Available via license: CC BY 4.0
Content may be subject to copyright.
Limited substitutability, relative price changes
and the uplifting of public natural capital values
Moritz A. Druppa,b, Zachary M. Turkc,d,
Ben Groome, Jonas Heckenhahnf*
aDepartment of Economics, University of Hamburg, Germany
bCESifo Munich, Germany
cDepartment of Geography and Environment,
London School of Economics and Political Science, UK
dLandscape Policy and Governance Team,
Manaaki Whenua - Landcare Research, New Zealand
eDepartment of Economics, University of Exeter, United Kingdom
fFaculty of Management and Economics, Ruhr-University Bochum, Germany
August 9, 2023
Abstract
As the global economy continues to grow, ecosystem services tend to stagnate or
degrow. Economic theory has shown how such shifts in relative scarcities can be re-
flected in the appraisal of public projects and environmental-economic accounting,
but empirical evidence has been lacking to put the theory into practice. To estimate
the relative price change in ecosystem services that can be used to make such ad-
justments, we perform a global meta-analysis of environmental valuation studies to
derive income elasticities of willingness to pay (WTP) for ecosystem services as a
proxy for the degree of limited substitutability. Based on 749 income-WTP pairs,
we estimate an income elasticity of WTP of around 0.78 (CI: 0.6 to 1.0). Combining
these results with a global data set on shifts in the relative scarcity of ecosystem ser-
vices, we estimate relative price change of ecosystem services of around 2.2 percent
per year. In an application to natural capital valuation of non-timber forest ecosys-
tem services by the World Bank, we show that their natural capital value should be
uplifted by more than 50 percent (CI: 32 to 78 percent), materially elevating the role
of public natural capital. We discuss implications for relative price adjustments in
policy appraisal and for improving estimates of comprehensive national accounts.
Keywords: Willingness to Pay; Ecosystem Services; Limited Substitutability; Growth;
Relative Prices; Contingent Valuation; Forests
JEL codes: D61, H43, Q51, Q54, Q58
*We thank Jasper Meya, Sjak Smulers, Daan van Soest and Martin Quaas as well as seminar audiences
at the World Bank, idiv Leipzig and EAERE 2023 for helpful discussions, and are grateful to Johanna
Darmstadt, Mark Lustig and Jasper R ¨
oder for excellent research assistance. We gratefully acknowledge
funding by the World Bank. M.D. additionally acknowledges support from the German Federal Ministry
of Education and Research (BMBF) under grant number 01UT2103B.
arXiv:2308.04400v1 [econ.GN] 8 Aug 2023
1 Introduction
Measuring economic progress towards sustainability requires addressing the limited
substitutability among the various constituents of comprehensive wealth (Smulders and
van Soest,2023). Potential limits to substitutability imply that society must strike a
balance between the two opposing paradigms of Weak and Strong Sustainability (e.g.,
Neumayer,2003;Hanley et al.,2015;Dasgupta,2021). Many contemporary measures
of economic progress and wealth have explicitly or implicitly followed a Weak Sustain-
ability approach. In doing so, they consider natural capital and ecosystem services as
largely substitutable—sometimes even perfectly substitutable—with human-made capi-
tal stocks. In light of the continued growth of human-made capital and the stagnation
or degradation of many natural capital stocks, the Weak Sustainability approach is in-
creasingly being called into question. From a theory perspective, we should consider
some degree of imperfect substitutability when estimating shadow prices. This is rele-
vant both for natural capital that serves as an intermediate input to various production
processes as well as public natural capital as a direct source of utility (see, e.g. Smul-
ders and van Soest,2023;Zhu et al.,2019). A common constraint to implementation,
however, has been a lack of sufficient empirical evidence on the limits of substitutability
of ecosystem services and natural capital that can inform the computation of shadow
prices (e.g., Cohen et al.,2019;Drupp,2018;Rouhi Rad et al.,2021).
