estimated values costs
increases in the available number of environmental valu-
instead of bene®ts. WTP is the conventional economic approach
ation studies, (2) the seemingly large differences in valu- to measure environmental values in money and hence make them
ation outcomes as a result of the use of different research commensurable with other market values ± costs and bene®ts ±
designs (Carson et al. 1996), and (3) the high costs of associated with decisions that have been made, are made or have
carrying out environmental valuation studies which tend to be made in the face of limited human and natural resources
48 Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
Table 1
Studies (with year of publication in parentheses) included in the meta-analysis
Type of publicationa nb
Authors Study year
1 Bateman et al. (1995) Journal article (EE) 1991 3
2 Bergstrom et al. (1990) Journal article (EE) 1986 1
3 Bishop and Boyle (1985) Consultancy report 1985 2
4 Bishop et al. (1987) Journal article (TAFS) 1985 7
5 Brouwer and Slangen (1998) Journal article (ERAE) 1994 3
6 Carson and Mitchell (1993) Journal article (WRR) 1983 3
7 Cooper and Loomis (1991) Book chapter 1987 3
8 Cummings et al. (1994) Journal article (AJAE) 1992 2
9 Desvousges et al. (1987) Journal article (JEEM) 1981 21
10 Farber (1988) Journal article (JEM) 1984 1
11 Garrod and Willis (1996) Journal article (JEPM) 1993 4
12 Green and Tunstall (1991) Journal article (AE) 1986 1
13 Greenley et al. (1981) Journal article (QJE) 1976 4
14 Silvander (1991) Dissertation 1989 2
15 Jordan and Elnagheeb (1993) Journal article (WRR) 1991 2
16 Kaoru (1993) Journal article (ERE) 1989 1
17 Kosz (1996) Journal article (EE) 1993 1
18 Lant and Roberts (1990) Journal article (EPA) 1987 6
19 Loomis et al. (1991) Book chapter 1989 10
20 Loomis (1987) Journal article (WRR) 1985 1
21 Olsen et al. (1991) Journal article (Rivers) 1989 3
22 Phillips et al. (1993) Journal article (CJAE) 1991 2
23 Sanders et al. (1990) Journal article (WRR) 1983 2
24 Schultz and Lindsay (1990) Journal article (WRR) 1988 1
25 Spaninks (1993) MSc thesis 1993 3
26 Spaninks et al. (1996) Scienti®c report 1995 2
27 Sutherland and Walsh (1985) Journal article (LE) 1981 2
28 Whitehead and Blomquist (1991) Journal article (WRR) 1989 6
29 Willis (1990) Journal article (AE) 1986 2
30 Willis et al. (1995) Journal article (JEM) 1992 2
a
Abbreviations: AE Applied Economics; AJAE American Journal Journal of Environmental Management; JEPM Journal of
of Agricultural Economics; CJAE Canadian Journal of Agri- Environmental Planning and Management; LE Land Economics;
cultural Economics; EE Ecological Economics; EPA Environment QJE Quarterly Journal of Economics; TAFS Transactions of the
and Planning A, ERAE European Review of Agricultural Eco- American Fisheries Society; WRR Water Resources Research
b
Number of observations taken from each study
nomics; ERE Environmental and Resource Economics; JEEM
Journal of Environmental Economics and Management; JEM
The studies included in the analysis focus primarily on
meta-analysis. First, a number of people have been
wetlands or wetland-type areas.2 The speci®c WTP
involved in several studies and related publications. This
may result in an `authorship' effect. Learning from pre- questions addressed in each study cover a large con-
vious studies, authors may use similar, perhaps slightly tinuum of activities, actions or projects related to wet-
adapted survey designs in subsequent studies. Secondly, lands, but in some cases (approximately a third of all
103 data points (observations) were extracted from 30 studies) also to water resources in general. These values
studies. This corresponds, on average, to three or four were kept in the analysis because they referred directly
observations per study. More than half of all studies to the hydrological wetland functions distinguished in
provided one or two observations. Outliers are the the analysis and were considered reliable estimates for
studies by Loomis et al. (1991) and Desvousges et al. these functions. The WTP questions range from outdoor
(1987), providing 10 and 21 observations respectively. recreational activities like birdwatching or ®shing, to
Studies provided more than one observation mainly
because of the use of split survey samples targeting
different wetland user and non-user groups and testing 2
Although there is little agreement among scientists on what
different survey designs. The possibility that results from
constitutes a wetland, a workable de®nition is given by the so-
the same study cluster together, for example as a result of
called Ramsar Convention (1975, article 1): `areas of marsh fen,
identical survey design or sample population, and that peatland or water, whether natural or arti®cial, permanent or
results from some studies may be more variable than temporary, with water that is static or ¯owing, fresh, brackish or
others was tested and accounted for in the meta-analysis salt including areas of marine water, the depth of which at low
(see `The model'). tide does not exceed 6 m'
49
Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
in all other cases wetland functions provided multiple
ground-water protection and complete wildlife habitat
bene®ts.
preservation. Two-thirds of the studies are carried out in
Most studies (70%) asked respondents for the use and
the USA, the rest in Europe. Half of the European
non-use values attached to the bene®ts derived from
studies were carried out in the UK.
wetland functions. One study (Schultz and Lindsay
Separating the heterogeneous complex of hydrological,
1990), elicited future use value only (option value),
biogeochemical and ecological functions performed by the
while another (Greenley et al. 1981) focused, among
wetlands considered in each study in the meta-analysis
others, on the ex ante `option price' of being able to
was very dif®cult. Based on the various functions ad-
make a better informed judgement in the future based
dressed in the reviewed studies, a simple distinction was
on more information becoming available regarding wa-
made therefore ®rst of all between four main wetland
ter quality.
ecosystem functions: ¯ood control, water generation, wa-
In eight studies an attempt was made to break down the
ter quality support and wildlife habitat provision (Fig. 1).
stated total economic value ex post in the questionnaire
Secondly, the main function valued in each study was as-
into the various components distinguished in the litera-
signed to one of each of these four groups. Hence, each
ture, e.g. use, option, philantrophic, bequest, stewardship
study was categorised as addressing one of these four main
and existence value. In two water quality studies (De-
wetland functions, unless a study explicitly generated
svousges et al. 1987; Carson and Mitchell 1993), respon-
distinct values for different wetland functions (for more
dents were presented ex ante with a `value card' which
details, see Brouwer et al. 1997).
described the main reasons why water quality might be
Obviously, wetland ecosystem structures and processes
valued. In another two studies use and non-use values
and the functions they provide are highly interrelated,
were elicited separately, by the use of either different
making it very hard, and in some cases impossible, to
questionnaires (Bishop et al. 1987) or separate questions
distinguish between individual functions. They often go
for use and non-use values in the same questionnaire
hand in hand and attempts to separate them, for example
(Greenley et al. 1981).
for economic valuation purposes in order to avoid double
Finally, two study quality indicators were included in
counting, are liable to be arbitrary. This implies that
the analysis: one for the quality of the studies included in
double counting is a real problem and attempts to aggre-
the meta-analysis and one for the quality of the meta-
gate up to system-level values are fraught with dif®culties.
analysis itself. The quality of individual studies is indicated
Also, in the case of the human bene®ts derived from the
by the study response rate and the quality of the meta-
wetland functions involved, complete separation of direct
analysis by the so-called scope test. Both indicators are
and indirect use and non-use bene®ts is dif®cult. Only in a
found back in the National Oceanic and Atmospheric
third of all studies could a single bene®t ¯ow be identi®ed;
Wetland ecosystem
structure and processes
Functions
Hydrological Biogeochemical Ecological
1) Flood water retention 3) Nutrient retention and 4) Nursery and habitat for plants,
2) Surface and groundwater export animals and micro-organisms
recharge and landscape structural
Fig. 1
diversity
Main wetland ecosystem functions
identi®ed in the meta-analysis and their
derived socio-economic bene®ts. The
analysis is based on stated WTP (will-
Socio-economic benefits
ingness to pay) for goods and services for
- Natural flood protection - Improved water quality - Fishing
which no market exists. The value of
alternative - Waste disposal - Wildfowl hunting
marketed products such as reed or ®sh is
- Reduced damage to infra- - Other recreational amenities
excluded from the analysis to avoid
structure, property and crops
double counting with the stated use and
- Water supply
non-use values
- Habitat maintenance
50 Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
gression technique called multilevel modelling was used
Administration's (NOAA's) `burden of proof' requirements
(Arrow et al. 1993).3 (Langford 1994; Langford et al. 1998). [The GLS regression
was carried out using the package MLN (Rasbash and
A high non-response, either to the entire survey instru-
Woodhouse 1995).]
ment or to the valuation question, raises concern regard-
Given the shape of the distribution of the WTP amounts,
ing the study's representativeness, and questions the
a logarithmic transformation was used. Hence, for the
validity of the survey design employed and the extent to
log(WTP) amounts of the studies Y, the GLS model is:
which the valuation scenario in the questionnaire was
comprehensible and credible. The scope test refers to the
Y Xb ZH 1
sensitivity of WTP measures to changes in the provision
level of the goods and services being valued, i.e. the where Xb consists of the design matrix X and associated
difference between reference and target provision levels. parameters b represent the mean or ®xed effects of the
Conforming to the strong monotonicity assumption in explanatory variables on the dependent variable Y. How-
neo-classical consumer theory, responses should reveal a ever, whereas in OLS regression there is a single vector of
smaller WTP for smaller amounts of an environmental error terms or residuals, here a more complex variance
commodity provided by an environmental programme. structure may be modelled where the values of residuals
Very few studies reported the extent of protest bids and are dependent on explanatory variables included in the
other questionable responses in the survey. Although most design matrix Z for the random part of the model. For
studies mention the survey response rates, it is in many example, using one explanatory variable for simplicity,
cases not clear what these response rates actually represent Eq. (1) can be written as:
or which criteria have been used to exclude responses
from further analysis. Where such information was yi b0 b1 x1i ui
!
! !
available, protest bids and questionable responses were
r2
ui 0 rum
excluded from the response rates. u
Y
m1i Y $N 2
r2
mi 0 rum
In order to carry out a scope test, the size of the affected m
study site and the difference between the reference and the
where ui is the residual associated with the intercept b0,
target levels of environmental service provision in the CV
and vi is the residual associated with the slope parameter
scenarios should ideally be considered. However, in two-
b1 of x1. While the variance of the responses in OLS is
thirds of all studies no information is provided about the
determined by a single residual term, in the basic GLS
size of the area involved. In about one-third of all studies,
model the variance is dependent on the explanatory
the study site size was estimated using geographical maps.
variable:
Problems accumulate when aiming to also include the
difference between the reference and target levels of the var yi r2 2rum x1 r2 x2 3
u m1
various wetland functions distinguished in the environ-
mental scenarios in each study. The multi-dimensional This can be done for any number of variables, hence
nature of these functions makes a comparison between making the variance of the responses a complex function
studies impossible. Hence, instead a `relative size' variable of the explanatory variables, accounting for he-
was compiled, referring to the share of each study site in teroscedasticity. This turns out to be highly relevant, as
the country's total stock of wetlands. there are signi®cant differences, for example, in the vari-
ance of responses within different studies (intra-study
effects). Using the subscript j to label different studies,
The model the basic previous GLS model can be rewritten as:
yij b0 b1 x1ij uij m1ij x1ij
The structure of the data used in the meta-analysis is ! ! !
r2 rum
complex. WTP values are generated by different studies, ui 0 u
sj Y $N Y Y
carried out in different geographical locations using dif- rum r2
mi 0 m
ferent valuation formats. Using the summary statistics of
sj N 0Y r2 4
these different studies in a pooled sample, the usual con- s
ditions required for ordinary least squares (OLS) regres- where r2 is the variance parameter that describes the
s
sion are likely to be violated. In order to account for differing variability of estimates within different studies.
heteroscedasticity, a generalised least squares (GLS) re- This latter model will be referred to as the `extended'
model in the results section.
