The economic value of wetland services: a meta-analysis
Ecological Economics 37 (2001) 257– 270
www.elsevier.com/locate/ecolecon
ANALYSIS
The economic value of wetland services: a meta-analysis
Richard T. Woodward *, Yong-Suhk Wui
Department of Agricultural Economics, Texas A&M Uni6ersity, 2l24 TAMU, College Station, TX 77843-2124, USA
Received 30 March 2000; received in revised form 15 November 2000; accepted 16 November 2000
Abstract
The number of studies quantify the value of wetlands and the services provided by these ecosystems is rapidly
expanding. The time is ripe for an assessment of what has been learned from this literature. Using results from 39
studies, we evaluate the relative value of different wetland services, the sources of bias in wetland valuation and the
returns to scale exhibited in wetland values. While some general trends are beginning to emerge, the prediction of a
wetland’s value based on previous studies remains highly uncertain and the need for site-specific valuation efforts
remains large. © 2001 Elsevier Science B.V. All rights reserved.
growing need to quantify the value of wetland
1. Introduction
services.
The services provided by wetlands include habi-
The valuation of wetlands’ ecological services is
tat for species, protection against floods, water
a relatively recent phenomenon. Historically, wet-
purification, amenities and recreational opportu-
lands were viewed as a waste of valuable land that
nities. Because these services typically have no
could only be ‘improved’ through drainage and
market price, a measure of their values can only
destruction of the wetland (Mitsch and Gosselink,
be obtained through non-market valuation tech-
1986). Today, while there is now widespread niques. Many wetland valuation studies have been
recognition that wetlands provide valuable eco- conducted and the range of the estimates is re-
logical services, there remain substantial debates markable. A recent review by Heimlich et al.
over whether particular areas are in their highest (1998) lists 33 studies over the last 26 years with
economic use as wetlands, and to what extent per acre values ranging from US$0.06 to
public and private resources should be used for US$22050. Even within the same study looking at
their protection and restoration. Hence, there is a a single ecosystem function, Batie and Wilson
(1978) find values per acre that differ by two
orders of magnitude from one site to another.
* Corresponding author. Tel.: + 1-979-8455864; fax: + 1-
The purpose of this paper is to assess whether
979-8454261.
any systematic trends can be distilled from the
E-mail address: r-woodward@tamu.edu (R.T. Woodward).
0921-8009/01/$ - see front matter © 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 1 - 8 0 0 9 ( 0 0 ) 0 0 2 7 6 - 7
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270
258
Estimates of the value of a wetland can also
breadth of wetland valuation studies conducted to
influence site-specific valuation efforts in two
date, and to shed light on what factors determine
ways. First, they might provide Bayesian priors
a wetland’s value. We maintain an assumption
that might be formally incorporated into the valu-
that there exists an unobserved valuation function
ation exercise. Secondly, they may give re-
that determines a wetland’s value given its physi-
searchers a sense of where the values at stake are
cal, economic and geographic characteristics. Af-
likely to be of greatest social importance and
ter reviewing 46 studies, data from 39 wetland
might, therefore, influence where detailed studies
valuation studies were identified that had suffi-
are carried out.
cient commonalties to allow inter-study compari-
The paper is organized as follows. In the next
sons. We used two techniques to learn about the
section we provide a brief overview of the eco-
valuation function, both of which can be broadly
nomics of wetland valuation. We survey the eco-
described as meta-analysis since many studies are
logical functions and economic services provided
used to identify general relationships. The first
by these areas, the basis for their valuation and
method that we employ uses bivariate graphical
the techniques that are used to place an economic
and standard techniques. This gives us both an
value on wetlands. Section 3 provides a brief
indication of the extent to which particular char-
summary of meta-analysis as a tool. In Sections 5
acteristics influence wetland values while also por-
and 6 we explore the trends in the data, identify-
traying the full distribution of the data. The
ing the sources of variability in wetland values.
second technique is more standard, using a multi-
We conclude by reflecting on the implications of
variate regression of wetland values on the char-
our analysis both for our understanding of wet-
acteristics of both the wetlands and the studies.
land values and for future research.
Together, these two techniques provide a richer
basis from which we can draw lessons on the
factors determining wetland value.
2. The value of wetland functions
There are numerous reasons why understanding
the value of wetland services might be useful. The
2.1. Wetland functions and ser6ices
most obvious is that if the value of these services
were known, benefits transfer efforts could be While an inclusive definition of wetlands is
substantially improved. As Deck and Chestnut difficult to state, they are generally characterized
(1993) point out, benefits transfer may play a as being moist during an extended period each
variety of different roles, ranging from an attempt year with soils, plants and animals that are dis-
to place a precise value on a particular resource to tinct from their aquatic and terrestrial neighbors.
providing information that feeds into the process These transition areas are highly diverse, ranging
of building support for projects already imple- from coastal mangroves that are inundated with
mented. While benefits transfer is rarely suitable water most of the year to areas that are moist for
for the former case, it might often be appropriate only a few months during the year. Partly because
for the latter. Another form of benefits transfer is they share features of both terrestrial and aquatic
the use of estimated values to predict the aggre- systems, wetlands are remarkably productive.
gate value of similar systems nationally or In assessing the value of wetlands, it is useful to
globally. Such estimates can be useful in setting distinguish the systems’ ecological functions from
national priorities or the evaluation of policies the associated services that are directly valued by
with impacts that are national or global. Costanza humans (Costanza et al., 1997). Larson et al.
et al. (1997), for example, placed a value on the (1989) list 17 services and functions provided by
entire globe’s ecosystems. While such expansive the world’s ecosystems (Table 1). In our data set,
efforts may be overambitious, the aggregate num- the services were grouped into ten categories as
bers do help to get the attention of policy makers indicated in the table. The measurement of the
and the public. value of these services varies substantially both in
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270 259
Table 1
Wetland functions, the associated economically valuable goods and services and the names of variables that capture the presence of
these in the dataa
Function Economically valuable good(s) and/or Technique(s) typically used to quantiiy
service(s) (variable names) the value of the service(s)
Recharge of ground water Increased water quantity (quantity) Net factor income or replacement cost
Discharge of ground water Increased productivity of downstream Net factor income, replacement cost or
fisheries (com.fish) travel cost
Water quality control Reduced costs of water purification Net factor income or replacement cost
(quality)
Retention, removal and transformation Reduced costs of water purification Net factor income or replacement cost
of nutrients (quality)
Habitat for aquatic species Improvements in commercial and/or Net factor income, replacement cost,
recreational fisheries either on or offsite travel cost or contingent valuation
(com.fish and rec.fish). Nonuse
appreciation of the species (habitat)
Habitat for terrestrial and avian species Recreational observation and hunting of Travel cost or contingent valuation
wildlife (birdwatch & birdhunt). Nonuse
appreciation of the species (habitat)
Biomass production and export (both Production of valuable food and fiber Net factor income
plant and animal) for harvest (birdhunt & com. fish)
Flood control and storm buffering Reduced damage due to flooding and Net factor income or replacement cost
severe storms (flood)
Stabilization of sediment Erosion reduction (storm) Net factor income or replacement cost
Overall environment Amenity values provided by proximity Hedonic pricing
to the environment (amenity)
a
The first two columns are adapted from Larson et al. (1989).
the methods that are used and the economic tice, it is often assumed that demand is perfectly
theory that underlies the valuation exercise. The elastic so that the impact of the wetland on con-
most common methods for measuring the eco- sumer surplus can be ignored. Other times the
nomic value of these services are also presented in producer surplus that is generated by a wetland is
the table. Though not indicated in the table, the estimated using the replacement cost (RC)
contingent valuation method can, in principle, be method. This approach values the wetland’s ser-
used to measure the value of all these services. vice based on the price of the cheapest alternative
way of obtaining that service. For example, the
2.2. The 6aluation of wetland ser6ices value of a wetland in the treatment of wastewater
might be estimated using the cost of chemical or
The economic value of resources such as wet- mechanical alternatives. As noted by Anderson
lands is equal to the benefits (net of costs) that and Rockel (1991), the replacement cost method
these systems provide to humans (Freeman, 1993). is actually an upper bound on the true value since
Four methods are listed in Table 1 as commonly the producer may not choose to actually use that
used to place a value on wetlands. The net factor alternative considered.
income (NFI) method is most appropriate when Non-market values can be measured using
the wetland provides a service that leads to an travel cost (TC), contingent valuation (CV), or
increase in producer surplus. In the NFI method hedonic pricing (HP) methods. There is, however,
the physical relationship between wetland area substantial variability within each of these ap-
and the economic activity is estimated. It is then proaches, much more than can be discussed here
possible to identify the increase in producer sur- (Freeman, 1993), and the literature is filled with
plus associated with the wetland’s area. In prac- reasons why results might be biased or otherwise
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270
260
deficient. While in principle each method can give the first to use meta-analysis in the context of
a correct estimate of economic value, it is easy to nonmarket valuation, looking at values placed on
misuse the method and obtain results that may outdoor recreation. Since then, meta-analysis has
have little relation to the true value. been used to study air pollution, recreational
Some wetland values are obtained using meth- fishing, visibility, health risks, endangered species
and wetlands.1
ods that do not estimate economic surplus and,
therefore, lack a foundation in standard economic The basic approach used in most valuation
theory. Of the studies identified in our review of meta-analyses is the same. A set of studies is
the literature, two values were estimated using selected yielding a number of values that become
energy analysis in which the value is based on the the dependent variable. The independent variables
gross primary production of the ecosystem, and are the characteristics of each study and study
five values were obtained using the market value site. If a single study reports numerous values,
of the products extracted. These methods have then several data points are obtained. Meta-anal-
been strongly criticized (Anderson and Rockel, ysis allows the evaluation of the effect of changes
1991). Energy analysis equates the energy embod- in the underlying environmental attribute on
ied in a wetland’s biota with the energy purchased value. Such analysis is usually not possible in the
in fossil fuels. Since the correlation between en- context of a single study since most such at-
ergy content and consumer preferences is quite tributes are held constant.