This paper makes a step towards closing this important empirical evidence gap to en-
able respecting a limited degree of substitutability and the change in the relative scarcity
of ecosystem services in policy appraisal and environmental-economic accounting, by
focussing on the limited sustainability of ecosystem services in utility. To this end, we
present the largest global database to estimate the degree of limited substitutability of
ecosystem services vis-a-vis human-made goods, via the income elasticities of WTP for
ecosystem services, in order to compute relative price changes of ecosystem services.
We use these estimates of relative price changes to derive adjustments to natural capital
accounting, taking the assessment of forest ecosystem service values in the Changing
Wealth of Nations (CWON) as a case study.
There are two general approaches to dealing with limited substitutability of ecosys-
tem services in a dynamic context, for example in project appraisals with future costs
and benefits or the assessment of comprehensive wealth. We can either apply differenti-
ated discount rates – often a lower discount rate for non-market ecosystem services, or
account for increasing relative scarcity by adjusting our valuation of ecosystem services
over time (e.g. Baumg¨
artner et al.,2015;Drupp,2018;Gollier,2010;Hoel and Sterner,
2007;Traeger,2011;Weikard and Zhu,2005). Sterner and Persson (2008) have put a
spotlight on the detrimental effects of climate change on non-market ecosystem services,
such as the loss of biodiversity or environmental amenities. Sterner and Persson (2008)
and subsequent work (Drupp and H¨
ansel,2021;Bastien-Olvera and Moore,2021) has
studied how the increasing scarcity and limited substitutability of non-market ecosystem
services vis-`
a-vis manufactured goods affects optimal climate policy via good-specific
2
discount rates or relative price changes. Drupp and H ¨
ansel (2021), for instance, estimate
that relative prices of non-market goods increase by around 2 to 4 percent per year.
Incorporating this scale of adjustment leads to social cost of carbon estimates that are
more than 50 percent higher compared to the case where goods are assumed perfectly
substitutable. Accounting for relative price changes of non-market goods is thus crucial
to the appraisal of climate change policy. It is likewise relevant for project appraisal and
environmental-economic accounting such as in the System of Environmental Economic
Accounting-Experimental Ecosystem Accounting (SEEA-EEA) and in CWON.
Previous empirical studies have estimated relative price changes of non-market goods
by calculating the elasticity of substitution indirectly via the income elasticity of willig-
ness to pay (WTP) from non-market valuation studies (Baumg¨
artner et al.,2015;Drupp,
2018;Heckenhahn and Drupp,2022). The second key component are good-specific
growth rates, estimated on historical time series (Baumg¨
artner et al.,2015;Heckenhahn
and Drupp,2022) or as endogenous outcomes in global integrated climate-economy
assessment models (e.g. Drupp and H¨
ansel,2021). Relative price changes are then
approximated by the income elasticity of WTP multiplied by the difference in good-
specific growth rares. Baumg¨
artner et al. (2015) were the first to estimate relative price
changes. They applied national growth rates to arrive at country-level results. How-
ever, they assumed that the elasticity of substitution is constant across all countries and
ecosystem service types. Specifically, they derive the elasticity of substitution based on
a single meta-analysis by Jacobsen and Hanley (2009), who estimate an income elas-
ticity of WTP for global biodiversity conservation based on 46 contingent valuation
(CV) studies. Heckenhahn and Drupp (2022) provide the first comprehensive country-
specific evidence, estimating growth rates of 15 ecosystem services and the degree of
limited substitutability based on a meta-analysis of 36 German WTP studies. They find
that ecosystem services are complements to manufactured goods and that their relative
price has, on aggregate, increased by around 4 percent per year.
Most government appraisal and environmental-economic accounting has yet to ex-
plicitly address the limited substitutability of non-market goods (Groom et al.,2022).