3
In 1992, the NOAA commissioned a prestigious `Blue-Ribbon
Panel' of economists and survey specialists, co-chaired by Nobel
Results
laureates Kenneth Arrow and Robert Solow, to investigate the CV
method. After carefully considering a wide range of issues, the
panel's report gave the method a quali®ed bill of health, but only
Summary statistics
if studies were conducted to a rigorous set of guidelines. The
A ®rst step in the meta-analysis was to make stated aver-
panel identi®ed a subset of issues which it called `burden of
age WTP amounts in each study comparable. The response
proof ' requirements
51
Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
surveys were carried out is used as an indicator for income
variable is average WTP per household per year for the
differences to account for people's capacity to pay. Income
preservation of speci®c wetland aspects. After expressing
taxation as a payment vehicle generates the highest aver-
WTP in national currencies in terms of their 1990 pur-
age WTP value, followed by stated WTP over and above
chasing power, these national currencies were converted in
actual trip expenditures to visit a wetland site (use value).
the International Monetary Fund's (IMF's) Special Draw-
The latter is part of a set of payment mechanisms that
ing Rights (SDRs), which is the Fund's of®cial monetary
present wetland functions to respondents as a private
unit of account (IMF 1996). Average WTP for wetland
good, i.e. to be `consumed' by the individual who is being
function preservation found in all studies taken together is
interviewed by asking him or her to pay, for example, an
62 SDRs (by the end of 1995, 1 SDR approximately
entrance fee. On the other hand, general income taxation is
equalled 1.5 US$). The median is considerably lower,
expected to prompt responses that consider the implica-
namely 34 SDRs.
tions of wetland preservation for society at large, not just
The breakdown of WTP values according to a number of
for the individual (Mitchell and Carson 1989; Blamey
possible explanatory factors is presented in Table 2. Mean
1995). Hence, the higher WTP value elicited through this
WTP values have been calculated for (1) wetland types as
payment mechanism is expected to re¯ect more than
identi®ed by Dugan (1990); (2) main wetland functions;
private use values only. (Obviously, `users' of speci®c
(3) relative wetland size; (4) the different value types
wetland functions, for instance people who visit a wetland
elicited in the studies (use and/or non-use values); (5) the
site for recreational ®shing or boating, may hold values
continent where the wetland sites are found; (6) the way
related to their non-use, e.g. preservation for future gen-
people were asked to pay for wetland function preserva-
erations, as well.) The high value for non-speci®ed pay-
tion in the CV survey as part of the institutional setting of
ment modes is due to outliers and the very low number of
the wetland conservation programmes (e.g. through
observations. Calculating through the value of wetland
general income taxation or otherwise); and (7) the way
function preservation in existing product prices yields a
the WTP question was elicited in the CV survey (e.g. in
signi®cantly higher mean WTP than the establishment of a
an open-ended question or otherwise).
private fund or raising entrance fees.
The calculated differences in mean WTP for each of these
Finally, corresponding to previous research results (e.g.
categories are statistically signi®cant at the 5% signi®-
McFadden 1994; Bateman et al. 1995; Willis et al. 1995),
cance level or stronger (see the outcomes of the Chi-
the open-ended (OE) elicitation format yields a signi®-
square test statistic in the last column of Table 2). The
cantly lower WTP than other formats. The dichotomous
range of values (minimum and maximum) found for the
choice format (yes or no to a given bid amount) yields the
factor levels across studies is considerable. Mean WTP
highest average WTP, followed by the iterative bidding
per household is more or less the same for salt- and
procedure (yes or no to a sequence of bid amounts).
fresh-water wetlands. However, the number of observa-
Possible explanations are the larger numbers of non-
tions for salt-water wetlands is very low. Almost all ob-
responses or protest responses OE elicitation tends to
servations refer to fresh-water wetlands. Within fresh-
produce (Desvousges et al. 1983) or the uncertainty
water wetlands, the value of wetlands fed by rivers (riv-
experienced in answering the unfamiliar WTP question for
erine) is twice as high as the value of lakes and ponds
non-market goods and services in an OE format (Bateman
(lacustrine) or marshes and swamps (palustrine).
et al. 1995).
Ground-water is valued highest, although the number of
observations is again low.
Regression results
The wetland function ¯ood control generates the highest
The ®ndings for the basic and extended GLS model in
mean WTP, followed by wildlife habitat provision and
which we account for study level effects are presented in
landscape structural diversity (labelled biodiversity in
Table 3. Only those variables are included that are statis-
Table 2). Surface and groundwater recharge (labelled
tically signi®cant at the 0.1 level. The ®xed part of the
water generation in Table 2) has the lowest value. As
model represents the ®xed or mean effects of each variable,
expected, larger sites result in higher WTP. An incon-
as for an OLS regression model, while the random part
sistency is found between the categories `small' and `very
displays the variance and covariance parameters that
small'.
model heteroscedasticity.
Use values associated with wetland functions are almost
The estimates for the regression results are obtained
twice as high as non-use values. However, a combination
through Maximum Likelihood techniques (e.g. Maddala
of the two is not equal to their sum, suggesting some non-
1983). The outcome of the likelihood ratio test
linear relationship between the two. Socio-psychological
(v2 96X51; P < 0.01) rejects the null hypothesis of zero
and related factors underpinning so-called embedding ef- 14
fects, where the sum of the valuations placed on the parts effects for all explanatory variables. A pseudo R-squared
of a commodity exceeds that for the whole (Bateman et al. was calculated from the log likelihood (LL) function. The
1997), may be one important reason. outcome corresponds with the goodness of ®t measures
North Americans are willing to pay, on average, more than usually found in CV studies. Since the pseudo R-squared
Europeans. Since average income data for the survey lacks the straightforward explained variance interpretation
samples are missing in most of the studies reviewed, the of R-squared in OLS regression (Hamilton 1993), it is used
country in which the wetland sites are found and the CV here as a rough indicator for the model's goodness-of-®t.
52 Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
Table 2
Summary statistics (WTP willingness to pay; SDR Special Drawing Rights)
na v2 (p<)b
Mean WTP (SDRs) Standard error Min (SDRs) Max (SDRs)
Wetland type 15.2 (0.05)
Saltwater 56.2 27.2 19 137 4
Marine 22.7 3.7 19 26 2
Lagoonal 136.6 ± ± ± 1
Lake 42.8 ± ± ± 1
Freshwater 58.9 6.1 1 267 97
Riverine 71.7 13.7 1 267 38
Lacustrine 36.8 9.4 12 88 9
Palustrine 36.9 4.3 9 117 31
Groundwater 125.7 24.3 99 174 3
Fresh- and saltwater 237.5 106.2 131 344 2
Wetland function 7.8 (0.05)
Flood control 92.6 24.4 24 177 5
Water generation 21.5 6.8 3 59 9
Water quality 52.5 5.9 9 174 43
Biodiversity 76.1 12.8 1 344 46
Relative wetland size 13.1 (0.01)
Very large 86.9 17.6 19 177 8
Large 70.3 21.6 12 344 16
Medium 67.0 8.9 3 267 58
Small 29.5 13.2 1 137 13
Very small 53.4 13.8 24 105 6
Value type 6.1 (0.05)
Use value 68.1 8.4 9 344 50
Non-use value 35.5 4.8 12 78 13
Use and non-use values 63.8 12.9 1 267 40
A3.0 (0.003)c
Country
USA and Canada 70.8 7.8 3 344 80
Europe 32.8 8.4 1 177 23
Payment mode 27.4 (0.001)
Income tax (1) 121.3 18.1 2 267 22
Entrance fee/private fund (2) 28.6 5.7 1 137 28
Product prices (3) 47.8 8.9 3 174 22
Combination of (1) and (3) 42.8 6.3 9 117 26
Trip expenditures 102.9 6.8 89 112 3
Not speci®ed 237.5 106.2 131 344 2
Elicitation format 10.1 (0.01)
Open-ended 37.4 6.5 1 137 35
Dichotomous choice 91.2 17.1 3 344 29
Iterative bidding 78.5 14.9 9 244 20
Payment card 47.1 8.4 10 174 19
a
Number of observations does not sum up to 103 in all cases as a result of missing values
b
Outcome of the non-parametric Kruskal±Wallis test statistic which has approximately a Chi-squared distribution under the null
hypothesis of equal average WTP in all groups
c
Outcome of the non-parametric Mann±Whitney test statistic for two independent samples which has approximately a standard
normal distribution under the same null hypothesis
WTP is reduced, on average, by 41% (ceteris paribus)
The estimated models account for approximately 37% of
when using studies in a value transfer exercise which are
the observed variability in the mean WTP values found in
based on an open-ended WTP question.
individual studies.
The basic model also indicates that study location has
For the ®xed part of the basic and extended model, the
a signi®cant impact on average WTP. The dummy
estimated coef®cients in the semi-log function represent
variable has a value of 1 if the research took place in North
the constant proportional rate of change in the dependent
America and zero if in Europe. As shown before, average
variable per unit change in the independent variables
WTP is substantially higher in North America than in
(Johnston 1984). Hence, the coef®cient estimated for the
Europe. The parameter estimates for the four main wet-
dummy variable `Payment vehicle' in the basic model re-
land functions are particularly interesting. These functions
¯ects, ceteris paribus, an almost twice as higher average
are found to have a statistically signi®cant role in
WTP for an increase in income tax than for any other
explaining variance in average WTP. The size of the
payment vehicle. Compared with other elicitation formats,
53
Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
Table 3
Generalised least squares results for the basic and extended model. As a result of missing values for explanatory variables,
number of observations is reduced from 103 to 92
Parameter Parameter de®nition Basic model Extended model
Estimate Standard error Estimate Standard error
Fixed effects
Constant Intercept 3.356*** 0.100 3.311*** 0.247
Payment vehicle Dummy: 1 = income tax; 0 = other 1.880*** 0.265 1.576*** 0.362
Elicitation format Dummy: 1 = open-ended; 0 = other A0.411** 0.130 A0.376* 0.183
Country Dummy: 1 = North America; 0 = other 1.861*** 0.217 1.629*** 0.363
Response rate (1) Dummy: 1 = 30±50%; 0 = other A2.253*** 0.326 A1.722*** 0.451
Response rate (2) Dummy: 1 = >50%; 0 = other A1.904*** 0.333 A1.461** 0.450
Flood control Dummy: 1 = ¯ood control; 0 = other 1.477*** 0.240 1.134* 0.456
Water generation Dummy: 1 = water generation; 0 = other 0.691* 0.342 0.441 0.479
0.545 0.282 0.659* 0.327
Water quality Dummy: 1 = water quality; 0 = other
Random effects
Between studies
r2 Variance ± ± 0.160* 0.071
constant
Between average WTP
r2 Variance 0.059* 0.029 0.045 0.028
constant
rpayment vehicle, constant Covariance 0.020 0.043 0.001 0.036
rcountry, constant Covariance 0.689** 0.222 0.351** 0.129
rcountry, payment vehicle Covariance A0.707** 0.226 A0.345** 0.134
r¯ood control, constant Covariance A0.013 0.050 0.027 0.060
A0.266
rwater generation, constant Covariance A0.637** 0.227 0.153
A0.424
rwater quality, constant Covariance 0.231 A0.188 0.135
LLunconstrained A83.907 A81.874
Pseudo R-squared 0.365 0.380
n 92 92
Signi®cant at 0.10; * signi®cant at 0.05; ** signi®cant at 0.01; *** signi®cant at 0.001
effects is as follows: the variance of the constant in the
estimated parameters indicates that average WTP is, as
basic and extended model is the variance associated with
before, highest for ¯ood control, but this time, whilst
the baseline case, i.e. where the value of all the explanatory
controlling for other explanatory factors, followed by
variables is zero. The variance of, for example, North
water generation and water quality and lowest for the
American studies in the basic model is
wetland function biodiversity supply. The latter is used as
(0.059 + 2 ´ 0.689) 1.437. Hence, North American
the baseline category in the regression analysis in order to
studies are more variable than European ones. In this way
avoid multicolinearity. The positive parameter estimates
heteroscedasticity can be modelled in the basic model.
for the three other wetland functions indicate that these
Another example is North American studies using income
functions generate higher values than the baseline function
taxation as a payment vehicle:
biodiversity supply.