weak, the technique is a poor predictor of eco- A good example of the benefits of meta-analysis
nomic value. The market value technique is also is seen in Smith and Osborne’s (1996) analysis of
flawed since it cannot capture consumer surplus the value of improvements in visibility. Since visi-
and can lead to over-estimate producer surplus if bility varies continuously from zero to one, the
cost of extracting the valued products is not authors are able to estimate the marginal benefit
subtracted. of improvements in visibility. One of the most
significant limitations of meta-analysis, however,
is the lack of comparability across studies (van
3. Meta-analysis as a tool in understanding den Bergh and Button, 1997). Characteristics of
non-market valuation the resource being valued are often presented in
such diverse fashion that the best that the analyst
First used by psychologists (Glass, 1976; can do is to use a binary variable to indicate
Schmidt and Hunter, 1977), meta-analysis has whether an attribute is reflected in each value.
proved to be a useful tool for synthesizing the Boyle et al. (1994), for example, have no data on
results of numerous studies. The method has re- the level of cancer risk in each of their eight
cently gained attention in economics as a way to studies, only an indication of whether such a risk
appreciate numerous studies that have placed eco- was mentioned in the study. Similarly, in this
nomic values on environmental goods and ser-
study wetland services are captured using qualita-
vices (see Brouwer (2000) for a review). The
tive variables.
central advantage of meta-analysis is that it pro-
Brouwer et al. (1997) is, to our knowledge, the
vides a rigorous statistical synthesis of the litera-
only other attempt to carry out meta-analysis of
ture that cannot be achieved using more
wetland valuation studies. In their work, only
qualitative analysis.
contingent valuation studies are considered. This
There are two main types of meta-analyses:
narrow focus allowed the authors to develop a
those that use the actual data from multiple stud-
ies, and those that use the results of multiple
1
The analysis of wetlands by Brouwer et al. (1997) looked
studies. It is the later method that has been ap-
at WTP estimates from CV studies and used a more expansive
plied to interpret valuation studies. Brouwer lists interpretation of ‘wetlands’ than we retain in this paper.
ten studies that have used meta-analysis to study Hence, their results are not directly comparable to the results
valuation efforts. Smith and Kaoru (1990) were here.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270 261
rich set of variables characterizing the qualities of ues: variation due to differing characteristics of
the studies. However, this is also a limitation in the wetlands, i.e. along the function; and variation
that it eliminates any variability associated with due to error in the estimation of the true value,
valuation method and reduces the variability in i.e. deviations from the function.
services that can be considered. In the next section Of course, the wetlands that have been valued
we discuss how we attempted to overcome these were almost certainly not chosen at random from
limitations in our data. the total population. First there is the problem of
selection bias. It seems likely that wetlands that
are considered valuable a priori are much more
likely to be studied and valued. This need not lead
4. The wetland valuation data
to errors in our estimation of the valuation func-
tion if all important variables are accurately mea-
After a lengthy review of the literature, we
sured, but given the limitations in the available
identified 39 studies that contained sufficient data
data, the likelihood of such bias should be taken
to allow inter-study comparisons. Many other
into account in benefits transfer exercises or any
studies were identified but could not be used. The
other attempt to extrapolate estimated values.
values are taken from published reports, ‘gray’
Similarly, the fact that many of the studies have
literature, and theses.2 Because of our desire to
been filtered by the peer review process might
synthesize wetland values from all different ser-
have excluded some estimates. Good estimates
vices, we use annual value per acre in 1990 US
that are either not statistically different from zero
dollars. This distinguishes our work from other
or are much higher than anticipated may not be
meta-analyses which typically use willingness to
published.
pay (WTP) per person (e.g. Brouwer et al., 1997).
The values in our data are also not independent
WTP per person is not applicable here because
draws. Numerous studies generate multiple mea-
some methods (e.g. NFI) do not lead to a WTP
sures of wetland value, so that, as pointed out by
per person measure. On the other hand, if WTP
Smith and Kaoru (1990), the data have panel
per person is available, then value per acre can be
characteristics. Furthermore, researchers who
calculated with knowledge of the relevant popula-
work closely together are likely to share practices
tion and the wetland’s size. When capitalized val-
that differ in important ways from others. Finally,
ues were reported, they were annualized assuming
since there has no doubt been learning over time,
constant value per year and using discount factors
both in terms of methodology and the values that
provided in the studies or a 6% rate in the two
are reasonable, the data also probably suffer from
studies that did not state a discount rate.
some autocorrelation.
In our analysis we assume that a wetland’s
value is a function of the system’s ecological
characteristics and its socio-economic environ-
5. Bivariate meta-analysis
ment. Each wetland in our data is interpreted as a
(not necessarily random) draw from the popula-
Using the available data, we now evaluate the
tion of all wetlands. We assume that there exists a
sources of variation in estimates of wetland value.
true public WTP at a given moment for a particu-
Two complementary techniques are used. In this
lar wetland. While this true WTP cannot be ob-
section we explore some of the relationships in the
served directly, it can be estimated using the
data using graphical presentation and bivariate
methods discussed above. Seen in this way, there
statistics. The advantage of this analysis is that it
are two sources of variability in the wetland val-
allows us to present the full data set graphically,
making possible a richer appreciation of the data.
2
The complete data set, including a description of each
However, the bivariate analysis ignores interac-
study and an explanation of the interpretations of the data
tions between explanatory variables. Hence, a sec-
that were made is available from the authors on request or via
ond and more standard technique is also used,
the internet at http://ageco.tamu.edu/faculty/woodward/.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270
262
that of estimating a valuation function using mul- difference is found it is sensible to retain the ‘bad’
tivariate regression techniques. studies because they contain other variation in
methods (such as different samples and locations)
5.1. Variation due to measurement error or bias that, by their inclusion, will help answer many
other questions surrounding the problem area.’’
As noted above, there are two types of varia- Following Cooper’s advice, some meta-analyses
tion with which we are concerned, deviations include objective indicators of study quality such
from the valuation function due to bias or errors as response rate or study format (Brouwer et al.,
in estimation, and variations along the valuation 1997; Loomis and White, 1996). Given the diver-
function attributable to different wetland charac- sity of studies considered in our analysis, no
teristics. We begin by looking at sources of sys-
standard objective indicator of quality was avail-
tematic error because of study weaknesses and
able; only a subjective assessments of study qual-
bias because of the valuation method used.
ity could be used. Each study was ranked on a
One might expect study quality to affect esti-
scale of 1–3 in four categories: the apparent
mates of wetland value. Attention to this issue is
quality of the data, the theoretical consistency of
potentially important because there is substantial
the methodology, econometric techniques and
variability in the quality of wetland valuation
statistical certainty. A study was given a rank of 1
studies. While some studies are characterized by
if we felt that this feature of the study made the
sound theoretical foundations and state-of-the-art
results highly questionable. Studies with a 1 in
econometric methods, others are crippled by
any of the quality categories are called ‘weak’ in
faulty logic, poor data or incorrect economic
the figures and econometric analysis below.3 A
analysis.
‘weakness’ in a study should not be interpreted as
The weakness of many wetland valuation ef-
a condemnation since valuation may not have
forts is widely recognized. In their review of wet-
been the authors’ primary objective or data limi-
land valuation studies, Anderson and Rockel
(1991) found only five studies that they deemed tations may have been prohibitive.
credible enough to list in their summary table. On average, the weak and strong studies do not
However, while it may appear obvious that only yield statistically different values. Excluding the
high quality analyses should be used in meta-anal- highest value in our data set, the average of the
ysis, it is also clear that the evaluation of quality weak studies is US$986 per acre versus US$915
is likely to be quite subjective. The problem of for the strong studies.4 When looking at the com-
subjectivity is particularly problematic in wetland plete distribution of these studies however, there
valuation studies because few efforts satisfy the do appear to be some systematic differences two
highest standards of quality, in large part because sets of values. Fig. 1 plots the rank of both the
of data limitations. In CV studies, for example, weak and the strong studies in their respective
strict adherence to the guidelines of the NOAA
panel (Arrow et al., 1993) is often impossible
because of budgetary restrictions. Hence, there is
3
For studies that are also evaluated by Anderson and
a great deal of subjectivity in assessing how good
Rockel (1991) our critique was generally consistent. Some
is good enough.
studies which we ranked as a 2 were questioned, but did not
As Cooper (1989, p. 67) points out, ‘‘The deci- appear to be completely rejected by Anderson and Rockel.
sion to include or exclude studies on an a priori 4
The highest value from the Amacher et al. (1989) study is
basis requires the reviewer to make an overall excluded as it is over 60 times the second largest value. After
excluding this value, the mean of the weak studies is not
judgment of quality that is often too subjective to
significantly different from the mean of the strong studies at
be credible.’’ He argues that it makes more sense
the 10% level. This value, one other value estimated using
to enumerate characteristics of each study and energy analysis and five values estimated using the market
then evaluate whether ‘good’ methods lead to value of the output are excluded from the econometric analysis
different results than ‘bad’ methods. ‘‘When no below.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270 263
Fig. 1. Cumulative distributions of wetland values broken down by study quality.
convergent validity of CV analysis relative to
categories on the vertical axis, against the esti-
other methods (e.g. Carson et al., 1996).
mated values per acre on the horizontal axis. The
Fig. 2 presents the distributions of the values
solid lines in the figure represent cumulative dis-
taken from the four primary methods used to
tribution functions (cdfs) of distributions from
measure wetland values. The means of the values
which the data appear to be drawn. While the
from these methods vary from a low of US$198
distribution of the strong data closely resembles a
for the travel cost method to a high of US$1555
log normal distribution, the weak data seem to be
for the replacement cost method. However, be-
drawn from a uniform distribution. There is also
cause of the substantial variability in the data,
slightly less variance in the strong studies, indicat-
none of the means are statistically different from
ing that the lack of quality may not bias the
each other. Still, some patterns are evident. At
estimated value, but it might have implications for
one extreme, the net-factor input method is a
the accuracy of the predictions. This result is
lower bound on the distribution of values. At the
confirmed in the multivariate analysis below.
other extreme, the distribution of values obtained
Nonetheless, we do not find the kind of dramatic
using CV nearly stochastically dominates the dis-
difference between the two groups that would
tributions of values from the other three methods.
justify discarding the weak studies from the data
These findings do not necessarily indicate biases
set.
in these techniques. Because of the small sample
Another potential reason that the estimated
size we cannot statistically reject the hypothesis
value may deviate from the valuation function
that the distributions are the same. Moreover,
might be bias due to the method that was used in
different methods are used to value different ser-
the study. In principle, if two methods seek to
vices. It may be that CV is used for high-value
estimate consumer surplus from the same wetland
then they should yield similar values. If there is no services while the NFI method is used for low-
systematic difference between two techniques, value services. Hence, the question of whether the
then they are said to satisfy the criterion of con- method itself is a source of bias can only be
vergent validity. Numerous studies have tested the explored using multivariate analysis.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270
264
Fig. 2. Cumulative distributions of wetland values broken down by valuation method.