Due to a lack of country-specific estimates of substitutability, global-level estimates of
relative price changes have recently been integrated into governmental policy guidance
(Groom and Hepburn,2017). For instance, The Netherlands consider relative price
changes of 1 percent per annum as part of their discounting guidance, and the UK
Department for Environment, Food and Rural Affairs (DEFRA) consider relative price
adjustments by ‘uplifting’ the damage costs of air pollution by 2 percent per year. Ad-
ditionally, guidelines by the Asian Development Bank and Canada suggest the use of
lower discount rates for environmental goods (Groom et al.,2022). The most recent
CWON report by the World Bank captures a few select non-market goods and capi-
tal stocks. Most prominently, forest ecosystem service values are featured based on a
meta-regression and benefit transfer analysis by Siikam¨
aki et al. (2015). Even though
Siikam¨
aki et al. (2015) find positive and large GDP elasticities of WTP, CWON does not
apply relative price adjustments in response to the increasing relative scarcity of forest
3
ecosystem service values. This occurs by assuming that per-hectare monetary values are
constant over time, solely adjusting for inflation. Given that a constant social discount
rate of 4 percent is also applied to forest ecosystem service values, such an approach
implicitly assumes that non-market environmental values do not increase with incomes
and thus follow a perfect substitutability assumption.
Against this background, we provide the first systematic global empirical evidence
basis to inform relative price adjustments of ecosystem services for application to CWON
and beyond. Our main focus is on improving the estimation of limited substitutability
of ecosystem services vis-a-vis human-made goods. To this end, we perform a meta-
analysis of environmental values derived by the contingent valuation (CV) approach to
estimating the income elasticity of WTP—a key parameter also for benefit transfer across
space (Baumg¨
artner et al.,2017;Smith,2023). This builds on a large-scale keyword-
based search strategy and an in-depth analysis of a large random sample of the known
population of peer-reviewed CV studies. The subsequent inputs to our analysis are
aggregate income and WTP estimates as well as recurring covariates from each study.
We find relative price changes of ecosystem services of around 2.2 percent per year
on aggregate. Relative price changes are slightly smaller (1.6 percent) for forest ecosys-
tem services, due to a lower rate of de-growth of forest area offset by a larger income
elasticity of WTP. These estimates can be employed to adjust WTP estimates for project
appraisal or environmental-economic accounting. In an application on natural cap-
ital valuation, taking the Changing Wealth of Nations (CWON) 2021 report by the
World Bank (2021) as a case study, we show that adjusting natural capital estimates for
non-timber ecosystem services for relative price changes results in uplifting the present
value over a 100 year time period by 52 percent (CI: 32 to 78 percent), materially ele-
vating the role of public natural capital. This echoes work on the importance of limited
substitutability in climate policy appraisal (Bastien-Olvera and Moore,2021;Drupp and
H¨
ansel,2021;Sterner and Persson,2008). We close by discussing the results of our
empirical analysis in context and summarizing insights for environmental-economic ap-
praisal and accounting.
2 Theoretical background
Well-being depends on the good and services that are derived in each period from
human-made and natural capital stocks. To provide the theoretical background for our
empirical analysis, we consider a simple model in which intertemporal well-being is
derived from both human-made goods, 𝐶𝑡and non-market environmental goods or
ecosystem services, 𝐸𝑡. In the general case of imperfect substitutability, ecosystem ser-
vices feature explicitly in the instantaneous utility function representing preferences
over market-traded human-made consumption goods and non-market goods, 𝑈(𝐶𝑡,𝐸𝑡).
4
A standard form of time-discounted Utilitarian social welfare function is given by:
𝑊=∫∞
𝑡=0
𝑈(𝐶𝑡,𝐸𝑡)𝑒−𝛿𝑡𝑑𝑡 . (1)
The theory of dual discounting or relative price changes has shown that there are
two approaches to addressing the intertemporal appraisal of non-market goods (e.g.,
Baumg¨
artner et al.,2015;Gollier,2010;Traeger,2011;Weikard and Zhu,2005):
1. Explicitly consider how the relative price of non-market goods vis-a-vis market-
traded consumption goods changes over time. Then, compute comprehensive con-
sumption equivalents at each point in time and use a single consumption discount
rate to on future comprehensive consumption equivalents.