This suggests the prominence of use over non-use moti-
r2
vations underpinning stated WTP amounts. The distinc- countryY payment vehicle 0X059 2 Â 0X689
tion between use and/or non-use values does not have a
2 Â 0X020 À 2 Â 0X707 0X023X
signi®cant impact on average WTP, probably because the
corresponding variance is already accounted for by the
distinction between wetland functions. Also, relative wet- The last term in this equation is the covariance between
land size is statistically not signi®cant. Higher response payment vehicle and country. So, wetland CV studies
rates, a rough indicator of better overall study quality, based on income taxation in the US appear to have a
appear to result in signi®cantly lower average WTP than particularly low variance.
low response rates. A practical explanation may be that Accounting for study-level effects in the basic model
low response rates are sometimes biased towards includ- signi®cantly reduces the sample variance or standard
ing a relatively large number of sample respondents with a deviation of average WTP in the extended model
(v2 4X06; P < 0.05). The extended, multilevel model ac-
greater interest than average in environmental protection 1
and corresponding WTP. counting for the random effects between studies hence
The model's random effects can be used to (1) model provides a signi®cant improvement over conventional
heteroscedasticity and (2) investigate the suitability of meta-analysis by allowing for the hierarchical structure of
using speci®c CV results in a value transfer exercise. This data implicit in clustering of multiple results from
will be explained below. The interpretation of the random single studies. As expected, having explained some of the
54 Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
with the resource. It is people's perception of a good's
variance in the model by study-level effects, the random
characteristics or functions that in¯uence their attitudes
effects between log(WTP) amounts have decreased, except
and behaviour, not necessarily the good's `objective'
for the covariance between ¯ood control and the intercept.
characteristics (e.g. see Adamowicz et al. 1997). Finally,
However, in the extended model the ®xed effects have
accounting for intra-study variability, the statistical anal-
slightly decreased, while the signi®cance level of half of the
ysis produces slightly different results regarding the sig-
explanatory variables is lower. Only the signi®cance of the
ni®cance and size of the effect of the main functions on the
wetland function water quality has increased. The wetland
WTP values. The low number of observations also has to
function water generation has become statistically insig-
be taken into consideration.
ni®cant and has switched place with the function water
On the other hand, the study progressed meta-analytical
quality supply in the order of size, suggesting that the
research in environmental economics by providing a
results must be viewed with some caution when properly
statistical multilevel model which accounts for the clus-
accounting for intra-study variability.
tering of results from the same studies, for example as a
If low variance is considered an estimate of quality in the
result of identical survey design or sample population,
sense that study results are better suited for comparison
and owing to the fact that results from some studies may
and hence can be more readily put together in a value
be more variable than others. In the GLS models used,
transfer exercise, then it can be concluded that, on the
the variance and covariance estimators not only enabled
whole, studies using income taxation as a payment vehicle
us to model heteroscedasticity, but also provided im-
are better suited than other payment vehicles, and that
portant background information for environmental value
studies valuing wetland biodiversity tend to be less
transfer.
variable than studies valuing wetlands in their capacity
Finally, although considerable effort has been put into
of generating water or maintaining water quality.
specifying the characteristics of the environmental func-
tions and correspondingly the environmental goods and
services involved, other important aspects that may have
Discussion and conclusions helped to explain differences in valuation outcomes re-
main unde®ned. This is a common problem in meta-an-
In this paper, estimates for socio-economic use and non- alyses in the ®eld of environmental valuation as a result of
use values attached to different hydro-ecological and bio- insuf®cient and inadequate information provided in pub-
lished valuation studies. Relevant information about the
geochemical wetland functions were compared and syn-
samples' socio-economic values is missing in many stud-
thesised in a meta-analysis of wetland CV studies. The
meta-analysis provides insights into the factors that have to ies, let alone respondents' socio-psychological and cultural
be considered when attempting to transfer environmental characteristics.
values on the basis of CV studies. A statistically signi®cant In meta-analysis, inferences are made on the basis of
breakdown of WTP values for four main wetland functions information on global statistics, such as the mean and
has been presented. Although single ecosystem character- standard deviations of parameter estimates. These may or
may not describe individual behaviour adequately. In or-
istics or functions are given meaning and value within
der to overcome this potential problem and to increase the
existing ecosystem structures, the distinction between
functions is essential for a valid transfer of the economic study's validity and reliability, a logical next step would be
to gather more information about sample population
values generated within an ecosystem's primary self-or-
characteristics by complementing the analysis with the
ganising capacity. It reduces the risk of double counting
underlying individual responses. This will provide an
when attempting to assess a natural resource's total eco-
important test of the appropriateness of meta-analysis as
nomic value on the basis of different valuation studies.
an instrument to synthesise CV outcomes for the purpose
From an anthropocentric point of view, the size of the
of value transfer.
estimated parameters in the estimated basic model is as
expected. Average WTP is highest for ¯ood control, be-
cause of the possible risks to life and livelihood as a result Acknowledgements The Centre for Social and Economic
of ¯ooding and the capacity of wetlands to reduce this risk, Research on the Global Environment (CSERGE) is a designated
research centre of the UK Economic and Social Research Council
followed by water supply and water quality and ®nally the (ESRC). This research paper is part of the European Commission
provision and maintenance of biodiversity. However, these Directorate General XII-funded project `Ecological±Economic
results have to be handled with care for a number of Analysis of Wetlands' (ECOWET), contract no. ENV4-CT96±
reasons. 0273, co-ordinated by CSERGE.
Ecosystem structures and processes provide a heteroge-
neous complex of highly interrelated socio-economic
functions. The analysis presents a simple and arbitrary
References
breakdown of these functions into independent compo-
nents. The distinction between four main wetland func- ADAMOWICZ W, SWAIT J, BOXALL P, LOUVIERE J, WILLIAMS M (1997)
DAMOWICZ WAIT OXALL OUVIERE ILLIAMS
tions does also not necessarily correspond with people's Perceptions versus objective measures of environmental quality
perception of the various functions wetlands perform, in combined revealed and stated preference models of envi-
which depends upon their own knowledge and experience ronmental valuation. J Environ Econ Manage 32:65±84
55
Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
DESVOUSGES WH, NAUGHTON MC, PARSONS GR (1992) Bene®t
ARROW K, SOLOW R, PORTNEY PR, LEAMER EE, RADNER R, SCHUMAN H
RROW OLOW ORTNEY EAMER ADNER CHUMAN ESVOUSGES AUGHTON ARSONS
(1993) Report of the NOAA Panel on Contingent Valuation. transfer: conceptual problems in estimating water quality ben-
Federal Register (15 January) 58(10):4601±4614 e®ts using existing studies. Water Resour Res 28(3):675±683
BARBIER EB (1993) Sustainable use of wetlands. Valuing tropical DUGAN PJ (ed) (1990) Wetland conservation. A review of current
ARBIER UGAN
wetland bene®ts: economic methodologies and applications. issues and required action. World Conservation Union, Gland,
Geogr J 159(1):22±32 Switzerland
BATEMAN IJ, LANGFORD IH, TURNER RK, WILLIS KG, GARROD GD FARBER S (1988) The value of coastal wetlands for recreation: an
ATEMAN ANGFORD URNER ILLIS ARROD ARBER
(1995) Elicitation and truncation effects in contingent valuation application of travel cost and contingent valuation methodolo-
studies. Ecol Econ 12(2):161±179 gies. J Environ Manage 26:299±312
BATEMAN IJ, MUNRO A, RHODES B, STARMER C, SUGDEN R (1997) Does GARROD GD, WILLIS KG (1996) Estimating the bene®ts of envi-
ATEMAN UNRO HODES TARMER UGDEN ARROD ILLIS
part±whole bias exist? An experimental investigation. Econ J ronmental enhancement: a case study of the River Darent. J
107(441):322±332 Environ Plann Manage 39(2):189±203
BERGH van den JCJM, BUTTON KJ, NIJKAMP P, PEPPING GC GLASS GV, MCGAW B, SMITH ML (1981) Meta-analysis in social
ERGH UTTON IJKAMP EPPING LASS C AW MITH
(1997) Meta-analysis in environmental economics. Kluwer, research. Sage Publications, Beverly Hills
Dordrecht GREEN CH, TUNSTALL SM (1991) The evaluation of river water
REEN UNSTALL
BERGSTROM JC, STOLL JR, TITRE JP, WRIGHT VL (1990) Economic quality improvements by the contingent valuation method. Appl
ERGSTROM TOLL ITRE RIGHT
value of wetlands-based recreation. Ecol Econ 2:129±147 Econ 23:1135±1146
BISHOP RC, BOYLE KJ (1985) The economic value of Illinois Beach GREENLEY DA, WALSH RG, YOUNG RA (1981) Option value: empir-
ISHOP OYLE REENLEY ALSH OUNG
State Nature Reserve. Consultancy report. HBRS, Madison, ical evidence from a case study of recreation and water quality.