5.2. Variation in 6alue due to wetland valued. The correlation between estimated value
characteristics per acre and the number of services is only 0.10
and, based on the Spearman rank criterion, the
We now turn an initial analysis of the sources hypothesis of no correlation cannot be rejected at
of variation in the valuation function. Ten vari- the 10% level.
ables were defined indicating whether a particular In addition to being affected by wetland ser-
wetland service was reflected in each study. These vices, one might also expect the value per acre to
are listed in Table 1. Identifying the services be a function of the wetland’s area. In this case
reflected in a study often involves some subjectiv- there is no clear a priori expectation as to the
ity, particularly in CV studies since respondents form that such a relationship might take. Eco-
might be aware of services other than those about nomic intuition would suggest that the marginal
which they had been explicitly asked in the value of each acre would tend to decline. On the
survey.5 other hand, based on ecological principles of
A relatively weak hypothesis would be that functional interdependence, one might expect that
increasing the number of services considered in a larger wetlands would provide a richer and more
valuation exercise would tend to increase a wet- valuable set of services. This relationship is plot-
land’s estimated value. This relationship is pre- ted in Fig. 4. There is no apparent relationship
sented in Fig. 3. While almost two-thirds of the between wetland area and value in the figure and,
studies measured the value of only one wetland once again, the hypothesis of no significant corre-
service, more than 30% of the studies measured lation cannot be rejected at the 10% level.
three or more services. Contrary to our hypothe- Our analysis to this point is quite inconclusive.
sis, there is no noticeable relationship between the There is some evidence that CV studies tend to
value of a wetland and the number of services yield greater values than any other method, but
no visible relationship between value per acre and
either the number of services or the size of the
5
In a few instances authors were contacted to assist us in
wetland. However, while we find the bivariate
obtaining the most accurate interpretation possible.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270 265
Fig. 3. Wetland values and the number services present.
analysis useful, it cannot distinguish how multiple estimate of producer’s surplus, PS; and whether
factors might be interacting to influence wetland the results had been published, published. The
value. In the next section we attempt to tease out variables data0, theory0 and metric0, are dummy
more understanding of the wetland valuation variables set at one if the data, theory or econo-
function using multivariate regression analysis. metrics used in the study were deemed highly
questionable.6
We should recognize that there are certainly
6. Multivariate meta-analysis of wetland values important variables that determine a wetland’s
value that are omitted from our model. Charac-
In this section we estimate a parametric specifi- teristics of the population near a wetland are
cation of the valuation function using the data particularly likely to influence the value placed on
discussed above. After excluding incomplete ob- the area. However, such data could not be iden-
servations and values based on either energy anal- tified in most of the studies; we were unable to
ysis or the market value methods, the 65 include any such variables in our model. While
observations of wetland values were obtained. the absence of these variables no doubt greatly
diminishes the explanatory power of our analysis,
6.1. The estimated model and results it need not bias the estimated coefficients if these
variables are uncorrelated with the included set
The dependent variable in all regressions is the (Kennedy, 1986).
natural log of the value per acre of wetland Our econometric model is based on a main-
converted to 1990 dollars, the mean of which is tained hypothesis that measured wetland value
4.92. In addition to the variables discussed above
representing services, area and study quality, we 6
Since only two studies were deemed weak based on statisti-
included variables indicating date of the study cal significance, and many studies did not report sufficient
(1960= 0), year; whether the wetland was a information to gauge the statistical accuracy of their estimates,
coastal wetland, coastal; whether the value was an this variable was excluded from the econometric analysis.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270
266
Fig. 4. Wetland area and values.
per acre, y, is a function of the services provided, values and the quality of the studies. Model C
xs, the methodology used, xm, the acres of the combines both the characteristics of the sites and
wetland xa, other variables describing the study variables related to how the values were
including year and location, x0, and a constant estimated.
term. The fit was substantially improved by using
the logs of both the per-acre value and the acres.
6.2. Do the study quality or 6aluation method
Hence, the estimated linear model is
affect the 6alue obtained?
%x % %
ln(y) =a+ba ln(xa)+ bs s +bmxm +b0x0 (1)
In our discussion of Fig. 1 we argued that there
where a is the constant term and the b’s are the was little evidence of bias as a result of the quality
estimated coefficients on the respective explana- of the studies, and our regression results largely
tory variables. confirm that conclusion. The coefficients on the
The results of several regressions are presented variables indicating poor quality theory and data
in Table 2. In each case the hypothesis of ho- are both statistically insignificant, as is the coeffi-
moskedasticity was rejected at the 5% significance cient indicating whether the study was published.
level using the BPG test. Accordingly, the stan- However, the variable indicating econometric
dard errors were estimated using White’s (1980) quality was strongly significant in both regressions
correction. Model A presents the estimated coeffi- B and C. Holding all else constant, the values
cients of a model in which it is assumed that the from studies with poor quality econometrics aver-
variability in the values is solely a function of the age 24–50 times greater than those from those
physical characteristics of the wetland systems, with comparatively strong econometric
ignoring any systematic variation due to the way foundations.
that the values were estimated. Model B takes the Study quality also has important consequences
opposite approach, explaining the values based for the confidence we place on predicted values.
solely on the methods used to measure those Using the results from model C evaluated at the
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270 267
Table 2 means of the variables the log of the wetland
Estimated models of the wetland valuation functiona (log of value predicted for a high-quality unpublished
value per acre dependent variable, standard errors in parenthe-
study is 5.68 with a standard error around the
ses)
prediction (|p) of 0.61. If it is assumed that the
study is published, however, |p falls to only 46%
Variable Mean A B C
of the original value. For studies that are weak in
7.945b 6.641b 7.872b
Intercept
the areas of theory or econometrics, |p increases
(1.07) (1.31) (1.74)
by 1.9 or 2.1-fold, respectively. On the other
Year 14.908 −0.052 −0.004 0.016
hand, the impact of study’s data being of poor
(0.03) (0.04) (0.04)
−0.168c −0.286b quality is slight, leading to a 4% decline in |p.7
Ln acres 9.281
(0.10) (0.11)
Study quality is important not so much because it
Coastal 0.431 −0.523 −0.117
might bias results, but because high quality stud-
(0.71) (0.68)
Flood 0.138 −0.358 0.678 ies lead to a much more precise basis for
(1.03) (0.77)
prediction.
1.494c
Quality 0.200 0.737
There is some evidence that the method used
(0.78) (0.75)
Quantity 0.062 0.514 −0.452 has a statistically significant effect on the value
(1.60) (1.54)
obtained. As in Fig. 2, in model B we find that
Rec. Fish 0.354 0.395 0.582
CV tends to dominate other methods as the signs
(0.55) (0.56)
on their coefficients are either negative or statisti-
Com. Fish 0.277 0.669 1.360
(0.79) (1.01) cally insignificant. However, when variables indi-
−1.311b −1.055b
Birdhunt 0.400
cating the wetland services are introduced in
(0.49) (0.52)
model C, the dominance of CV disappears and
1.704b 1.804b
Birdwatch 0.277
(0.52) (0.59) the sign on HP and RC methods becomes signifi-
−3.352b −4.303b
Amenity 0.154
cantly positive. Hence, relative to these methods,
(0.93) (0.95)
CV studies tend to find a lower value per acre and
Habitat 0.308 0.577 0.427
we cannot conclude that this method is biased
(0.56) (0.59)
Storm 0.031 0.310 0.173 relative to the TC or NFI method.8
(2.37) (1.66)
Publish 0.769 −0.669 −0.154
6.3. Do wetlands 6alues exhibit returns to scale?
(0.72) (0.71)
Data0 0.246 0.302 0.000
(0.56) (0.60)
The coefficient on LnAcres is consistently nega-
Theory0 0.215 −1.020 −1.045
tive and statistically significant across the models
(0.84) (0.84)
−4.030b −3.186b
Metric0 0.123 reported in Table 2, indicating significant decreas-
(1.21) (1.22)
ing returns to scale. However, because of the
−2.416b −2.034b −3.140b
PS 0.277
double-log functional form, the scale effect is
(0.83) (0.72) (0.86)
5.043b extremely small for large wetlands. From Eq. (1),
HP 0.031 0.441
(1.02) (1.12) the marginal effect of an increase in the size of a
NFI 0.246 −0.724 0.273
wetland is
(0.82) (0.90)
2.232b
RC 0.277 1.376
(y/(xa = aax (aa − 1)e (a0 + asxs + amxm + a0x0)
(0.86) (0.89) a
−1.196c
TC 0.108 −0.341
(0.64) (1.05)
n 65 65 65 65
R2 0.373 0.364 0.582
a
Standard errors were calculated using White’s (1980) cor- 7
The predicted values and standard errors around the pre-
rection for heteroskedasticity. All results were obtained using dictor were calculated following Goldberger (1991, p.175).