2. Use differentiated, good-specific consumption discount rates. In the most general
application, one discount rate is used for manufactured consumption, 𝑟𝐶, and
another for non-market goods, 𝑟𝐸.
In the first approach, we compute the value of non-market goods in terms of the mar-
ket good numeraire. This is given by the marginal rate of substitution (MRS), 𝑈𝐸𝑡/𝑈𝐶𝑡,
which is the implicit price of non-market goods. The MRS tells us by how much the
consumption of market goods would need to increase in reply to a marginal decrease
in non-market goods to hold utility constant. The 𝑅𝑃𝐶𝑡measures the relative change
in the valuation of non-market goods – the change in the MRS between non-market
and market goods – and thus the change in their relative scarcity over time (Hoel and
Sterner,2007):
𝑅𝑃𝐶𝑡=
𝑑
𝑑𝑡 𝑈𝐸𝑡
𝑈𝐶𝑡!/ 𝑈𝐸𝑡
𝑈𝐶𝑡!. (2)
Future expected non-market values can then be adjusted using the 𝑅𝑃𝐶𝑡and a single
SDR can be used to discount future flows of private and non-market consumption.
In the second approach, we compute good-specific (dual) discount rates as:
𝑟𝐶𝑡=𝛿+𝜂𝐶𝐶𝑡𝑔𝐶𝑡+𝜂𝐶𝐸𝑡𝑔𝐸𝑡(3)
𝑟𝐸𝑡=𝛿+𝜂𝐸𝐸𝑡𝑔𝐸𝑡+𝜂𝐸𝐶𝑡𝑔𝐶𝑡(4)
where 𝑔𝐸and 𝑔𝐶are the growth rates, 𝜂𝐶𝐶𝑡(𝜂𝐸𝐸𝑡) the elasticity of marginal utility of
private-good (non-market good) consumption with respect to private-good (non-market
good) consumption, and 𝜂𝐶 𝐸𝑡(𝜂𝐸𝐶𝑡) denotes the cross-elasticity of marginal utility of
private-good (non-market good) consumption with respect to non-market good (private-
good) consumption (see, e.g., Baumg¨
artner et al.,2015). Expanding their applicability,
these dual rates can also be used in cases where non-market goods are not evaluated in
monetary units such as satellite accounts in national accounting and biophysical impact
assessments. It is important to stress that this approach also implies that we have to
adjust the ‘standard’ discount rate for private consumption with an addition to the
Simple Ramsey Rule by a substitutability effect (𝜂𝐶𝐸𝑡𝑔𝐸𝑡).
5
To make this more concrete, let us consider the constant-elasticity-of-substitution
(CES) utility function, capturing various degrees of substitutability or complementarity:
𝑈(𝐶𝑡,𝐸𝑡)=𝛼𝐶
𝜎−1
𝜎
𝑡+ (1−𝛼)𝐸
𝜎−1
𝜎
𝑡𝜎
𝜎−1, (5)
0< 𝜎 < +∞, is the constant elasticity of substitution between the two goods, and
0< 𝛼 < 1 is the utility share parameter for private consumption. The utility function
given by equation 5is strictly concave, represent homothetic preferences, and both the
private good, 𝐶𝑡, and non-market good, 𝐸𝑡, are normal. It turns out that with CES
preferences and imperfect complements, i.e. 𝜎 > 0, we get the following straightforward
equivalence between both approaches (Weikard and Zhu,2005):
𝑅𝑃𝐶𝑡=1
𝜎[𝑔𝐶𝑡−𝑔𝐸𝑡]=𝑟𝐶𝑡−𝑟𝐸𝑡. (6)
Accordingly, the choice of whether one adjusts the numerator via a relative price effect
adjustment or the denominator via the use of dual discount rates is not of theoretical
importance in intertemporal valuation exercises.