Wisconsin Q J Econ 96(4):657±673
È
BISHOP RC, BOYLE KJ, WELSH MP (1987) Toward total economic GREN I-M, SoDERQVIST T (1994) Economic valuation of wetlands:
ISHOP OYLE ELSH REN
valuation of Great Lakes ®shery resources. Trans Am Fish Soc a survey. Beijer Discussion Paper Series no. 54, Beijer Institute,
116:339±345 Stockholm
BLAMEY RK (1995) Citizens, consumers and contingent valuation: HAMILTON LC (1993) Statistics with Stata 3. Duxbury Press,
LAMEY AMILTON
an investigation into respondent behaviour. PhD Diss, Austra- Belmont, California
lian National University, Canberra IMF (1996) International ®nancial statistics yearbook. Interna-
BROUWER R, LANGFORD IH, BATEMAN IJ, CROWARDS TC, TURNER RK tional Monetary Fund, Washington
ROUWER ANGFORD ATEMAN ROWARDS URNER
(1997) A meta-analysis of wetland contingent valuation studies. JOHNSTON J (1984) Econometric methods, 3rd edn. McGraw-Hill,
OHNSTON
Global Environmental Change Working Paper 97±20, Centre for Maidenhead
Social and Economic Research on the Global Environment JORDAN JL, ELNAGHEEB AH (1993) Willingness to pay for improve-
ORDAN LNAGHEEB
(CSERGE), University of East Anglia, Norwich, and University ments in drinking water quality. Water Resour Res 29(2):237±
College, London 245
BROUWER R, SLANGEN LHG (1998) Contingent valuation of the KAORU Y (1993) Differentiating use and nonuse values for coastal
ROUWER LANGEN AORU
public bene®ts of agricultural wildlife management: the case of pond water quality improvements. Environ Resour Econ 3:487±
Dutch peat meadow land. Eur Rev Agric Econ 25:53±72 494
BROUWER R (1998) Future research priorities for valid and reliable KOSZ M (1996) Valuing riverside wetlands: the case of the
ROUWER OSZ
environmental value transfer. Global Environmental Change ``Donau-Auen'' National Park. Ecol Econ 16:109±127
Working Paper 98±28, Centre for Social and Economic Research LANGFORD IH (1994) Using a generalized linear mixed model to
ANGFORD
on the Global Environment (CSERGE), University of East An- analyze dichotomous choice contingent valuation data. Land
glia, Norwich, and University College, London Econ 70(4):507±514
CARSON RT, MITCHELL RC (1993) The value of clean water: the LANGFORD IH, BATEMAN IJ, JONES AP, LANGFORD HD, GEORGIOU S
ARSON ITCHELL ANGFORD ATEMAN ONES ANGFORD EORGIOU
public's willingness to pay for boatable, ®shable, and swimm- (1998) Improved estimation of willingness to pay in dichoto-
able quality water. Water Resour Res 29(7):2445±2454 mous choice contingent valuation studies. Land Econ 74(1):65±
CARSON RT, FLORES NE, MARTIN KM, WRIGHT JL (1996) Contingent 75
ARSON LORES ARTIN RIGHT
valuation and revealed preference methodologies: comparing LANT CL, ROBERTS RS (1990) Greenbelts in the cornbelt: riparian
ANT OBERTS
the estimates for quasi-public goods. Land Econ 72:80±99 wetlands, intrinsic values, and market failure. Environ Plan A
COOPER J, LOOMIS JB (1991) Economic value of wildlife resources in 22:1375±1388
OOPER OOMIS
the San Joaquin Valley: hunting and viewing values. In: Dinar A, LOOMIS JB (1987) Balancing public trust resources of Mono Lake
OOMIS
ZILBERMAN D (eds) The economics and management of water and Los Angeles' water right: an economic approach. Water
ILBERMAN
drainage. Kluwer, Dordrecht Resour Res 23(8):1449±1456
COSTANZA R, d'ARGE R, de GROOT R, FARBER S, GRASSO M, HANNON B, LOOMIS JB, HANEMANN M, KANNINEN B, WEGGE T (1991) Willingness
OSTANZA ARGE ROOT ARBER RASSO ANNON OOMIS ANEMANN ANNINEN EGGE
LIMBURG K, NAEEM S, O'NEILL RV, PARUELO J, RASKIN RG, SUTTON P, to pay to protect wetlands and reduce wildlife contamination
IMBURG AEEM EILL ARUELO ASKIN UTTON
van den BELT M (1997) The value of the world's ecosystem from agricultural drainage. In: Dinar A, Zilberman D (eds) The
ELT
services and natural capital. Nature 387:253±260 economics and management of water drainage. Kluwer, Do-
CROWARDS TC, TURNER RK (1996) FAEWE sub-project report: rdrecht
ROWARDS URNER
economic valuation of wetlands. Centre for Social and Economic MADDALA GS (1983) Limited-dependent and qualitative variables
ADDALA
Research on the Global Environment (CSERGE), University of in econometrics. Cambridge University Press, Cambridge
East Anglia, Norwich, and University College, London MCFADDEN D (1994) Contingent valuation and social choice. Am J
C ADDEN
CUMMINGS RG, GANDERTON PT, MCGUCKIN T (1994) Substitution Agric Econ 76:689±708
UMMINGS ANDERTON C UCKIN
effects in cvm values. Am J Agric Econ 76:205±214 MITCHELL RC, CARSON RT (1989) Using surveys to value public
ITCHELL ARSON
DESVOUSGES WH, SMITH VK, MCGIVNEY MP (1983) A comparison of goods: the contingent valuation method. Resources for the Fu-
ESVOUSGES MITH C IVNEY
alternative approaches for estimation of recreational and related ture, Washington, DC
bene®ts of water quality improvement. Report to the US Envi- OLSEN D, RICHARDS J, SCOTT RD (1991) Existence and sport values
LSEN ICHARDS COTT
ronmental Protection Agency, Washington, DC for doubling the size of Columbia River Basin salmon and
DESVOUSGES WH, SMITH VK, FISHER A (1987) Option price estimates steelhead runs. Rivers 2(1):44±56
ESVOUSGES MITH ISHER
for water quality improvements: a contingent valuation study PEARCE DW, TURNER RK (1990) Economics of natural resources
EARCE URNER
for the Monongahela River. J Environ Econ Manage 14:248±267 and the environment. Harvester Wheatsheaf, Hemel Hempstead
56 Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
PHILLIPS WE, HANEY TJ, ADAMOWICZ WL (1993) An economic SUTHERLAND RJ, WALSH RG (1985) Effect of distance on the
HILLIPS ANEY DAMOWICZ UTHERLAND ALSH
analysis of wildlife habitat preservation in Alberta. Can J Agric preservation value of water quality. Land Econ 61(3):281±291
Econ 41:411±418 TOLBA MK, EL-KHOLY OA (1992) The world environment 1972±
OLBA L HOLY
RAMSAR (1975) Convention on wetlands of international impor- 1992. Chapman and Hall on behalf of the UN Environment
AMSAR
tance especially as waterfowl habitat. Ramsar Convention, Iraq Programme, London
RASBASH J, WOODHOUSE G (1995) MLn command reference. TURNER RK (1992) Policy failures in managing wetlands. In: OECD
ASBASH OODHOUSE URNER
Institute of Education, University of London market and government failures in environmental management.
SANDERS LD, WALSH RG, LOOMIS JB (1990) Toward empirical esti- OECD, Paris
ANDERS ALSH OOMIS
mation of the total value of protecting rivers. Water Resour Res WHITEHEAD JC, BLOMQUIST GC (1991) Measuring contingent values
HITEHEAD LOMQUIST
26(7):1345±1357 for wetlands: effects of information about related environmental
SCHULTZ SD, LINDSAY BE (1990) The willingness to pay for goods. Water Resour Res 27(10):2523±2531
CHULTZ INDSAY
groundwater protection. Water Resour Res 26(9):1869±1875 WILLIS KG (1990) Valuing non-market wildlife commodities: an
ILLIS
SILVANDER U (1991) The willingness to pay for angling and ground evaluation and comparison of bene®ts and costs. Appl Econ
ILVANDER
water in Sweden. PhD Diss, Swedish University of Agricultural 22:13±30
Sciences, Uppsala WILLIS KG, GARROD GD, SAUNDERS CM (1995) Bene®ts of environ-
ILLIS ARROD AUNDERS
SPANINKS FA (1993) Een Schatting van de Sociale Baten van Be- mentally sensitive area policy in England: a contingent valuation
PANINKS
heersovereenkomsten met behulp van de Contingent Valuation assessment. J Environ Manage 44:105±125
Methode. MSc Diss, Department of Agricultural Economics and WOLF FM (1986) Meta-analysis. Sage Publications, Beverly Hills,
OLF
Policy, Wageningen Agricultural University California
SPANINKS FA, KUIK OJ, HOOGEVEEN JGM (1996) Willingness to pay of
PANINKS UIK OOGEVEEN
Dutch households for a natural Wadden Sea. An application of
the contingent valuation method. Report E-96/6, Institute for
Environmental Studies, Amsterdam
57
Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
A meta-analysis of wetland
contingent valuation studies
R. Brouwer á I.H. Langford á I.J. Bateman á R.K. Turner
asked for either their willingness to pay (WTP) or their
willingness to accept (WTA) compensation for the gains or
Abstract There is growing interest in the potential
losses involved (Mitchell and Carson 1989). Of these op-
for producing generally applicable models for
tions the WTP approach has become the most frequently
valuing non-market environmental services which
applied and has been given peer review endorsement
do not rely upon expensive and time-consuming
through a variety of studies (see, for example, Arrow et al.
survey work, but rather extrapolate results from
1993). When aggregated across those who will be affected
previous studies. This paper presents a meta-
by the suggested environmental changes, this stated WTP
analysis for the use and non-use values generated by
amount is used as a socio-economic indicator of the
wetlands across North America and Europe. The
environmental values involved.
study assesses the socio-economic values
Given the substantial indirect, often off-site, use and non-
attributable to the hydrological, biogeochemical and
use values involved, wetlands have been the focus of at-
ecological functions provided by such complex
tention in several CV studies (Crowards and Turner 1996).
environmental assets. The clustering of multiple
Many of these studies try to estimate the total economic
values derived from single studies is examined
value of wetlands. Total economic value, not to be
through the application of multilevel modelling
confused with total ecosystem value, consists of use and
methods allowing for the hierarchical structure of
non-use values (Pearce and Turner 1990). CV is the only
such data.
economic method to date that is able, in principle, to
account for possible non-use motivations underlying
Key words Meta-analysis á Contingent valuation á
people's value statements. Whereas use values refer to the
Wetlands á Ecosystem functions
values associated with the actual use of the various goods
and services wetlands provide, non-use values are unre-
lated to any actual or potential use of these goods and
services.
Introduction Wetlands are complex hydro-ecological systems, whose
structure provides us with goods or products involving
some direct utilisation of one or more wetland character-
This paper addresses the socio-economic values of the
istics, while wetland ecosystem processes provide us with
various functions performed by wetland ecosystems.
hydrological and ecological services, supporting or pro-
Environmental economists have developed a variety of
techniques for measuring such values, of which the con- tecting human activities or human properties without
tingent valuation (CV) method is probably the most widely being used directly. The stock of wetlands is a multi-
applied in contemporary research. CV is a survey method functional resource with signi®cant economic value, as
also has been suggested by Costanza et al. (1997). Fifteen
where individuals are presented with information about
percent of the value of the world's ecosystem services and
speci®c environmental changes, and their perception,
natural capital is generated by wetlands (Costanza et al.
attitudes and preferences regarding these changes are
1997). However, all over the world countries have experi-
elicited. In order to measure the effect of the suggested
enced severe wetland losses (Tolba and El-Kholy 1992;
changes on people's welfare, respondents are typically
Turner 1992). Sustainable management of these assets is
highly relevant. Since this management process is not
costless, they require accurate and meaningful valuation in
order to be able to weigh the costs and bene®ts of their
Received: 24 February 1999 á Accepted: 6 June 1999
conservation.
R. Brouwer (8) á I.H. Langford á I.J. Bateman á R.K. Turner In this paper, the main ®ndings of CV studies of
Centre for Social and Economic Research on the Global
wetlands in temperate climate zones in developed
Environment (CSERGE), School of Environmental Sciences,
economies will be investigated. The main objective is
University of East Anglia, Norwich, UK
to quantify the socio-economic values associated with
Tel.: 01603 593741; fax: 01603 593739,
wetland ecosystem functioning in a meta-analysis of
e-mail: r.brouwer@uea.ac.uk
47
Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
to increase policymaker demand for transferable valuation
wetland CV studies, supplementing qualitative analyses
È results.
provided, for example, by Gren and Soderqvist (1994)
Meta-analysis enables researchers to identify criteria for
or Crowards and Turner (1996). Natural and social
valid environmental value transfer1 or to test the conver-
science are brought together by relating the various
hydrological, biogeochemical and ecological wetland gent validity of value estimates. In the ®rst case the data-
functions to the societal bene®ts derived from these set is entirely used to determine the factors that help to
functions and the socio-economic values attached signi®cantly explain variances in valuation outcomes. In
to these bene®ts. the second case the data-set can be split, for example into
two parts, one of which is used for the ®rst purpose and
the other to test whether the value estimates based on the
signi®cant factors fall within the con®dence interval of the
Approach other half's estimates.