8
Shazam version 8.0 (White, 1997). The coefficient on the HP method should be interpreted
b
Significantly different from zero at the 5% level. with extra caution since it reflects only two studies in the data
c
Significantly different from zero at the 10% level. set that used this method.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270
268
Table 3
While a negative value for aa means that an
Predicted values per acre of single-service wetlandsa
increase in the size of a wetland pushes down the
value per acre, this effect diminishes rapidly as Service E[ln y] 90% confidence interval around y
ˆ
wetland size increases. Using the coefficient from (1990 US$’s per acre)
model C, a 1% increase in area leads to a 2.9% fall
Lower Mean Upper
in value for a ten-acre wetland. This effect de-
clines geometrically, and for a wetland of 1000
Flood 5.97 89 393 1747
acres the elasticity is only − 0.029. This confirms Quality 6.03 126 417 1378
what we see in Fig. 4 where wetland area appears Quantity 4.84 6 127 2571
to have little impact on value per acre. Rec.fish 5.88 95 357 1342
Com.fish 6.66 108 778 5618
Birdhunt 4.24 25 70 197
6.4. How do wetland ser6ices affect wetland Birdwatch 7.10 528 1212 2782
6alue? Amenity 0.99 1 3 14
Habitat 5.72 95 306 981
Storm 5.47 11 237 5142
The final and central question that we seek to
answer is how wetland services influence wetland a
The results presented in Table 3 are obtained from model
value. The coefficients on the wetland service vari- C. The predicted values are obtained at the means of year and
ables are estimates of the extent to which the acre variables. Except for the variables indicating the respec-
presence of each service changes the value per tive services, all other binary variables are set to zero so that
the prediction reflects a high-quality CV study estimating
acre. A very small coefficient on the habitat vari-
consumer surplus.
able, for example, does not mean that this service
has no value, but that the value of wetlands that
mercial fishing services are among the highest
provide this service are very close to the average
three valued services while amenity services are
value for all wetlands.
the least valued among all wetland services. The
Most of the wetland service variables are not
confidence intervals are extraordinary, spanning
statistically significant. In models A and C, only
thousands of dollars. Clearly it would be highly
the coefficient on the birdwatch variable is signifi-
speculative to use of a single point from this
cant and greater than zero while those on the
distribution in a benefits transfer exercise.
birdhunt and amenity variables are significant and
less than zero. Hence, the data indicate that a
wetland that provides bird watching opportunities
7. Conclusions
is more valuable than the average wetland, while
those that offer bird hunting or amenity services
We have seen that wetland valuation studies are
are less valuable.
remarkably diverse in terms of the values ob-
As one would expect, based on the explanatory
tained, the wetlands evaluated, and the character-
variables in the model, only very imprecise predic-
istics of the studies. Our goal in this study was to
tions of wetland values are possible. Using the
isolate the sources of the variability in the wetland
estimates from model C, Table 3 presents the
value.
predicted values per acre for each possible single-
There is some evidence that the method em-
service wetland and 90% confidence intervals
ployed affects the value obtained. Relative to the
around those estimates.9 Some strong conclusions
HP or RC methods, using the CV method tends
can be drawn from the results. Looking not only
to yield a lower estimated value while there is no
at the mean, but at the upper and lower bounds
statistically significant difference between the CV
of the confidence interval, bird watching and com-
and the TC or NFI methods. While it is perhaps
comforting that the method that is used does not
9
We emphasize that the values in Table 3 do not represent
appear to be a primary determinant of value, the
marginal values and cannot be summed to obtain the value of
unimportance of study quality is not so reassur-
multiple function wetlands.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270 269
Application of wetland valuation techniques: examples
ing. As we saw in Fig. 1, the distribution of weak
from Great Lakes coastal wetlands. University of Michi-
studies is quite similar to that of the values from
gan, School of Natural Resources, Ann Arbor, MI.
strong studies. However, econometric quality was Anderson R., Rockel M., 1991. Economic valuation of wet-
found to be statistically significant in Table 2, and lands. Discussion Paper No. 065. American Petroleum
studies with weak econometrics tended to yield Institute, Washington, DC.
higher values. Study quality also has a substantial Arrow, K., Solow, R., Portney, P.R., Learner, E.E., Radner,
R., Schuman, H., 1993. Report of the NOAA Panel on
impact on the standard error around our predic-
Contingent Valuation. Fed. Reg. 58 (10), 4601 – 4614.
tion, suggesting that quality is important for the Batie S.S., Wilson J.R., 1978. Economic values attributable to
precision of our results. Virginia’s coastal wetlands as inputs in oyster production,
This leads us to our final point: the use of So. J. Agric. Econ., 111-118.
benefits transfer to estimate wetland values faces van den Bergh, J.C.J.M., Button, K.J., 1997. Meta-analysis of
environmental issues in regional, urban and transport eco-
substantial challenges. From our analysis it is
nomics. Urban Studies 34 (5-6), 927 – 944.
clear that the prediction of a wetland’s value
Boyle, K.J., Poe, G.L., Bergstrom, J.C., 1994. What do we
based on previous studies is, at best, an imprecise know about groundwater values? Preliminary implications
science. The need for site-specific studies remains. from meta analysis of contingent valuation studies. Am. J.
Part of the problem lies in the lack of uniformity Agric. Econ. 76 (5), 1055 – 1061.
Brouwer R., Langford I.H., Bateman I.J., Crowards T.C.,
across studies. A better understanding of wetland
Turner R.K., 1997. A meta-analysis of wetland contingent
values might be achieved if future researchers
valuation studies. CSERGE Working Paper GEC 97-20.
follow the suggestions of David (1993) in provid- Centre for Social and Economic Research on the Global
ing more information about their studies and Environment, University of East Anglia, UK.
centralizing the supporting documentation. Until Brouwer, R., 2000. Environmental value transfer: state of the
an improved foundation can be established, it is art and future prospects. Ecol. Econ. 32 (1), 137 – 152.
Carson, R.T., Flores, N.E., Martin, K.M., Wright, J.L., 1996.
important to emphasize the enormous uncertain-
Contingent valuation and revealed preference methodolo-
ties that are present in benefits transfer exercises
gies: comparing the estimates for quasi-public goods. Land
applied to wetlands. In the interim, our analysis Econ. 72 (1), 80 – 99.
provides some guidance as to the wetland services Cooper, H.M., 1989. Integrating Research: a Guide for Litera-
that are most valuable, and the potential biases of ture Reviews, 2nd edn. Sage, Newbury Park, CA.
Costanza, R., d’Arge, R., de Groot, R., Farber, S., Grasso,
some of the valuation methods.
M., Hannon, B., et al., 1997. The value of the world’s
ecosystem services and natural capital. Nature 387, 253 –
260.
Acknowledgements David M.H., 1993. Benefiting benefits transfer: information
systems for complex scientific data. In: Benefits Transfer
Procedures, Problems, and Research Needs, United States
This research was funded in part by a grant
Environmental Protection Agency, Office of Policy, Plan-
from the Texas Water Resources Institute to the ning, and Evaluation (PM-221), Report No. EPA 230-R-
Center for Public Leadership Studies, George 93-018.
Bush School of Government and Public Service, Deck L.B., Chestnut L.G., 1993. Benefits transfer: how good is
Texas A&M University, and by the Texas Agri- good enough? In: Benefits Transfer Procedures, Problems,
and Research Needs, United States Environmental Protec-
cultural Experiment Station. Helpful comments
tion Agency, Office of Policy, Planning, and Evaluation
were provided by Letitia Alston, April Henry, (PM-221), Report No. EPA 230-R-93-018.
Thomas Lacher, R. Douglas Slack, Arnold Freeman, A.M., 1993. The Measurement of Environmental
Vedlitz, an anonymous reviewer and, especially and Resource Values. Resources for the Future. Washing-
Roy Brouwer. Valuable editorial assistance was ton, DC.
Glass G.V., 1976. Primary, secondary and meta-analysis of
provided by Michele Zinn.
research, Educ. Res., 3-8.
Goldberger, A.S., 1991. A Course in Econometrics. Harvard
University Press, Cambridge, MA.
References Heimlich R.E., Weibe K.D., Claassen R., Gadsy D., House
R.M., 1998. Wetlands and agriculture: private interests and
Amacher G.S., Brazee R.J., Bulkley J.W., Moll R.A., 1989. public benefits, Resource Economics Division, E.R.S.,
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270
270
USDA, Agricultural Economic Report 765.10. Schmidt F.L., Hunter J.E, 1977. Development of a general
Kennedy, P.A., 1986. Guide to Econometrics, 2nd edn. MIT solution to the problem of validity generalization, J. Appl.
Press, Cambridge, MA. Psychol., 529-540.
Larson J.S., Adamus P.R., Clairain E.J., 1989. Functional Smith, V.K., Kaoru, Y., 1990. Signals or noise? Explaining the
assessment of freshwater wetlands: a manual and training variation in recreation benefit estimates. Am. J. Agric. Econ.
outline. University of Massachusetts, Amherst, MA 72 (2), 419 – 433.
Loomis, J.B., White, D.S., 1996. Economic benefits of rare and Smith, V.K., Osborne, L.L., 1996e. Do contingent valuation
endangered species: summary and meta-analysis. Ecol. estimates pass a ‘scope’ test? A meta analysis. J. Environ.
Econ. 18 (3), 197 – 206. Econ. Mgmt. 31, 287 – 301.
Mitsch, W.J., Gosselink, J.G., 1986. Wetlands. Van Nostrand White, K.J., 1997. SHAZAM User’s Reference Manual Version
Reinhold, New York. 8.0, McGraw-Hill, Vancouver.