In the setting of CES preferences, Ebert (2003) has shown that the constant elasticity
of substitution between a human-made consumption goods and a non-market good is
directly and inversely related to the income elasticity of WTP, 𝜉, of the non-market good
(cf. Baumg¨
artner et al.,2017). We can accordingly write Equation 6as:
𝑅𝑃𝐶𝑡=𝜉[𝑔𝐶𝑡−𝑔𝐸𝑡]. (7)
Our empirical approach is subsequently designed to estimate the income elasticity
of WTP for a variety of applications. We estimate 𝜉at global and regional scales, for a
few nations where sufficient data is available, and by specific types of non-market good.
3 Empirical strategy
We build on previous work by Jacobsen and Hanley (2009), Heckenhahn and Drupp
(2022), Subroy et al. (2019), Richardson and Loomis (2009), and Barrio and Loureiro
(2010) to estimate income elasticities of WTP for ecosystem services based non-market
valuation studies in the academic literature. Our meta-analysis collects mean WTP and
income estimates at the valuation exercise scale. The resulting dataset is then used to es-
timate income elasticities of WTP and subsequently the elasticity of substitution between
ecosystem services and human-made goods (cf. Heckenhahn and Drupp,2022). In this
section we first discuss the meta-analysis, then the empirical strategy we implement to
arrive at estimates of 𝜉.
6
3.1 Meta-analysis of mean WTP-income value pairs
The data basis for this study is a meta-analysis of existing WTP studies. In the first
phase, we identify potentially relevant non-market valuation studies through a keyword-
based search string provided in Appendix A.1 based on the authors’ experience and
beta testing. To ensure better comparability of ecosystem service valuation estimates,
we focus our search on contingent valuation (CV) studies that were published in peer-
reviewed, English-language literature since the year 2000. The keyword-based search
results in a preliminary data set where each row is a peer-reviewed journal article in
which we expect to find relevant contingent valuation-based mean WTP estimates and
respondent income data.
The search string employed is intended to cast a wide net. That is, we expect to
drop several studies due to irrelevance and informational shortcomings. The data is
then evaluated using the exclusion criteria reported in Appendix A.2. After application
of the first exclusion criteria—including whether each article has been cited at least
once in SCOPUS—2,174 articles remain. The next exclusion criteria step is an abstract
screening to check whether the articles potentially report new, contingent valuation-
based WTP estimates at all. Strictly theoretical papers as well as reviews, secondary
source estimates, and benefits transfer approach (BTA)-focused papers are excluded
to avoid double-counting estimates. Naturally, whether we can access the articles is
important and rarely proved to be an issue. At this stage, 1,165 studies remain on which
to conduct a detailed screening and subsequently data harvesting.
From the data set of 1,165 candidate WTP studies, we selected a random sample of
100 studies as the basis to fine tune key steps in the screening and coding processes and
improve consistency between our two independent coders. Each paper is carefully scru-
tinized for appropriate WTP and income data. A recurring issue is that several papers
do not report whether income data is net of taxes or report gross income. We have sub-
sequently contacted each paper’s corresponding author in search of clarification. The
review of each paper and harvesting of relevant data was found to be a particularly
time-intensive process. However, we found it easier to first screen for the inclusion of
both mean WTP and mean income estimates—or the information necessary to derive
such estimates—before harvesting relevant data. We also found that there is an impor-
tant distinction between contingent valuation estimates presented on a timescale basis
versus per-use estimates. Namely, without data on frequency of use at the respondent
scale, per-use estimates are not comparable to estimates based on timescales, which is
why we chose to set them aside. We then further constrain our data set to peer-reviewed,
contingent valuation studies that survey respondents on values based on timescales such
as annually, monthly, etc and convert estimates to an annual scale.
Our main analysis subsequently builds on a random sample of studies surviving
our exclusion criteria and containing at least the minimum necessary information—at
least one mean WTP estimate and mean respondent income estimate. An unfortunate
but necessary result of our focus on comparability is a substantially reduced number
7
of studies contributing to the end result. Of 1,000 randomly selected studies of the
1,165 passing the first two rounds of screening, 351 studies containing 749 distinct WTP-
income pairs are of use. Table 1provides summary statistics of our sample.