Environmental value transfer is commonly de®ned as the
transposition of monetary environmental values estimated
The results from 30 different CV studies of wetlands in
at one site (study site) through market-based or non-
temperate climate zones in developed economies were
market-based economic valuation techniques to another
compared and synthesised in a meta-analysis. Only very
site (policy site). The most important reason for using
few tropical wetland valuation studies exist (Barbier 1993).
previous research results in new policy contexts is cost
Tropical wetland studies are excluded from the
effectiveness. Applying previous research ®ndings to
meta-analysis presented here because of the enormous
similar decision situations is a very attractive alternative
differences between population samples in developed and
to expensive and time-consuming original research to
developing countries regarding socio-cultural and demo-
quickly inform decision making.
graphic-economic characteristics.
The criteria for selecting studies for environmental value
Since the beginning of the 1990s, meta-analysis has been
transfer suggested in the literature focus on the environ-
playing an increasingly important role in environmental
mental goods involved, the sites in which the goods are
economics research (van den Bergh et al. 1997). Originally
found, the stakeholders and the study quality (Desvousges
a technique used in experimental medical treatment and
et al. 1992). However, very little published evidence exists
psychotherapy, meta-analysis is the statistical evaluation
of studies that test the validity of environmental value
of the summary ®ndings of empirical studies, helping to
transfer. Moreover, in the few studies that have been
extract information from large masses of data in order to
carried out, the transfer errors are substantial (Brouwer
quantify a more comprehensive assessment. It enables
1998).
researchers to explain differences in outcomes found in
As more information about factors in¯uencing environ-
single studies on the basis of differences in underlying
mental valuation outcomes becomes available, for in-
assumptions, standards of design and/or measurement. As
stance through the meta-analysis presented here,
such, meta-analysis is an important extension of quanti-
transfers across populations and sites become more
tative analyses and can be seen as a supplement to quali-
practicable.
tative analysis.
Compared to qualitative analysis, important advantages
of meta-analysis are that on the `input' side it does not
prejudge research ®ndings on the basis of the original
Dataset and study characteristics
study's quality, while it avoids a differential subjective
weighting of studies in the interpretation of a set of
®ndings on the `output' side (Glass et al. 1981). How- The list of wetland CV studies included in the meta-
ever, one drawback is that it may be biased towards analysis is presented in Table 1. Most studies have been
including signi®cant study results only, since it may well published in journals. Half of all studies were carried
be that insigni®cant study results will not be published. out between 1985 and 1989, with most being published in
Furthermore, multiple results from the same study are the ®rst 3 years of the 1990s. One study was carried
often treated as individual, independent observations out in the 1970s, 19 in the 1980s and 10 in the 1990s.
without explicit testing for intra-study correlation (Wolf Besides the inclusion of published signi®cant results,
1986). Table 1 illustrates two other problems in this
In the ®eld of environmental valuation, meta-analyses
have focused on a range of environmental issues from
outdoor recreation to urban air pollution, based on single
1
The term `environmental value transfer' is used here instead of
or multiple valuation techniques. The increase in meta-
the popular term `bene®ts transfer', because CV can also measure
analytical research seems to be triggered principally by (1)
the bene®ts foregone, which makes the estimated values costs
increases in the available number of environmental valu-
instead of bene®ts. WTP is the conventional economic approach
ation studies, (2) the seemingly large differences in valu- to measure environmental values in money and hence make them
ation outcomes as a result of the use of different research commensurable with other market values ± costs and bene®ts ±
designs (Carson et al. 1996), and (3) the high costs of associated with decisions that have been made, are made or have
carrying out environmental valuation studies which tend to be made in the face of limited human and natural resources
48 Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
Table 1
Studies (with year of publication in parentheses) included in the meta-analysis
Type of publicationa nb
Authors Study year
1 Bateman et al. (1995) Journal article (EE) 1991 3
2 Bergstrom et al. (1990) Journal article (EE) 1986 1
3 Bishop and Boyle (1985) Consultancy report 1985 2
4 Bishop et al. (1987) Journal article (TAFS) 1985 7
5 Brouwer and Slangen (1998) Journal article (ERAE) 1994 3
6 Carson and Mitchell (1993) Journal article (WRR) 1983 3
7 Cooper and Loomis (1991) Book chapter 1987 3
8 Cummings et al. (1994) Journal article (AJAE) 1992 2
9 Desvousges et al. (1987) Journal article (JEEM) 1981 21
10 Farber (1988) Journal article (JEM) 1984 1
11 Garrod and Willis (1996) Journal article (JEPM) 1993 4
12 Green and Tunstall (1991) Journal article (AE) 1986 1
13 Greenley et al. (1981) Journal article (QJE) 1976 4
14 Silvander (1991) Dissertation 1989 2
15 Jordan and Elnagheeb (1993) Journal article (WRR) 1991 2
16 Kaoru (1993) Journal article (ERE) 1989 1
17 Kosz (1996) Journal article (EE) 1993 1
18 Lant and Roberts (1990) Journal article (EPA) 1987 6
19 Loomis et al. (1991) Book chapter 1989 10
20 Loomis (1987) Journal article (WRR) 1985 1
21 Olsen et al. (1991) Journal article (Rivers) 1989 3
22 Phillips et al. (1993) Journal article (CJAE) 1991 2
23 Sanders et al. (1990) Journal article (WRR) 1983 2
24 Schultz and Lindsay (1990) Journal article (WRR) 1988 1
25 Spaninks (1993) MSc thesis 1993 3
26 Spaninks et al. (1996) Scienti®c report 1995 2
27 Sutherland and Walsh (1985) Journal article (LE) 1981 2
28 Whitehead and Blomquist (1991) Journal article (WRR) 1989 6
29 Willis (1990) Journal article (AE) 1986 2
30 Willis et al. (1995) Journal article (JEM) 1992 2
a
Abbreviations: AE Applied Economics; AJAE American Journal Journal of Environmental Management; JEPM Journal of
of Agricultural Economics; CJAE Canadian Journal of Agri- Environmental Planning and Management; LE Land Economics;
cultural Economics; EE Ecological Economics; EPA Environment QJE Quarterly Journal of Economics; TAFS Transactions of the
and Planning A, ERAE European Review of Agricultural Eco- American Fisheries Society; WRR Water Resources Research
b
Number of observations taken from each study
nomics; ERE Environmental and Resource Economics; JEEM
Journal of Environmental Economics and Management; JEM
The studies included in the analysis focus primarily on
meta-analysis. First, a number of people have been
wetlands or wetland-type areas.2 The speci®c WTP
involved in several studies and related publications. This
may result in an `authorship' effect. Learning from pre- questions addressed in each study cover a large con-
vious studies, authors may use similar, perhaps slightly tinuum of activities, actions or projects related to wet-
adapted survey designs in subsequent studies. Secondly, lands, but in some cases (approximately a third of all
103 data points (observations) were extracted from 30 studies) also to water resources in general. These values
studies. This corresponds, on average, to three or four were kept in the analysis because they referred directly
observations per study. More than half of all studies to the hydrological wetland functions distinguished in
provided one or two observations. Outliers are the the analysis and were considered reliable estimates for
studies by Loomis et al. (1991) and Desvousges et al. these functions. The WTP questions range from outdoor
(1987), providing 10 and 21 observations respectively. recreational activities like birdwatching or ®shing, to
Studies provided more than one observation mainly
because of the use of split survey samples targeting
different wetland user and non-user groups and testing 2
Although there is little agreement among scientists on what
different survey designs. The possibility that results from
constitutes a wetland, a workable de®nition is given by the so-
the same study cluster together, for example as a result of
called Ramsar Convention (1975, article 1): `areas of marsh fen,
identical survey design or sample population, and that peatland or water, whether natural or arti®cial, permanent or
results from some studies may be more variable than temporary, with water that is static or ¯owing, fresh, brackish or
others was tested and accounted for in the meta-analysis salt including areas of marine water, the depth of which at low
(see `The model'). tide does not exceed 6 m'
49
Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
in all other cases wetland functions provided multiple
ground-water protection and complete wildlife habitat
bene®ts.
preservation. Two-thirds of the studies are carried out in
Most studies (70%) asked respondents for the use and
the USA, the rest in Europe. Half of the European
non-use values attached to the bene®ts derived from
studies were carried out in the UK.
wetland functions. One study (Schultz and Lindsay
Separating the heterogeneous complex of hydrological,
1990), elicited future use value only (option value),
biogeochemical and ecological functions performed by the
while another (Greenley et al. 1981) focused, among
wetlands considered in each study in the meta-analysis
others, on the ex ante `option price' of being able to
was very dif®cult. Based on the various functions ad-
make a better informed judgement in the future based
dressed in the reviewed studies, a simple distinction was
on more information becoming available regarding wa-
made therefore ®rst of all between four main wetland
ter quality.
ecosystem functions: ¯ood control, water generation, wa-
In eight studies an attempt was made to break down the
ter quality support and wildlife habitat provision (Fig. 1).
stated total economic value ex post in the questionnaire
Secondly, the main function valued in each study was as-
into the various components distinguished in the litera-
signed to one of each of these four groups. Hence, each
ture, e.g. use, option, philantrophic, bequest, stewardship
study was categorised as addressing one of these four main
and existence value. In two water quality studies (De-
wetland functions, unless a study explicitly generated
svousges et al. 1987; Carson and Mitchell 1993), respon-
distinct values for different wetland functions (for more
dents were presented ex ante with a `value card' which
details, see Brouwer et al. 1997).
described the main reasons why water quality might be
Obviously, wetland ecosystem structures and processes
valued. In another two studies use and non-use values
and the functions they provide are highly interrelated,
were elicited separately, by the use of either different
making it very hard, and in some cases impossible, to
questionnaires (Bishop et al. 1987) or separate questions
distinguish between individual functions. They often go
for use and non-use values in the same questionnaire
hand in hand and attempts to separate them, for example
(Greenley et al. 1981).
for economic valuation purposes in order to avoid double
Finally, two study quality indicators were included in
counting, are liable to be arbitrary. This implies that
the analysis: one for the quality of the studies included in
double counting is a real problem and attempts to aggre-
the meta-analysis and one for the quality of the meta-
gate up to system-level values are fraught with dif®culties.