.
www.elsevier.com/locate/ecolecon
ANALYSIS
The economic value of wetland services: a meta-analysis
Richard T. Woodward *, Yong-Suhk Wui
Department of Agricultural Economics, Texas A&M Uni6ersity, 2l24 TAMU, College Station, TX 77843-2124, USA
Received 30 March 2000; received in revised form 15 November 2000; accepted 16 November 2000
Abstract
The number of studies quantify the value of wetlands and the services provided by these ecosystems is rapidly
expanding. The time is ripe for an assessment of what has been learned from this literature. Using results from 39
studies, we evaluate the relative value of different wetland services, the sources of bias in wetland valuation and the
returns to scale exhibited in wetland values. While some general trends are beginning to emerge, the prediction of a
wetland’s value based on previous studies remains highly uncertain and the need for site-specific valuation efforts
remains large. © 2001 Elsevier Science B.V. All rights reserved.
growing need to quantify the value of wetland
1. Introduction
services.
The services provided by wetlands include habi-
The valuation of wetlands’ ecological services is
tat for species, protection against floods, water
a relatively recent phenomenon. Historically, wet-
purification, amenities and recreational opportu-
lands were viewed as a waste of valuable land that
nities. Because these services typically have no
could only be ‘improved’ through drainage and
market price, a measure of their values can only
destruction of the wetland (Mitsch and Gosselink,
be obtained through non-market valuation tech-
1986). Today, while there is now widespread niques. Many wetland valuation studies have been
recognition that wetlands provide valuable eco- conducted and the range of the estimates is re-
logical services, there remain substantial debates markable. A recent review by Heimlich et al.
over whether particular areas are in their highest (1998) lists 33 studies over the last 26 years with
economic use as wetlands, and to what extent per acre values ranging from US$0.06 to
public and private resources should be used for US$22050. Even within the same study looking at
their protection and restoration. Hence, there is a a single ecosystem function, Batie and Wilson
(1978) find values per acre that differ by two
orders of magnitude from one site to another.
* Corresponding author. Tel.: + 1-979-8455864; fax: + 1-
The purpose of this paper is to assess whether
979-8454261.
any systematic trends can be distilled from the
E-mail address: r-woodward@tamu.edu (R.T. Woodward).
0921-8009/01/$ - see front matter © 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 1 - 8 0 0 9 ( 0 0 ) 0 0 2 7 6 - 7
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270
258
Estimates of the value of a wetland can also
breadth of wetland valuation studies conducted to
influence site-specific valuation efforts in two
date, and to shed light on what factors determine
ways. First, they might provide Bayesian priors
a wetland’s value. We maintain an assumption
that might be formally incorporated into the valu-
that there exists an unobserved valuation function
ation exercise. Secondly, they may give re-
that determines a wetland’s value given its physi-
searchers a sense of where the values at stake are
cal, economic and geographic characteristics. Af-
likely to be of greatest social importance and
ter reviewing 46 studies, data from 39 wetland
might, therefore, influence where detailed studies
valuation studies were identified that had suffi-
are carried out.
cient commonalties to allow inter-study compari-
The paper is organized as follows. In the next
sons. We used two techniques to learn about the
section we provide a brief overview of the eco-
valuation function, both of which can be broadly
nomics of wetland valuation. We survey the eco-
described as meta-analysis since many studies are
logical functions and economic services provided
used to identify general relationships. The first
by these areas, the basis for their valuation and
method that we employ uses bivariate graphical
the techniques that are used to place an economic
and standard techniques. This gives us both an
value on wetlands. Section 3 provides a brief
indication of the extent to which particular char-
summary of meta-analysis as a tool. In Sections 5
acteristics influence wetland values while also por-
and 6 we explore the trends in the data, identify-
traying the full distribution of the data. The
ing the sources of variability in wetland values.
second technique is more standard, using a multi-
We conclude by reflecting on the implications of
variate regression of wetland values on the char-
our analysis both for our understanding of wet-
acteristics of both the wetlands and the studies.
land values and for future research.
Together, these two techniques provide a richer
basis from which we can draw lessons on the
factors determining wetland value.
2. The value of wetland functions
There are numerous reasons why understanding
the value of wetland services might be useful. The
2.1. Wetland functions and ser6ices
most obvious is that if the value of these services
were known, benefits transfer efforts could be While an inclusive definition of wetlands is
substantially improved. As Deck and Chestnut difficult to state, they are generally characterized
(1993) point out, benefits transfer may play a as being moist during an extended period each
variety of different roles, ranging from an attempt year with soils, plants and animals that are dis-
to place a precise value on a particular resource to tinct from their aquatic and terrestrial neighbors.
providing information that feeds into the process These transition areas are highly diverse, ranging
of building support for projects already imple- from coastal mangroves that are inundated with
mented. While benefits transfer is rarely suitable water most of the year to areas that are moist for
for the former case, it might often be appropriate only a few months during the year. Partly because
for the latter. Another form of benefits transfer is they share features of both terrestrial and aquatic
the use of estimated values to predict the aggre- systems, wetlands are remarkably productive.
gate value of similar systems nationally or In assessing the value of wetlands, it is useful to
globally. Such estimates can be useful in setting distinguish the systems’ ecological functions from
national priorities or the evaluation of policies the associated services that are directly valued by
with impacts that are national or global. Costanza humans (Costanza et al., 1997). Larson et al.
et al. (1997), for example, placed a value on the (1989) list 17 services and functions provided by
entire globe’s ecosystems. While such expansive the world’s ecosystems (Table 1). In our data set,
efforts may be overambitious, the aggregate num- the services were grouped into ten categories as
bers do help to get the attention of policy makers indicated in the table. The measurement of the
and the public. value of these services varies substantially both in
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270 259
Table 1
Wetland functions, the associated economically valuable goods and services and the names of variables that capture the presence of
these in the dataa
Function Economically valuable good(s) and/or Technique(s) typically used to quantiiy
service(s) (variable names) the value of the service(s)
Recharge of ground water Increased water quantity (quantity) Net factor income or replacement cost
Discharge of ground water Increased productivity of downstream Net factor income, replacement cost or
fisheries (com.fish) travel cost
Water quality control Reduced costs of water purification Net factor income or replacement cost
(quality)
Retention, removal and transformation Reduced costs of water purification Net factor income or replacement cost
of nutrients (quality)
Habitat for aquatic species Improvements in commercial and/or Net factor income, replacement cost,
recreational fisheries either on or offsite travel cost or contingent valuation
(com.fish and rec.fish). Nonuse
appreciation of the species (habitat)
Habitat for terrestrial and avian species Recreational observation and hunting of Travel cost or contingent valuation
wildlife (birdwatch & birdhunt). Nonuse
appreciation of the species (habitat)
Biomass production and export (both Production of valuable food and fiber Net factor income
plant and animal) for harvest (birdhunt & com. fish)
Flood control and storm buffering Reduced damage due to flooding and Net factor income or replacement cost
severe storms (flood)
Stabilization of sediment Erosion reduction (storm) Net factor income or replacement cost
Overall environment Amenity values provided by proximity Hedonic pricing
to the environment (amenity)
a
The first two columns are adapted from Larson et al. (1989).
the methods that are used and the economic tice, it is often assumed that demand is perfectly
theory that underlies the valuation exercise. The elastic so that the impact of the wetland on con-
most common methods for measuring the eco- sumer surplus can be ignored. Other times the
nomic value of these services are also presented in producer surplus that is generated by a wetland is
the table. Though not indicated in the table, the estimated using the replacement cost (RC)
contingent valuation method can, in principle, be method. This approach values the wetland’s ser-
used to measure the value of all these services. vice based on the price of the cheapest alternative
way of obtaining that service. For example, the
2.2. The 6aluation of wetland ser6ices value of a wetland in the treatment of wastewater
might be estimated using the cost of chemical or
The economic value of resources such as wet- mechanical alternatives. As noted by Anderson
lands is equal to the benefits (net of costs) that and Rockel (1991), the replacement cost method
these systems provide to humans (Freeman, 1993). is actually an upper bound on the true value since
Four methods are listed in Table 1 as commonly the producer may not choose to actually use that
used to place a value on wetlands. The net factor alternative considered.
income (NFI) method is most appropriate when Non-market values can be measured using
the wetland provides a service that leads to an travel cost (TC), contingent valuation (CV), or
increase in producer surplus. In the NFI method hedonic pricing (HP) methods. There is, however,
the physical relationship between wetland area substantial variability within each of these ap-
and the economic activity is estimated. It is then proaches, much more than can be discussed here
possible to identify the increase in producer sur- (Freeman, 1993), and the literature is filled with
plus associated with the wetland’s area. In prac- reasons why results might be biased or otherwise
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270
260
deficient. While in principle each method can give the first to use meta-analysis in the context of
a correct estimate of economic value, it is easy to nonmarket valuation, looking at values placed on
misuse the method and obtain results that may outdoor recreation. Since then, meta-analysis has
have little relation to the true value. been used to study air pollution, recreational
Some wetland values are obtained using meth- fishing, visibility, health risks, endangered species
and wetlands.1
ods that do not estimate economic surplus and,
therefore, lack a foundation in standard economic The basic approach used in most valuation
theory. Of the studies identified in our review of meta-analyses is the same. A set of studies is
the literature, two values were estimated using selected yielding a number of values that become
energy analysis in which the value is based on the the dependent variable. The independent variables
gross primary production of the ecosystem, and are the characteristics of each study and study
five values were obtained using the market value site. If a single study reports numerous values,
of the products extracted. These methods have then several data points are obtained. Meta-anal-
been strongly criticized (Anderson and Rockel, ysis allows the evaluation of the effect of changes
1991). Energy analysis equates the energy embod- in the underlying environmental attribute on
ied in a wetland’s biota with the energy purchased value. Such analysis is usually not possible in the
in fossil fuels. Since the correlation between en- context of a single study since most such at-
ergy content and consumer preferences is quite tributes are held constant.