Table 1: Prepared data set description
Variable Context Value
Countries represented Count 68
Continent Observations
North America 86
South America 34
Africa 33
Europe 244
Asia 345
Australia 7
Study year Mean (s.d.) 2010 (6.8)
Income Mean annual, 2020 USD (s.d.) 3,176 (3,421)
WTP Mean annual, 2020 USD (s.d.) 162 (538)
Survey sample size Mean (s.d.) 616 (834)
Respondent age Mean (s.d.) 43 (6.4)
Respondent household size Mean (s.d.) 4.2 (1.5)
Forest-relevant estimates Share of observations 0.31
Notes: s.d. is the standard deviation of the data referenced. Based on N=749
WTP-income pairs contained in 351 unique studies.
3.2 Estimation strategy
Our main result is based on a log-log specification of WTP and income values while
accounting for the structure of our data, and clustering standard errors at the study
level. The basic univariate version of estimation equation is thus given by
𝑙𝑛(𝑊𝑇𝑃𝑖𝑗)=𝛼+𝜉𝑙𝑛(𝐼𝑁 𝐶 𝑖 𝑗 ) + 𝜖𝑖(8)
We suspect a number of covariates might bias the estimated coefficient on income if
omitted. These variables would have a direct effect on WTP and as such should be in-
cluded in the model. For example, populations of different ages may have different pref-
erences impacting their WTP as their lifestyles—and interaction with nature—may differ
systematically. We might also expect that people value different types of ecosystem ser-
vices differently and that different nations—with individual and unique histories—may
8
systematically differ in their resulting levels of WTP. Importantly, the income (and WTP)
is not always consistently on a household level, but sometimes elicited at an individual
level. We therefore rely as a default on the multivariate estimate that contains controls
for these differences across estimates. The model specification then becomes:
𝑙𝑛(𝑊𝑇𝑃𝑖𝑗)=𝛼+𝜉𝑙𝑛(𝐼𝑁 𝐶 𝑖 𝑗 ) +
𝑛
Õ
𝑘=1
𝛽𝑘𝑥𝑖𝑗 +𝜖𝑖(9)
where Í𝑛
𝑘=1𝛽𝑘𝑥𝑖𝑗 is our list of 𝑛covariates. These include potentially relevant factors
about the survey environment and respondents (survey year, mean respondent, house-
hold size, nation or continent), respondent incomes (income and WTP per-person or
household, gross or net income), WTP terms (annual, monthly, repeated), and survey
methods (elicitation format, data collection method). We also test for different time peri-
ods (for example pre- and post-2010) and for differences between regulating and cultural
services as well as differences when estimates are relevant to forests through the use of
indicator variables. Interactions, for example between income and country indicators
can also be tests. We also prepared a set of indicator variables for ecosystem service
types based on the ecosystem services listed in the Millenium Ecosystem Assessment
MEA (2005) framework. These indicators are at the WTP estimate-level as some papers
report estimates specific to certain ecosystem service types.
We are interested in arriving at defensible estimates of 𝜉, including whether it differs
by region and ecosystem service type. Inclusion of covariates supports that effort. As
is common practice, we conduct a sensitivity analysis on our coefficient of interest by
estimating a large set of models with different variations of covariates included in our
main model in Equation 9. So, we estimate an 213 =8, 192 versions of our main log-log
specification.
As noted, we also estimate the income elasticity of the WTP based on important sub-
samples. These include regional, by broad categories of regulating and cultural services,
and pre- and post-2010. As study sample size also varies substantially—note a mean
sample size of 616 with a standard deviation of 834—the result of alternative observation
weights are compared. However, our preferred weighting approach is to apply the
square-root of the samples size used at the WTP estimate scale. This implies that we
put some weight on sample size but avoid the risk that a few studies with particularly
large sample sizes drive our result entirely. In an additional analysis, we investigate
numerous specifications that have been used in the literature so far to estimate income
elasticity of WTP: From simple OLS, to random effects (e.g. Jacobsen and Hanley,2009;
Heckenhahn and Drupp,2022) fixed effects, to clustered random regression as used here
with sample size, inverse of square root of sample size (cf. Subroy et al.,2019) to our
preferred main specification that uses the square root of sample size to weigh estimates.