analysis itself. The quality of individual studies is indicated
Also, in the case of the human bene®ts derived from the
by the study response rate and the quality of the meta-
wetland functions involved, complete separation of direct
analysis by the so-called scope test. Both indicators are
and indirect use and non-use bene®ts is dif®cult. Only in a
found back in the National Oceanic and Atmospheric
third of all studies could a single bene®t ¯ow be identi®ed;
Wetland ecosystem
structure and processes
Functions
Hydrological Biogeochemical Ecological
1) Flood water retention 3) Nutrient retention and 4) Nursery and habitat for plants,
2) Surface and groundwater export animals and micro-organisms
recharge and landscape structural
Fig. 1
diversity
Main wetland ecosystem functions
identi®ed in the meta-analysis and their
derived socio-economic bene®ts. The
analysis is based on stated WTP (will-
Socio-economic benefits
ingness to pay) for goods and services for
- Natural flood protection - Improved water quality - Fishing
which no market exists. The value of
alternative - Waste disposal - Wildfowl hunting
marketed products such as reed or ®sh is
- Reduced damage to infra- - Other recreational amenities
excluded from the analysis to avoid
structure, property and crops
double counting with the stated use and
- Water supply
non-use values
- Habitat maintenance
50 Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
gression technique called multilevel modelling was used
Administration's (NOAA's) `burden of proof' requirements
(Arrow et al. 1993).3 (Langford 1994; Langford et al. 1998). [The GLS regression
was carried out using the package MLN (Rasbash and
A high non-response, either to the entire survey instru-
Woodhouse 1995).]
ment or to the valuation question, raises concern regard-
Given the shape of the distribution of the WTP amounts,
ing the study's representativeness, and questions the
a logarithmic transformation was used. Hence, for the
validity of the survey design employed and the extent to
log(WTP) amounts of the studies Y, the GLS model is:
which the valuation scenario in the questionnaire was
comprehensible and credible. The scope test refers to the
Y Xb ZH 1
sensitivity of WTP measures to changes in the provision
level of the goods and services being valued, i.e. the where Xb consists of the design matrix X and associated
difference between reference and target provision levels. parameters b represent the mean or ®xed effects of the
Conforming to the strong monotonicity assumption in explanatory variables on the dependent variable Y. How-
neo-classical consumer theory, responses should reveal a ever, whereas in OLS regression there is a single vector of
smaller WTP for smaller amounts of an environmental error terms or residuals, here a more complex variance
commodity provided by an environmental programme. structure may be modelled where the values of residuals
Very few studies reported the extent of protest bids and are dependent on explanatory variables included in the
other questionable responses in the survey. Although most design matrix Z for the random part of the model. For
studies mention the survey response rates, it is in many example, using one explanatory variable for simplicity,
cases not clear what these response rates actually represent Eq. (1) can be written as:
or which criteria have been used to exclude responses
from further analysis. Where such information was yi b0 b1 x1i ui
!
! !
available, protest bids and questionable responses were
r2
ui 0 rum
excluded from the response rates. u
Y
m1i Y $N 2
r2
mi 0 rum
In order to carry out a scope test, the size of the affected m
study site and the difference between the reference and the
where ui is the residual associated with the intercept b0,
target levels of environmental service provision in the CV
and vi is the residual associated with the slope parameter
scenarios should ideally be considered. However, in two-
b1 of x1. While the variance of the responses in OLS is
thirds of all studies no information is provided about the
determined by a single residual term, in the basic GLS
size of the area involved. In about one-third of all studies,
model the variance is dependent on the explanatory
the study site size was estimated using geographical maps.
variable:
Problems accumulate when aiming to also include the
difference between the reference and target levels of the var yi r2 2rum x1 r2 x2 3
u m1
various wetland functions distinguished in the environ-
mental scenarios in each study. The multi-dimensional This can be done for any number of variables, hence
nature of these functions makes a comparison between making the variance of the responses a complex function
studies impossible. Hence, instead a `relative size' variable of the explanatory variables, accounting for he-
was compiled, referring to the share of each study site in teroscedasticity. This turns out to be highly relevant, as
the country's total stock of wetlands. there are signi®cant differences, for example, in the vari-
ance of responses within different studies (intra-study
effects). Using the subscript j to label different studies,
The model the basic previous GLS model can be rewritten as:
yij b0 b1 x1ij uij m1ij x1ij
The structure of the data used in the meta-analysis is ! ! !
r2 rum
complex. WTP values are generated by different studies, ui 0 u
sj Y $N Y Y
carried out in different geographical locations using dif- rum r2
mi 0 m
ferent valuation formats. Using the summary statistics of
sj N 0Y r2 4
these different studies in a pooled sample, the usual con- s
ditions required for ordinary least squares (OLS) regres- where r2 is the variance parameter that describes the
s
sion are likely to be violated. In order to account for differing variability of estimates within different studies.
heteroscedasticity, a generalised least squares (GLS) re- This latter model will be referred to as the `extended'
model in the results section.
3
In 1992, the NOAA commissioned a prestigious `Blue-Ribbon
Panel' of economists and survey specialists, co-chaired by Nobel
Results
laureates Kenneth Arrow and Robert Solow, to investigate the CV
method. After carefully considering a wide range of issues, the
panel's report gave the method a quali®ed bill of health, but only
Summary statistics
if studies were conducted to a rigorous set of guidelines. The
A ®rst step in the meta-analysis was to make stated aver-
panel identi®ed a subset of issues which it called `burden of
age WTP amounts in each study comparable. The response
proof ' requirements
51
Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
surveys were carried out is used as an indicator for income
variable is average WTP per household per year for the
differences to account for people's capacity to pay. Income
preservation of speci®c wetland aspects. After expressing
taxation as a payment vehicle generates the highest aver-
WTP in national currencies in terms of their 1990 pur-
age WTP value, followed by stated WTP over and above
chasing power, these national currencies were converted in
actual trip expenditures to visit a wetland site (use value).
the International Monetary Fund's (IMF's) Special Draw-
The latter is part of a set of payment mechanisms that
ing Rights (SDRs), which is the Fund's of®cial monetary
present wetland functions to respondents as a private
unit of account (IMF 1996). Average WTP for wetland
good, i.e. to be `consumed' by the individual who is being
function preservation found in all studies taken together is
interviewed by asking him or her to pay, for example, an
62 SDRs (by the end of 1995, 1 SDR approximately
entrance fee. On the other hand, general income taxation is
equalled 1.5 US$). The median is considerably lower,
expected to prompt responses that consider the implica-
namely 34 SDRs.
tions of wetland preservation for society at large, not just
The breakdown of WTP values according to a number of
for the individual (Mitchell and Carson 1989; Blamey
possible explanatory factors is presented in Table 2. Mean
1995). Hence, the higher WTP value elicited through this
WTP values have been calculated for (1) wetland types as
payment mechanism is expected to re¯ect more than
identi®ed by Dugan (1990); (2) main wetland functions;
private use values only. (Obviously, `users' of speci®c
(3) relative wetland size; (4) the different value types
wetland functions, for instance people who visit a wetland
elicited in the studies (use and/or non-use values); (5) the
site for recreational ®shing or boating, may hold values
continent where the wetland sites are found; (6) the way
related to their non-use, e.g. preservation for future gen-
people were asked to pay for wetland function preserva-
erations, as well.) The high value for non-speci®ed pay-
tion in the CV survey as part of the institutional setting of
ment modes is due to outliers and the very low number of
the wetland conservation programmes (e.g. through
observations. Calculating through the value of wetland
general income taxation or otherwise); and (7) the way
function preservation in existing product prices yields a
the WTP question was elicited in the CV survey (e.g. in
signi®cantly higher mean WTP than the establishment of a
an open-ended question or otherwise).
private fund or raising entrance fees.
The calculated differences in mean WTP for each of these
Finally, corresponding to previous research results (e.g.
categories are statistically signi®cant at the 5% signi®-
McFadden 1994; Bateman et al. 1995; Willis et al. 1995),
cance level or stronger (see the outcomes of the Chi-
the open-ended (OE) elicitation format yields a signi®-
square test statistic in the last column of Table 2). The
cantly lower WTP than other formats. The dichotomous
range of values (minimum and maximum) found for the
choice format (yes or no to a given bid amount) yields the
factor levels across studies is considerable. Mean WTP
highest average WTP, followed by the iterative bidding
per household is more or less the same for salt- and
procedure (yes or no to a sequence of bid amounts).
fresh-water wetlands. However, the number of observa-
Possible explanations are the larger numbers of non-
tions for salt-water wetlands is very low. Almost all ob-
responses or protest responses OE elicitation tends to
servations refer to fresh-water wetlands. Within fresh-
produce (Desvousges et al. 1983) or the uncertainty
water wetlands, the value of wetlands fed by rivers (riv-
experienced in answering the unfamiliar WTP question for
erine) is twice as high as the value of lakes and ponds
non-market goods and services in an OE format (Bateman
(lacustrine) or marshes and swamps (palustrine).
et al. 1995).
Ground-water is valued highest, although the number of
observations is again low.
Regression results
The wetland function ¯ood control generates the highest
The ®ndings for the basic and extended GLS model in
mean WTP, followed by wildlife habitat provision and
which we account for study level effects are presented in
landscape structural diversity (labelled biodiversity in
Table 3. Only those variables are included that are statis-
Table 2). Surface and groundwater recharge (labelled
tically signi®cant at the 0.1 level. The ®xed part of the
water generation in Table 2) has the lowest value. As
model represents the ®xed or mean effects of each variable,
expected, larger sites result in higher WTP. An incon-
as for an OLS regression model, while the random part
sistency is found between the categories `small' and `very
displays the variance and covariance parameters that
small'.
model heteroscedasticity.
Use values associated with wetland functions are almost
The estimates for the regression results are obtained
twice as high as non-use values. However, a combination
through Maximum Likelihood techniques (e.g. Maddala
of the two is not equal to their sum, suggesting some non-
1983). The outcome of the likelihood ratio test
linear relationship between the two. Socio-psychological
(v2 96X51; P < 0.01) rejects the null hypothesis of zero
and related factors underpinning so-called embedding ef- 14
fects, where the sum of the valuations placed on the parts effects for all explanatory variables. A pseudo R-squared
of a commodity exceeds that for the whole (Bateman et al. was calculated from the log likelihood (LL) function. The
1997), may be one important reason. outcome corresponds with the goodness of ®t measures
North Americans are willing to pay, on average, more than usually found in CV studies. Since the pseudo R-squared
Europeans. Since average income data for the survey lacks the straightforward explained variance interpretation
samples are missing in most of the studies reviewed, the of R-squared in OLS regression (Hamilton 1993), it is used
country in which the wetland sites are found and the CV here as a rough indicator for the model's goodness-of-®t.
52 Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
Table 2
Summary statistics (WTP willingness to pay; SDR Special Drawing Rights)
na v2 (p<)b
Mean WTP (SDRs) Standard error Min (SDRs) Max (SDRs)
Wetland type 15.2 (0.05)
Saltwater 56.2 27.2 19 137 4
Marine 22.7 3.7 19 26 2
Lagoonal 136.6 ± ± ± 1
Lake 42.8 ± ± ± 1
Freshwater 58.9 6.1 1 267 97
Riverine 71.7 13.7 1 267 38
Lacustrine 36.8 9.4 12 88 9
Palustrine 36.9 4.3 9 117 31
Groundwater 125.7 24.3 99 174 3
Fresh- and saltwater 237.5 106.2 131 344 2
Wetland function 7.8 (0.05)
Flood control 92.6 24.4 24 177 5
Water generation 21.5 6.8 3 59 9
Water quality 52.5 5.9 9 174 43
Biodiversity 76.1 12.8 1 344 46
Relative wetland size 13.1 (0.01)
Very large 86.9 17.6 19 177 8
Large 70.3 21.6 12 344 16
Medium 67.0 8.9 3 267 58
Small 29.5 13.2 1 137 13
Very small 53.4 13.8 24 105 6
Value type 6.1 (0.05)
Use value 68.1 8.4 9 344 50
Non-use value 35.5 4.8 12 78 13
Use and non-use values 63.8 12.9 1 267 40
A3.0 (0.003)c
Country
USA and Canada 70.8 7.8 3 344 80
Europe 32.8 8.4 1 177 23
Payment mode 27.4 (0.001)
Income tax (1) 121.3 18.1 2 267 22
Entrance fee/private fund (2) 28.6 5.7 1 137 28
Product prices (3) 47.8 8.9 3 174 22
Combination of (1) and (3) 42.8 6.3 9 117 26
Trip expenditures 102.9 6.8 89 112 3
Not speci®ed 237.5 106.2 131 344 2
Elicitation format 10.1 (0.01)
Open-ended 37.4 6.5 1 137 35
Dichotomous choice 91.2 17.1 3 344 29
Iterative bidding 78.5 14.9 9 244 20
Payment card 47.1 8.4 10 174 19
a
Number of observations does not sum up to 103 in all cases as a result of missing values
b
Outcome of the non-parametric Kruskal±Wallis test statistic which has approximately a Chi-squared distribution under the null
hypothesis of equal average WTP in all groups
c
Outcome of the non-parametric Mann±Whitney test statistic for two independent samples which has approximately a standard
normal distribution under the same null hypothesis
WTP is reduced, on average, by 41% (ceteris paribus)
The estimated models account for approximately 37% of
when using studies in a value transfer exercise which are
the observed variability in the mean WTP values found in
based on an open-ended WTP question.
individual studies.