weak, the technique is a poor predictor of eco- A good example of the benefits of meta-analysis
nomic value. The market value technique is also is seen in Smith and Osborne’s (1996) analysis of
flawed since it cannot capture consumer surplus the value of improvements in visibility. Since visi-
and can lead to over-estimate producer surplus if bility varies continuously from zero to one, the
cost of extracting the valued products is not authors are able to estimate the marginal benefit
subtracted. of improvements in visibility. One of the most
significant limitations of meta-analysis, however,
is the lack of comparability across studies (van
3. Meta-analysis as a tool in understanding den Bergh and Button, 1997). Characteristics of
non-market valuation the resource being valued are often presented in
such diverse fashion that the best that the analyst
First used by psychologists (Glass, 1976; can do is to use a binary variable to indicate
Schmidt and Hunter, 1977), meta-analysis has whether an attribute is reflected in each value.
proved to be a useful tool for synthesizing the Boyle et al. (1994), for example, have no data on
results of numerous studies. The method has re- the level of cancer risk in each of their eight
cently gained attention in economics as a way to studies, only an indication of whether such a risk
appreciate numerous studies that have placed eco- was mentioned in the study. Similarly, in this
nomic values on environmental goods and ser-
study wetland services are captured using qualita-
vices (see Brouwer (2000) for a review). The
tive variables.
central advantage of meta-analysis is that it pro-
Brouwer et al. (1997) is, to our knowledge, the
vides a rigorous statistical synthesis of the litera-
only other attempt to carry out meta-analysis of
ture that cannot be achieved using more
wetland valuation studies. In their work, only
qualitative analysis.
contingent valuation studies are considered. This
There are two main types of meta-analyses:
narrow focus allowed the authors to develop a
those that use the actual data from multiple stud-
ies, and those that use the results of multiple
1
The analysis of wetlands by Brouwer et al. (1997) looked
studies. It is the later method that has been ap-
at WTP estimates from CV studies and used a more expansive
plied to interpret valuation studies. Brouwer lists interpretation of ‘wetlands’ than we retain in this paper.
ten studies that have used meta-analysis to study Hence, their results are not directly comparable to the results
valuation efforts. Smith and Kaoru (1990) were here.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270 261
rich set of variables characterizing the qualities of ues: variation due to differing characteristics of
the studies. However, this is also a limitation in the wetlands, i.e. along the function; and variation
that it eliminates any variability associated with due to error in the estimation of the true value,
valuation method and reduces the variability in i.e. deviations from the function.
services that can be considered. In the next section Of course, the wetlands that have been valued
we discuss how we attempted to overcome these were almost certainly not chosen at random from
limitations in our data. the total population. First there is the problem of
selection bias. It seems likely that wetlands that
are considered valuable a priori are much more
likely to be studied and valued. This need not lead
4. The wetland valuation data
to errors in our estimation of the valuation func-
tion if all important variables are accurately mea-
After a lengthy review of the literature, we
sured, but given the limitations in the available
identified 39 studies that contained sufficient data
data, the likelihood of such bias should be taken
to allow inter-study comparisons. Many other
into account in benefits transfer exercises or any
studies were identified but could not be used. The
other attempt to extrapolate estimated values.
values are taken from published reports, ‘gray’
Similarly, the fact that many of the studies have
literature, and theses.2 Because of our desire to
been filtered by the peer review process might
synthesize wetland values from all different ser-
have excluded some estimates. Good estimates
vices, we use annual value per acre in 1990 US
that are either not statistically different from zero
dollars. This distinguishes our work from other
or are much higher than anticipated may not be
meta-analyses which typically use willingness to
published.
pay (WTP) per person (e.g. Brouwer et al., 1997).
The values in our data are also not independent
WTP per person is not applicable here because
draws. Numerous studies generate multiple mea-
some methods (e.g. NFI) do not lead to a WTP
sures of wetland value, so that, as pointed out by
per person measure. On the other hand, if WTP
Smith and Kaoru (1990), the data have panel
per person is available, then value per acre can be
characteristics. Furthermore, researchers who
calculated with knowledge of the relevant popula-
work closely together are likely to share practices
tion and the wetland’s size. When capitalized val-
that differ in important ways from others. Finally,
ues were reported, they were annualized assuming
since there has no doubt been learning over time,
constant value per year and using discount factors
both in terms of methodology and the values that
provided in the studies or a 6% rate in the two
are reasonable, the data also probably suffer from
studies that did not state a discount rate.
some autocorrelation.
In our analysis we assume that a wetland’s
value is a function of the system’s ecological
characteristics and its socio-economic environ-
5. Bivariate meta-analysis
ment. Each wetland in our data is interpreted as a
(not necessarily random) draw from the popula-
Using the available data, we now evaluate the
tion of all wetlands. We assume that there exists a
sources of variation in estimates of wetland value.
true public WTP at a given moment for a particu-
Two complementary techniques are used. In this
lar wetland. While this true WTP cannot be ob-
section we explore some of the relationships in the
served directly, it can be estimated using the
data using graphical presentation and bivariate
methods discussed above. Seen in this way, there
statistics. The advantage of this analysis is that it
are two sources of variability in the wetland val-
allows us to present the full data set graphically,
making possible a richer appreciation of the data.
2
The complete data set, including a description of each
However, the bivariate analysis ignores interac-
study and an explanation of the interpretations of the data
tions between explanatory variables. Hence, a sec-
that were made is available from the authors on request or via
ond and more standard technique is also used,
the internet at http://ageco.tamu.edu/faculty/woodward/.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270
262
that of estimating a valuation function using mul- difference is found it is sensible to retain the ‘bad’
tivariate regression techniques. studies because they contain other variation in
methods (such as different samples and locations)
5.1. Variation due to measurement error or bias that, by their inclusion, will help answer many
other questions surrounding the problem area.’’
As noted above, there are two types of varia- Following Cooper’s advice, some meta-analyses
tion with which we are concerned, deviations include objective indicators of study quality such
from the valuation function due to bias or errors as response rate or study format (Brouwer et al.,
in estimation, and variations along the valuation 1997; Loomis and White, 1996). Given the diver-
function attributable to different wetland charac- sity of studies considered in our analysis, no
teristics. We begin by looking at sources of sys-
standard objective indicator of quality was avail-
tematic error because of study weaknesses and
able; only a subjective assessments of study qual-
bias because of the valuation method used.
ity could be used. Each study was ranked on a
One might expect study quality to affect esti-
scale of 1–3 in four categories: the apparent
mates of wetland value. Attention to this issue is
quality of the data, the theoretical consistency of
potentially important because there is substantial
the methodology, econometric techniques and
variability in the quality of wetland valuation
statistical certainty. A study was given a rank of 1
studies. While some studies are characterized by
if we felt that this feature of the study made the
sound theoretical foundations and state-of-the-art
results highly questionable. Studies with a 1 in
econometric methods, others are crippled by
any of the quality categories are called ‘weak’ in
faulty logic, poor data or incorrect economic
the figures and econometric analysis below.3 A
analysis.
‘weakness’ in a study should not be interpreted as
The weakness of many wetland valuation ef-
a condemnation since valuation may not have
forts is widely recognized. In their review of wet-
been the authors’ primary objective or data limi-
land valuation studies, Anderson and Rockel
(1991) found only five studies that they deemed tations may have been prohibitive.
credible enough to list in their summary table. On average, the weak and strong studies do not
However, while it may appear obvious that only yield statistically different values. Excluding the
high quality analyses should be used in meta-anal- highest value in our data set, the average of the
ysis, it is also clear that the evaluation of quality weak studies is US$986 per acre versus US$915
is likely to be quite subjective. The problem of for the strong studies.4 When looking at the com-
subjectivity is particularly problematic in wetland plete distribution of these studies however, there
valuation studies because few efforts satisfy the do appear to be some systematic differences two
highest standards of quality, in large part because sets of values. Fig. 1 plots the rank of both the
of data limitations. In CV studies, for example, weak and the strong studies in their respective
strict adherence to the guidelines of the NOAA
panel (Arrow et al., 1993) is often impossible
because of budgetary restrictions. Hence, there is
3
For studies that are also evaluated by Anderson and
a great deal of subjectivity in assessing how good
Rockel (1991) our critique was generally consistent. Some
is good enough.
studies which we ranked as a 2 were questioned, but did not
As Cooper (1989, p. 67) points out, ‘‘The deci- appear to be completely rejected by Anderson and Rockel.
sion to include or exclude studies on an a priori 4
The highest value from the Amacher et al. (1989) study is
basis requires the reviewer to make an overall excluded as it is over 60 times the second largest value. After
excluding this value, the mean of the weak studies is not
judgment of quality that is often too subjective to
significantly different from the mean of the strong studies at
be credible.’’ He argues that it makes more sense
the 10% level. This value, one other value estimated using
to enumerate characteristics of each study and energy analysis and five values estimated using the market
then evaluate whether ‘good’ methods lead to value of the output are excluded from the econometric analysis
different results than ‘bad’ methods. ‘‘When no below.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270 263
Fig. 1. Cumulative distributions of wetland values broken down by study quality.
convergent validity of CV analysis relative to
categories on the vertical axis, against the esti-
other methods (e.g. Carson et al., 1996).
mated values per acre on the horizontal axis. The
Fig. 2 presents the distributions of the values
solid lines in the figure represent cumulative dis-
taken from the four primary methods used to
tribution functions (cdfs) of distributions from
measure wetland values. The means of the values
which the data appear to be drawn. While the
from these methods vary from a low of US$198
distribution of the strong data closely resembles a
for the travel cost method to a high of US$1555
log normal distribution, the weak data seem to be
for the replacement cost method. However, be-
drawn from a uniform distribution. There is also
cause of the substantial variability in the data,
slightly less variance in the strong studies, indicat-
none of the means are statistically different from
ing that the lack of quality may not bias the
each other. Still, some patterns are evident. At
estimated value, but it might have implications for
one extreme, the net-factor input method is a
the accuracy of the predictions. This result is
lower bound on the distribution of values. At the
confirmed in the multivariate analysis below.
other extreme, the distribution of values obtained
Nonetheless, we do not find the kind of dramatic
using CV nearly stochastically dominates the dis-
difference between the two groups that would
tributions of values from the other three methods.
justify discarding the weak studies from the data
These findings do not necessarily indicate biases
set.
in these techniques. Because of the small sample
Another potential reason that the estimated
size we cannot statistically reject the hypothesis
value may deviate from the valuation function
that the distributions are the same. Moreover,
might be bias due to the method that was used in
different methods are used to value different ser-
the study. In principle, if two methods seek to
vices. It may be that CV is used for high-value
estimate consumer surplus from the same wetland
then they should yield similar values. If there is no services while the NFI method is used for low-
systematic difference between two techniques, value services. Hence, the question of whether the
then they are said to satisfy the criterion of con- method itself is a source of bias can only be
vergent validity. Numerous studies have tested the explored using multivariate analysis.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270
264
Fig. 2. Cumulative distributions of wetland values broken down by valuation method.