The basic model also indicates that study location has
For the ®xed part of the basic and extended model, the
a signi®cant impact on average WTP. The dummy
estimated coef®cients in the semi-log function represent
variable has a value of 1 if the research took place in North
the constant proportional rate of change in the dependent
America and zero if in Europe. As shown before, average
variable per unit change in the independent variables
WTP is substantially higher in North America than in
(Johnston 1984). Hence, the coef®cient estimated for the
Europe. The parameter estimates for the four main wet-
dummy variable `Payment vehicle' in the basic model re-
land functions are particularly interesting. These functions
¯ects, ceteris paribus, an almost twice as higher average
are found to have a statistically signi®cant role in
WTP for an increase in income tax than for any other
explaining variance in average WTP. The size of the
payment vehicle. Compared with other elicitation formats,
53
Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
Table 3
Generalised least squares results for the basic and extended model. As a result of missing values for explanatory variables,
number of observations is reduced from 103 to 92
Parameter Parameter de®nition Basic model Extended model
Estimate Standard error Estimate Standard error
Fixed effects
Constant Intercept 3.356*** 0.100 3.311*** 0.247
Payment vehicle Dummy: 1 = income tax; 0 = other 1.880*** 0.265 1.576*** 0.362
Elicitation format Dummy: 1 = open-ended; 0 = other A0.411** 0.130 A0.376* 0.183
Country Dummy: 1 = North America; 0 = other 1.861*** 0.217 1.629*** 0.363
Response rate (1) Dummy: 1 = 30±50%; 0 = other A2.253*** 0.326 A1.722*** 0.451
Response rate (2) Dummy: 1 = >50%; 0 = other A1.904*** 0.333 A1.461** 0.450
Flood control Dummy: 1 = ¯ood control; 0 = other 1.477*** 0.240 1.134* 0.456
Water generation Dummy: 1 = water generation; 0 = other 0.691* 0.342 0.441 0.479
0.545 0.282 0.659* 0.327
Water quality Dummy: 1 = water quality; 0 = other
Random effects
Between studies
r2 Variance ± ± 0.160* 0.071
constant
Between average WTP
r2 Variance 0.059* 0.029 0.045 0.028
constant
rpayment vehicle, constant Covariance 0.020 0.043 0.001 0.036
rcountry, constant Covariance 0.689** 0.222 0.351** 0.129
rcountry, payment vehicle Covariance A0.707** 0.226 A0.345** 0.134
r¯ood control, constant Covariance A0.013 0.050 0.027 0.060
A0.266
rwater generation, constant Covariance A0.637** 0.227 0.153
A0.424
rwater quality, constant Covariance 0.231 A0.188 0.135
LLunconstrained A83.907 A81.874
Pseudo R-squared 0.365 0.380
n 92 92
Signi®cant at 0.10; * signi®cant at 0.05; ** signi®cant at 0.01; *** signi®cant at 0.001
effects is as follows: the variance of the constant in the
estimated parameters indicates that average WTP is, as
basic and extended model is the variance associated with
before, highest for ¯ood control, but this time, whilst
the baseline case, i.e. where the value of all the explanatory
controlling for other explanatory factors, followed by
variables is zero. The variance of, for example, North
water generation and water quality and lowest for the
American studies in the basic model is
wetland function biodiversity supply. The latter is used as
(0.059 + 2 ´ 0.689) 1.437. Hence, North American
the baseline category in the regression analysis in order to
studies are more variable than European ones. In this way
avoid multicolinearity. The positive parameter estimates
heteroscedasticity can be modelled in the basic model.
for the three other wetland functions indicate that these
Another example is North American studies using income
functions generate higher values than the baseline function
taxation as a payment vehicle:
biodiversity supply.
This suggests the prominence of use over non-use moti-
r2
vations underpinning stated WTP amounts. The distinc- countryY payment vehicle 0X059 2 Â 0X689
tion between use and/or non-use values does not have a
2 Â 0X020 À 2 Â 0X707 0X023X
signi®cant impact on average WTP, probably because the
corresponding variance is already accounted for by the
distinction between wetland functions. Also, relative wet- The last term in this equation is the covariance between
land size is statistically not signi®cant. Higher response payment vehicle and country. So, wetland CV studies
rates, a rough indicator of better overall study quality, based on income taxation in the US appear to have a
appear to result in signi®cantly lower average WTP than particularly low variance.
low response rates. A practical explanation may be that Accounting for study-level effects in the basic model
low response rates are sometimes biased towards includ- signi®cantly reduces the sample variance or standard
ing a relatively large number of sample respondents with a deviation of average WTP in the extended model
(v2 4X06; P < 0.05). The extended, multilevel model ac-
greater interest than average in environmental protection 1
and corresponding WTP. counting for the random effects between studies hence
The model's random effects can be used to (1) model provides a signi®cant improvement over conventional
heteroscedasticity and (2) investigate the suitability of meta-analysis by allowing for the hierarchical structure of
using speci®c CV results in a value transfer exercise. This data implicit in clustering of multiple results from
will be explained below. The interpretation of the random single studies. As expected, having explained some of the
54 Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
with the resource. It is people's perception of a good's
variance in the model by study-level effects, the random
characteristics or functions that in¯uence their attitudes
effects between log(WTP) amounts have decreased, except
and behaviour, not necessarily the good's `objective'
for the covariance between ¯ood control and the intercept.
characteristics (e.g. see Adamowicz et al. 1997). Finally,
However, in the extended model the ®xed effects have
accounting for intra-study variability, the statistical anal-
slightly decreased, while the signi®cance level of half of the
ysis produces slightly different results regarding the sig-
explanatory variables is lower. Only the signi®cance of the
ni®cance and size of the effect of the main functions on the
wetland function water quality has increased. The wetland
WTP values. The low number of observations also has to
function water generation has become statistically insig-
be taken into consideration.
ni®cant and has switched place with the function water
On the other hand, the study progressed meta-analytical
quality supply in the order of size, suggesting that the
research in environmental economics by providing a
results must be viewed with some caution when properly
statistical multilevel model which accounts for the clus-
accounting for intra-study variability.
tering of results from the same studies, for example as a
If low variance is considered an estimate of quality in the
result of identical survey design or sample population,
sense that study results are better suited for comparison
and owing to the fact that results from some studies may
and hence can be more readily put together in a value
be more variable than others. In the GLS models used,
transfer exercise, then it can be concluded that, on the
the variance and covariance estimators not only enabled
whole, studies using income taxation as a payment vehicle
us to model heteroscedasticity, but also provided im-
are better suited than other payment vehicles, and that
portant background information for environmental value
studies valuing wetland biodiversity tend to be less
transfer.
variable than studies valuing wetlands in their capacity
Finally, although considerable effort has been put into
of generating water or maintaining water quality.
specifying the characteristics of the environmental func-
tions and correspondingly the environmental goods and
services involved, other important aspects that may have
Discussion and conclusions helped to explain differences in valuation outcomes re-
main unde®ned. This is a common problem in meta-an-
In this paper, estimates for socio-economic use and non- alyses in the ®eld of environmental valuation as a result of
use values attached to different hydro-ecological and bio- insuf®cient and inadequate information provided in pub-
lished valuation studies. Relevant information about the
geochemical wetland functions were compared and syn-
samples' socio-economic values is missing in many stud-
thesised in a meta-analysis of wetland CV studies. The
meta-analysis provides insights into the factors that have to ies, let alone respondents' socio-psychological and cultural
be considered when attempting to transfer environmental characteristics.
values on the basis of CV studies. A statistically signi®cant In meta-analysis, inferences are made on the basis of
breakdown of WTP values for four main wetland functions information on global statistics, such as the mean and
has been presented. Although single ecosystem character- standard deviations of parameter estimates. These may or
may not describe individual behaviour adequately. In or-
istics or functions are given meaning and value within
der to overcome this potential problem and to increase the
existing ecosystem structures, the distinction between
functions is essential for a valid transfer of the economic study's validity and reliability, a logical next step would be
to gather more information about sample population
values generated within an ecosystem's primary self-or-
characteristics by complementing the analysis with the
ganising capacity. It reduces the risk of double counting
underlying individual responses. This will provide an
when attempting to assess a natural resource's total eco-
important test of the appropriateness of meta-analysis as
nomic value on the basis of different valuation studies.
an instrument to synthesise CV outcomes for the purpose
From an anthropocentric point of view, the size of the
of value transfer.
estimated parameters in the estimated basic model is as
expected. Average WTP is highest for ¯ood control, be-
cause of the possible risks to life and livelihood as a result Acknowledgements The Centre for Social and Economic
of ¯ooding and the capacity of wetlands to reduce this risk, Research on the Global Environment (CSERGE) is a designated
research centre of the UK Economic and Social Research Council
followed by water supply and water quality and ®nally the (ESRC). This research paper is part of the European Commission
provision and maintenance of biodiversity. However, these Directorate General XII-funded project `Ecological±Economic
results have to be handled with care for a number of Analysis of Wetlands' (ECOWET), contract no. ENV4-CT96±
reasons. 0273, co-ordinated by CSERGE.
Ecosystem structures and processes provide a heteroge-
neous complex of highly interrelated socio-economic
functions. The analysis presents a simple and arbitrary
References
breakdown of these functions into independent compo-
nents. The distinction between four main wetland func- ADAMOWICZ W, SWAIT J, BOXALL P, LOUVIERE J, WILLIAMS M (1997)
DAMOWICZ WAIT OXALL OUVIERE ILLIAMS
tions does also not necessarily correspond with people's Perceptions versus objective measures of environmental quality
perception of the various functions wetlands perform, in combined revealed and stated preference models of envi-
which depends upon their own knowledge and experience ronmental valuation. J Environ Econ Manage 32:65±84
55
Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
DESVOUSGES WH, NAUGHTON MC, PARSONS GR (1992) Bene®t
ARROW K, SOLOW R, PORTNEY PR, LEAMER EE, RADNER R, SCHUMAN H
RROW OLOW ORTNEY EAMER ADNER CHUMAN ESVOUSGES AUGHTON ARSONS
(1993) Report of the NOAA Panel on Contingent Valuation. transfer: conceptual problems in estimating water quality ben-
Federal Register (15 January) 58(10):4601±4614 e®ts using existing studies. Water Resour Res 28(3):675±683
BARBIER EB (1993) Sustainable use of wetlands. Valuing tropical DUGAN PJ (ed) (1990) Wetland conservation. A review of current
ARBIER UGAN
wetland bene®ts: economic methodologies and applications. issues and required action. World Conservation Union, Gland,
Geogr J 159(1):22±32 Switzerland
BATEMAN IJ, LANGFORD IH, TURNER RK, WILLIS KG, GARROD GD FARBER S (1988) The value of coastal wetlands for recreation: an
ATEMAN ANGFORD URNER ILLIS ARROD ARBER
(1995) Elicitation and truncation effects in contingent valuation application of travel cost and contingent valuation methodolo-
studies. Ecol Econ 12(2):161±179 gies. J Environ Manage 26:299±312
BATEMAN IJ, MUNRO A, RHODES B, STARMER C, SUGDEN R (1997) Does GARROD GD, WILLIS KG (1996) Estimating the bene®ts of envi-
ATEMAN UNRO HODES TARMER UGDEN ARROD ILLIS
part±whole bias exist? An experimental investigation. Econ J ronmental enhancement: a case study of the River Darent. J
107(441):322±332 Environ Plann Manage 39(2):189±203
BERGH van den JCJM, BUTTON KJ, NIJKAMP P, PEPPING GC GLASS GV, MCGAW B, SMITH ML (1981) Meta-analysis in social
ERGH UTTON IJKAMP EPPING LASS C AW MITH
(1997) Meta-analysis in environmental economics. Kluwer, research. Sage Publications, Beverly Hills
Dordrecht GREEN CH, TUNSTALL SM (1991) The evaluation of river water
REEN UNSTALL
BERGSTROM JC, STOLL JR, TITRE JP, WRIGHT VL (1990) Economic quality improvements by the contingent valuation method. Appl
ERGSTROM TOLL ITRE RIGHT
value of wetlands-based recreation. Ecol Econ 2:129±147 Econ 23:1135±1146
BISHOP RC, BOYLE KJ (1985) The economic value of Illinois Beach GREENLEY DA, WALSH RG, YOUNG RA (1981) Option value: empir-
ISHOP OYLE REENLEY ALSH OUNG
State Nature Reserve. Consultancy report. HBRS, Madison, ical evidence from a case study of recreation and water quality.
Wisconsin Q J Econ 96(4):657±673
È
BISHOP RC, BOYLE KJ, WELSH MP (1987) Toward total economic GREN I-M, SoDERQVIST T (1994) Economic valuation of wetlands:
ISHOP OYLE ELSH REN
valuation of Great Lakes ®shery resources. Trans Am Fish Soc a survey. Beijer Discussion Paper Series no. 54, Beijer Institute,
116:339±345 Stockholm
BLAMEY RK (1995) Citizens, consumers and contingent valuation: HAMILTON LC (1993) Statistics with Stata 3. Duxbury Press,
LAMEY AMILTON
an investigation into respondent behaviour. PhD Diss, Austra- Belmont, California
lian National University, Canberra IMF (1996) International ®nancial statistics yearbook. Interna-
BROUWER R, LANGFORD IH, BATEMAN IJ, CROWARDS TC, TURNER RK tional Monetary Fund, Washington
ROUWER ANGFORD ATEMAN ROWARDS URNER
(1997) A meta-analysis of wetland contingent valuation studies. JOHNSTON J (1984) Econometric methods, 3rd edn. McGraw-Hill,
OHNSTON
Global Environmental Change Working Paper 97±20, Centre for Maidenhead
Social and Economic Research on the Global Environment JORDAN JL, ELNAGHEEB AH (1993) Willingness to pay for improve-
ORDAN LNAGHEEB
(CSERGE), University of East Anglia, Norwich, and University ments in drinking water quality. Water Resour Res 29(2):237±
College, London 245
BROUWER R, SLANGEN LHG (1998) Contingent valuation of the KAORU Y (1993) Differentiating use and nonuse values for coastal
ROUWER LANGEN AORU
public bene®ts of agricultural wildlife management: the case of pond water quality improvements. Environ Resour Econ 3:487±
Dutch peat meadow land. Eur Rev Agric Econ 25:53±72 494
BROUWER R (1998) Future research priorities for valid and reliable KOSZ M (1996) Valuing riverside wetlands: the case of the
ROUWER OSZ
environmental value transfer. Global Environmental Change ``Donau-Auen'' National Park. Ecol Econ 16:109±127
Working Paper 98±28, Centre for Social and Economic Research LANGFORD IH (1994) Using a generalized linear mixed model to
ANGFORD
on the Global Environment (CSERGE), University of East An- analyze dichotomous choice contingent valuation data. Land
glia, Norwich, and University College, London Econ 70(4):507±514
CARSON RT, MITCHELL RC (1993) The value of clean water: the LANGFORD IH, BATEMAN IJ, JONES AP, LANGFORD HD, GEORGIOU S
ARSON ITCHELL ANGFORD ATEMAN ONES ANGFORD EORGIOU
public's willingness to pay for boatable, ®shable, and swimm- (1998) Improved estimation of willingness to pay in dichoto-
able quality water. Water Resour Res 29(7):2445±2454 mous choice contingent valuation studies. Land Econ 74(1):65±
CARSON RT, FLORES NE, MARTIN KM, WRIGHT JL (1996) Contingent 75
ARSON LORES ARTIN RIGHT
valuation and revealed preference methodologies: comparing LANT CL, ROBERTS RS (1990) Greenbelts in the cornbelt: riparian
ANT OBERTS
the estimates for quasi-public goods. Land Econ 72:80±99 wetlands, intrinsic values, and market failure. Environ Plan A
COOPER J, LOOMIS JB (1991) Economic value of wildlife resources in 22:1375±1388
OOPER OOMIS
the San Joaquin Valley: hunting and viewing values. In: Dinar A, LOOMIS JB (1987) Balancing public trust resources of Mono Lake
OOMIS
ZILBERMAN D (eds) The economics and management of water and Los Angeles' water right: an economic approach. Water
ILBERMAN
drainage. Kluwer, Dordrecht Resour Res 23(8):1449±1456
COSTANZA R, d'ARGE R, de GROOT R, FARBER S, GRASSO M, HANNON B, LOOMIS JB, HANEMANN M, KANNINEN B, WEGGE T (1991) Willingness
OSTANZA ARGE ROOT ARBER RASSO ANNON OOMIS ANEMANN ANNINEN EGGE
LIMBURG K, NAEEM S, O'NEILL RV, PARUELO J, RASKIN RG, SUTTON P, to pay to protect wetlands and reduce wildlife contamination
IMBURG AEEM EILL ARUELO ASKIN UTTON
van den BELT M (1997) The value of the world's ecosystem from agricultural drainage. In: Dinar A, Zilberman D (eds) The
ELT
services and natural capital. Nature 387:253±260 economics and management of water drainage. Kluwer, Do-
CROWARDS TC, TURNER RK (1996) FAEWE sub-project report: rdrecht
ROWARDS URNER
economic valuation of wetlands. Centre for Social and Economic MADDALA GS (1983) Limited-dependent and qualitative variables
ADDALA
Research on the Global Environment (CSERGE), University of in econometrics. Cambridge University Press, Cambridge
East Anglia, Norwich, and University College, London MCFADDEN D (1994) Contingent valuation and social choice. Am J
C ADDEN
CUMMINGS RG, GANDERTON PT, MCGUCKIN T (1994) Substitution Agric Econ 76:689±708
UMMINGS ANDERTON C UCKIN
effects in cvm values. Am J Agric Econ 76:205±214 MITCHELL RC, CARSON RT (1989) Using surveys to value public
ITCHELL ARSON
DESVOUSGES WH, SMITH VK, MCGIVNEY MP (1983) A comparison of goods: the contingent valuation method. Resources for the Fu-
ESVOUSGES MITH C IVNEY
alternative approaches for estimation of recreational and related ture, Washington, DC
bene®ts of water quality improvement. Report to the US Envi- OLSEN D, RICHARDS J, SCOTT RD (1991) Existence and sport values
LSEN ICHARDS COTT
ronmental Protection Agency, Washington, DC for doubling the size of Columbia River Basin salmon and
DESVOUSGES WH, SMITH VK, FISHER A (1987) Option price estimates steelhead runs. Rivers 2(1):44±56
ESVOUSGES MITH ISHER
for water quality improvements: a contingent valuation study PEARCE DW, TURNER RK (1990) Economics of natural resources
EARCE URNER
for the Monongahela River. J Environ Econ Manage 14:248±267 and the environment. Harvester Wheatsheaf, Hemel Hempstead
56 Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag
Original article
PHILLIPS WE, HANEY TJ, ADAMOWICZ WL (1993) An economic SUTHERLAND RJ, WALSH RG (1985) Effect of distance on the
HILLIPS ANEY DAMOWICZ UTHERLAND ALSH
analysis of wildlife habitat preservation in Alberta. Can J Agric preservation value of water quality. Land Econ 61(3):281±291
Econ 41:411±418 TOLBA MK, EL-KHOLY OA (1992) The world environment 1972±
OLBA L HOLY
RAMSAR (1975) Convention on wetlands of international impor- 1992. Chapman and Hall on behalf of the UN Environment
AMSAR
tance especially as waterfowl habitat. Ramsar Convention, Iraq Programme, London
RASBASH J, WOODHOUSE G (1995) MLn command reference. TURNER RK (1992) Policy failures in managing wetlands. In: OECD
ASBASH OODHOUSE URNER
Institute of Education, University of London market and government failures in environmental management.
SANDERS LD, WALSH RG, LOOMIS JB (1990) Toward empirical esti- OECD, Paris
ANDERS ALSH OOMIS
mation of the total value of protecting rivers. Water Resour Res WHITEHEAD JC, BLOMQUIST GC (1991) Measuring contingent values
HITEHEAD LOMQUIST
26(7):1345±1357 for wetlands: effects of information about related environmental
SCHULTZ SD, LINDSAY BE (1990) The willingness to pay for goods. Water Resour Res 27(10):2523±2531
CHULTZ INDSAY
groundwater protection. Water Resour Res 26(9):1869±1875 WILLIS KG (1990) Valuing non-market wildlife commodities: an
ILLIS
SILVANDER U (1991) The willingness to pay for angling and ground evaluation and comparison of bene®ts and costs. Appl Econ
ILVANDER
water in Sweden. PhD Diss, Swedish University of Agricultural 22:13±30
Sciences, Uppsala WILLIS KG, GARROD GD, SAUNDERS CM (1995) Bene®ts of environ-
ILLIS ARROD AUNDERS
SPANINKS FA (1993) Een Schatting van de Sociale Baten van Be- mentally sensitive area policy in England: a contingent valuation
PANINKS
heersovereenkomsten met behulp van de Contingent Valuation assessment. J Environ Manage 44:105±125
Methode. MSc Diss, Department of Agricultural Economics and WOLF FM (1986) Meta-analysis. Sage Publications, Beverly Hills,
OLF
Policy, Wageningen Agricultural University California
SPANINKS FA, KUIK OJ, HOOGEVEEN JGM (1996) Willingness to pay of
PANINKS UIK OOGEVEEN
Dutch households for a natural Wadden Sea. An application of
the contingent valuation method. Report E-96/6, Institute for
Environmental Studies, Amsterdam
57
Regional Environmental Change 1 (1) November 1999 á ã Springer-Verlag