5.2. Variation in 6alue due to wetland valued. The correlation between estimated value
characteristics per acre and the number of services is only 0.10
and, based on the Spearman rank criterion, the
We now turn an initial analysis of the sources hypothesis of no correlation cannot be rejected at
of variation in the valuation function. Ten vari- the 10% level.
ables were defined indicating whether a particular In addition to being affected by wetland ser-
wetland service was reflected in each study. These vices, one might also expect the value per acre to
are listed in Table 1. Identifying the services be a function of the wetland’s area. In this case
reflected in a study often involves some subjectiv- there is no clear a priori expectation as to the
ity, particularly in CV studies since respondents form that such a relationship might take. Eco-
might be aware of services other than those about nomic intuition would suggest that the marginal
which they had been explicitly asked in the value of each acre would tend to decline. On the
survey.5 other hand, based on ecological principles of
A relatively weak hypothesis would be that functional interdependence, one might expect that
increasing the number of services considered in a larger wetlands would provide a richer and more
valuation exercise would tend to increase a wet- valuable set of services. This relationship is plot-
land’s estimated value. This relationship is pre- ted in Fig. 4. There is no apparent relationship
sented in Fig. 3. While almost two-thirds of the between wetland area and value in the figure and,
studies measured the value of only one wetland once again, the hypothesis of no significant corre-
service, more than 30% of the studies measured lation cannot be rejected at the 10% level.
three or more services. Contrary to our hypothe- Our analysis to this point is quite inconclusive.
sis, there is no noticeable relationship between the There is some evidence that CV studies tend to
value of a wetland and the number of services yield greater values than any other method, but
no visible relationship between value per acre and
either the number of services or the size of the
5
In a few instances authors were contacted to assist us in
wetland. However, while we find the bivariate
obtaining the most accurate interpretation possible.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270 265
Fig. 3. Wetland values and the number services present.
analysis useful, it cannot distinguish how multiple estimate of producer’s surplus, PS; and whether
factors might be interacting to influence wetland the results had been published, published. The
value. In the next section we attempt to tease out variables data0, theory0 and metric0, are dummy
more understanding of the wetland valuation variables set at one if the data, theory or econo-
function using multivariate regression analysis. metrics used in the study were deemed highly
questionable.6
We should recognize that there are certainly
6. Multivariate meta-analysis of wetland values important variables that determine a wetland’s
value that are omitted from our model. Charac-
In this section we estimate a parametric specifi- teristics of the population near a wetland are
cation of the valuation function using the data particularly likely to influence the value placed on
discussed above. After excluding incomplete ob- the area. However, such data could not be iden-
servations and values based on either energy anal- tified in most of the studies; we were unable to
ysis or the market value methods, the 65 include any such variables in our model. While
observations of wetland values were obtained. the absence of these variables no doubt greatly
diminishes the explanatory power of our analysis,
6.1. The estimated model and results it need not bias the estimated coefficients if these
variables are uncorrelated with the included set
The dependent variable in all regressions is the (Kennedy, 1986).
natural log of the value per acre of wetland Our econometric model is based on a main-
converted to 1990 dollars, the mean of which is tained hypothesis that measured wetland value
4.92. In addition to the variables discussed above
representing services, area and study quality, we 6
Since only two studies were deemed weak based on statisti-
included variables indicating date of the study cal significance, and many studies did not report sufficient
(1960= 0), year; whether the wetland was a information to gauge the statistical accuracy of their estimates,
coastal wetland, coastal; whether the value was an this variable was excluded from the econometric analysis.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270
266
Fig. 4. Wetland area and values.
per acre, y, is a function of the services provided, values and the quality of the studies. Model C
xs, the methodology used, xm, the acres of the combines both the characteristics of the sites and
wetland xa, other variables describing the study variables related to how the values were
including year and location, x0, and a constant estimated.
term. The fit was substantially improved by using
the logs of both the per-acre value and the acres.
6.2. Do the study quality or 6aluation method
Hence, the estimated linear model is
affect the 6alue obtained?
%x % %
ln(y) =a+ba ln(xa)+ bs s +bmxm +b0x0 (1)
In our discussion of Fig. 1 we argued that there
where a is the constant term and the b’s are the was little evidence of bias as a result of the quality
estimated coefficients on the respective explana- of the studies, and our regression results largely
tory variables. confirm that conclusion. The coefficients on the
The results of several regressions are presented variables indicating poor quality theory and data
in Table 2. In each case the hypothesis of ho- are both statistically insignificant, as is the coeffi-
moskedasticity was rejected at the 5% significance cient indicating whether the study was published.
level using the BPG test. Accordingly, the stan- However, the variable indicating econometric
dard errors were estimated using White’s (1980) quality was strongly significant in both regressions
correction. Model A presents the estimated coeffi- B and C. Holding all else constant, the values
cients of a model in which it is assumed that the from studies with poor quality econometrics aver-
variability in the values is solely a function of the age 24–50 times greater than those from those
physical characteristics of the wetland systems, with comparatively strong econometric
ignoring any systematic variation due to the way foundations.
that the values were estimated. Model B takes the Study quality also has important consequences
opposite approach, explaining the values based for the confidence we place on predicted values.
solely on the methods used to measure those Using the results from model C evaluated at the
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270 267
Table 2 means of the variables the log of the wetland
Estimated models of the wetland valuation functiona (log of value predicted for a high-quality unpublished
value per acre dependent variable, standard errors in parenthe-
study is 5.68 with a standard error around the
ses)
prediction (|p) of 0.61. If it is assumed that the
study is published, however, |p falls to only 46%
Variable Mean A B C
of the original value. For studies that are weak in
7.945b 6.641b 7.872b
Intercept
the areas of theory or econometrics, |p increases
(1.07) (1.31) (1.74)
by 1.9 or 2.1-fold, respectively. On the other
Year 14.908 −0.052 −0.004 0.016
hand, the impact of study’s data being of poor
(0.03) (0.04) (0.04)
−0.168c −0.286b quality is slight, leading to a 4% decline in |p.7
Ln acres 9.281
(0.10) (0.11)
Study quality is important not so much because it
Coastal 0.431 −0.523 −0.117
might bias results, but because high quality stud-
(0.71) (0.68)
Flood 0.138 −0.358 0.678 ies lead to a much more precise basis for
(1.03) (0.77)
prediction.
1.494c
Quality 0.200 0.737
There is some evidence that the method used
(0.78) (0.75)
Quantity 0.062 0.514 −0.452 has a statistically significant effect on the value
(1.60) (1.54)
obtained. As in Fig. 2, in model B we find that
Rec. Fish 0.354 0.395 0.582
CV tends to dominate other methods as the signs
(0.55) (0.56)
on their coefficients are either negative or statisti-
Com. Fish 0.277 0.669 1.360
(0.79) (1.01) cally insignificant. However, when variables indi-
−1.311b −1.055b
Birdhunt 0.400
cating the wetland services are introduced in
(0.49) (0.52)
model C, the dominance of CV disappears and
1.704b 1.804b
Birdwatch 0.277
(0.52) (0.59) the sign on HP and RC methods becomes signifi-
−3.352b −4.303b
Amenity 0.154
cantly positive. Hence, relative to these methods,
(0.93) (0.95)
CV studies tend to find a lower value per acre and
Habitat 0.308 0.577 0.427
we cannot conclude that this method is biased
(0.56) (0.59)
Storm 0.031 0.310 0.173 relative to the TC or NFI method.8
(2.37) (1.66)
Publish 0.769 −0.669 −0.154
6.3. Do wetlands 6alues exhibit returns to scale?
(0.72) (0.71)
Data0 0.246 0.302 0.000
(0.56) (0.60)
The coefficient on LnAcres is consistently nega-
Theory0 0.215 −1.020 −1.045
tive and statistically significant across the models
(0.84) (0.84)
−4.030b −3.186b
Metric0 0.123 reported in Table 2, indicating significant decreas-
(1.21) (1.22)
ing returns to scale. However, because of the
−2.416b −2.034b −3.140b
PS 0.277
double-log functional form, the scale effect is
(0.83) (0.72) (0.86)
5.043b extremely small for large wetlands. From Eq. (1),
HP 0.031 0.441
(1.02) (1.12) the marginal effect of an increase in the size of a
NFI 0.246 −0.724 0.273
wetland is
(0.82) (0.90)
2.232b
RC 0.277 1.376
(y/(xa = aax (aa − 1)e (a0 + asxs + amxm + a0x0)
(0.86) (0.89) a
−1.196c
TC 0.108 −0.341
(0.64) (1.05)
n 65 65 65 65
R2 0.373 0.364 0.582
a
Standard errors were calculated using White’s (1980) cor- 7
The predicted values and standard errors around the pre-
rection for heteroskedasticity. All results were obtained using dictor were calculated following Goldberger (1991, p.175).
8
Shazam version 8.0 (White, 1997). The coefficient on the HP method should be interpreted
b
Significantly different from zero at the 5% level. with extra caution since it reflects only two studies in the data
c
Significantly different from zero at the 10% level. set that used this method.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270
268
Table 3
While a negative value for aa means that an
Predicted values per acre of single-service wetlandsa
increase in the size of a wetland pushes down the
value per acre, this effect diminishes rapidly as Service E[ln y] 90% confidence interval around y
ˆ
wetland size increases. Using the coefficient from (1990 US$’s per acre)
model C, a 1% increase in area leads to a 2.9% fall
Lower Mean Upper
in value for a ten-acre wetland. This effect de-
clines geometrically, and for a wetland of 1000
Flood 5.97 89 393 1747
acres the elasticity is only − 0.029. This confirms Quality 6.03 126 417 1378
what we see in Fig. 4 where wetland area appears Quantity 4.84 6 127 2571
to have little impact on value per acre. Rec.fish 5.88 95 357 1342
Com.fish 6.66 108 778 5618
Birdhunt 4.24 25 70 197
6.4. How do wetland ser6ices affect wetland Birdwatch 7.10 528 1212 2782
6alue? Amenity 0.99 1 3 14
Habitat 5.72 95 306 981
Storm 5.47 11 237 5142
The final and central question that we seek to
answer is how wetland services influence wetland a
The results presented in Table 3 are obtained from model
value. The coefficients on the wetland service vari- C. The predicted values are obtained at the means of year and
ables are estimates of the extent to which the acre variables. Except for the variables indicating the respec-
presence of each service changes the value per tive services, all other binary variables are set to zero so that
the prediction reflects a high-quality CV study estimating
acre. A very small coefficient on the habitat vari-
consumer surplus.
able, for example, does not mean that this service
has no value, but that the value of wetlands that
mercial fishing services are among the highest
provide this service are very close to the average
three valued services while amenity services are
value for all wetlands.
the least valued among all wetland services. The
Most of the wetland service variables are not
confidence intervals are extraordinary, spanning
statistically significant. In models A and C, only
thousands of dollars. Clearly it would be highly
the coefficient on the birdwatch variable is signifi-
speculative to use of a single point from this
cant and greater than zero while those on the
distribution in a benefits transfer exercise.
birdhunt and amenity variables are significant and
less than zero. Hence, the data indicate that a
wetland that provides bird watching opportunities
7. Conclusions
is more valuable than the average wetland, while
those that offer bird hunting or amenity services
We have seen that wetland valuation studies are
are less valuable.
remarkably diverse in terms of the values ob-
As one would expect, based on the explanatory
tained, the wetlands evaluated, and the character-
variables in the model, only very imprecise predic-
istics of the studies. Our goal in this study was to
tions of wetland values are possible. Using the
isolate the sources of the variability in the wetland
estimates from model C, Table 3 presents the
value.
predicted values per acre for each possible single-
There is some evidence that the method em-
service wetland and 90% confidence intervals
ployed affects the value obtained. Relative to the
around those estimates.9 Some strong conclusions
HP or RC methods, using the CV method tends
can be drawn from the results. Looking not only
to yield a lower estimated value while there is no
at the mean, but at the upper and lower bounds
statistically significant difference between the CV
of the confidence interval, bird watching and com-
and the TC or NFI methods. While it is perhaps
comforting that the method that is used does not
9
We emphasize that the values in Table 3 do not represent
appear to be a primary determinant of value, the
marginal values and cannot be summed to obtain the value of
unimportance of study quality is not so reassur-
multiple function wetlands.
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270 269
Application of wetland valuation techniques: examples
ing. As we saw in Fig. 1, the distribution of weak
from Great Lakes coastal wetlands. University of Michi-
studies is quite similar to that of the values from
gan, School of Natural Resources, Ann Arbor, MI.
strong studies. However, econometric quality was Anderson R., Rockel M., 1991. Economic valuation of wet-
found to be statistically significant in Table 2, and lands. Discussion Paper No. 065. American Petroleum
studies with weak econometrics tended to yield Institute, Washington, DC.
higher values. Study quality also has a substantial Arrow, K., Solow, R., Portney, P.R., Learner, E.E., Radner,
R., Schuman, H., 1993. Report of the NOAA Panel on
impact on the standard error around our predic-
Contingent Valuation. Fed. Reg. 58 (10), 4601 – 4614.
tion, suggesting that quality is important for the Batie S.S., Wilson J.R., 1978. Economic values attributable to
precision of our results. Virginia’s coastal wetlands as inputs in oyster production,
This leads us to our final point: the use of So. J. Agric. Econ., 111-118.
benefits transfer to estimate wetland values faces van den Bergh, J.C.J.M., Button, K.J., 1997. Meta-analysis of
environmental issues in regional, urban and transport eco-
substantial challenges. From our analysis it is
nomics. Urban Studies 34 (5-6), 927 – 944.
clear that the prediction of a wetland’s value
Boyle, K.J., Poe, G.L., Bergstrom, J.C., 1994. What do we
based on previous studies is, at best, an imprecise know about groundwater values? Preliminary implications
science. The need for site-specific studies remains. from meta analysis of contingent valuation studies. Am. J.
Part of the problem lies in the lack of uniformity Agric. Econ. 76 (5), 1055 – 1061.
Brouwer R., Langford I.H., Bateman I.J., Crowards T.C.,
across studies. A better understanding of wetland
Turner R.K., 1997. A meta-analysis of wetland contingent
values might be achieved if future researchers
valuation studies. CSERGE Working Paper GEC 97-20.
follow the suggestions of David (1993) in provid- Centre for Social and Economic Research on the Global
ing more information about their studies and Environment, University of East Anglia, UK.
centralizing the supporting documentation. Until Brouwer, R., 2000. Environmental value transfer: state of the
an improved foundation can be established, it is art and future prospects. Ecol. Econ. 32 (1), 137 – 152.
Carson, R.T., Flores, N.E., Martin, K.M., Wright, J.L., 1996.
important to emphasize the enormous uncertain-
Contingent valuation and revealed preference methodolo-
ties that are present in benefits transfer exercises
gies: comparing the estimates for quasi-public goods. Land
applied to wetlands. In the interim, our analysis Econ. 72 (1), 80 – 99.
provides some guidance as to the wetland services Cooper, H.M., 1989. Integrating Research: a Guide for Litera-
that are most valuable, and the potential biases of ture Reviews, 2nd edn. Sage, Newbury Park, CA.
Costanza, R., d’Arge, R., de Groot, R., Farber, S., Grasso,
some of the valuation methods.
M., Hannon, B., et al., 1997. The value of the world’s
ecosystem services and natural capital. Nature 387, 253 –
260.
Acknowledgements David M.H., 1993. Benefiting benefits transfer: information
systems for complex scientific data. In: Benefits Transfer
Procedures, Problems, and Research Needs, United States
This research was funded in part by a grant
Environmental Protection Agency, Office of Policy, Plan-
from the Texas Water Resources Institute to the ning, and Evaluation (PM-221), Report No. EPA 230-R-
Center for Public Leadership Studies, George 93-018.
Bush School of Government and Public Service, Deck L.B., Chestnut L.G., 1993. Benefits transfer: how good is
Texas A&M University, and by the Texas Agri- good enough? In: Benefits Transfer Procedures, Problems,
and Research Needs, United States Environmental Protec-
cultural Experiment Station. Helpful comments
tion Agency, Office of Policy, Planning, and Evaluation
were provided by Letitia Alston, April Henry, (PM-221), Report No. EPA 230-R-93-018.
Thomas Lacher, R. Douglas Slack, Arnold Freeman, A.M., 1993. The Measurement of Environmental
Vedlitz, an anonymous reviewer and, especially and Resource Values. Resources for the Future. Washing-
Roy Brouwer. Valuable editorial assistance was ton, DC.
Glass G.V., 1976. Primary, secondary and meta-analysis of
provided by Michele Zinn.
research, Educ. Res., 3-8.
Goldberger, A.S., 1991. A Course in Econometrics. Harvard
University Press, Cambridge, MA.
References Heimlich R.E., Weibe K.D., Claassen R., Gadsy D., House
R.M., 1998. Wetlands and agriculture: private interests and
Amacher G.S., Brazee R.J., Bulkley J.W., Moll R.A., 1989. public benefits, Resource Economics Division, E.R.S.,
R.T. Woodward, Y.-S. Wui / Ecological Economics 37 (2001) 257–270
270
USDA, Agricultural Economic Report 765.10. Schmidt F.L., Hunter J.E, 1977. Development of a general
Kennedy, P.A., 1986. Guide to Econometrics, 2nd edn. MIT solution to the problem of validity generalization, J. Appl.
Press, Cambridge, MA. Psychol., 529-540.
Larson J.S., Adamus P.R., Clairain E.J., 1989. Functional Smith, V.K., Kaoru, Y., 1990. Signals or noise? Explaining the
assessment of freshwater wetlands: a manual and training variation in recreation benefit estimates. Am. J. Agric. Econ.
outline. University of Massachusetts, Amherst, MA 72 (2), 419 – 433.
Loomis, J.B., White, D.S., 1996. Economic benefits of rare and Smith, V.K., Osborne, L.L., 1996e. Do contingent valuation
endangered species: summary and meta-analysis. Ecol. estimates pass a ‘scope’ test? A meta analysis. J. Environ.
Econ. 18 (3), 197 – 206. Econ. Mgmt. 31, 287 – 301.
Mitsch, W.J., Gosselink, J.G., 1986. Wetlands. Van Nostrand White, K.J., 1997. SHAZAM User’s Reference Manual Version
Reinhold, New York. 8.0, McGraw-Hill, Vancouver.
.