9
3.3 Growth rates
We next assemble growth rates of ecosystem services to obtain a proxy for a global
measure of the shift in the relative scarcity of ecosystem services vis-a-vis human-made
goods. These estimates extend and update prior work by (Baumg¨
artner et al.,2015),
who found that ecosystem services have overall declined by half a percent in the last
decades. We focus on non-market (and non-rivalrous) ecosystem services, i.e. we do not
consider provisioning services but capture regulating and cultural services. In a first
step, we update the data sources employed by (Baumg¨
artner et al.,2015), notably: Forest
cover, Living Planet Index (LPI), and IUCN’s Red List Index (RLI). We complement
this with two additional measures for regulating services that capture highly salient
aspects of environmental quality: air quality regulation and climate regulation. We
proxy the former by the negative of changes in PM2.5 emissions, i.e. counting reductions
in emission as an improvement in air quality. We proxy for the latter with the change in
the 2C global mean temperature budget – the upper target of the UN Paris Agreement.
Table 2shows the individual components, units of measurement, and data sources.
Table 2: Components and data sources for estimates of growth rates
Component Unit of measurement Data source
Forest area Hectare WorldBank (2023)
Living Planet Index (LPI) Dimensionless Zoological Society of London,
and WWF 2022
Red List Index (RLI) Various IUCN RedList (2023),
based on Butchart et al. (2010)
Air quality Micrograms per m3WorldBank (2023)
(mean annual PM2.5)
Climate regulation Degrees Celsius NOAA (2023)
GDP per capita US dollars WorldBank (2023)
Within regulating (forest, LPI, RLI, PM2.5, temperature) and cultural services (for-
est, LPI, RLI) as well as aggregate ecosystem services we take the arithmetic mean of
relevant individual components. To calculate growth rates, we use the time span with
the longest comparable data across all indicators (1993 to 2016) and estimate exponen-
tial growth rates, including standard errors. We use the largest standard error of the
individual growth rate components – climate for regulating and aggregate services, and
the living planet index for cultural services – when aggregating standard errors. Akin
to estimating growth rates of ecosystem services, we also estimate the growth rate of
global GDP per capita. In contrast to (Baumg ¨
artner et al.,2015), we do not subtract
provisioning services, and measure economic growth including its standard error.1
1All time series show a clear trend over the time period, except for air quality, which deteriorates from
10
4 Results
We present here estimates of income elasticities of WTP, 𝜉, for ecosystem services glob-
ally as well as select regions. We also estimate 𝜉based on subcategories of ecosystem
services as well as different time frames. We subsequently couple the estimates of in-
come elasticities with estimates of good-specific growth rates to compute relative price
changes of ecosystem services.
4.1 Income elasticity of WTP for ecosystem services
We first estimate the income elasticity of WTP for aggregate ecosystem services on our
full sample via univariate regression (Equation 8) and by adding key controls (different
permutations of Equation 9). Our univariate result is an estimate of the income elasticity
of WTP for ecosystem services of 0.57 (95-CI: 0.31 to 0.82) , see Table 3. Adding controls
changes the estimate of the income elasticity of WTP to 0.78 (95-CI: 0.56 to 0.99). This
change is almost entirely attributable to the inclusion of an indicator of whether the in-
come measure is at the household or individual level. Respondent measures of income
potentially overlook the dynamics around household size or multiple streams of income
resulting in seemingly more elastic estimates of the income elasticity of willingness to
pay for ecosystem services. As such, we select our coefficient from multivariate estima-
tion as our main result. We develop a specification graph to investigate the sensitivity
of our estimate to various combinations of control variables. The result of 8,192 alter-
native specifications represented in Figure 5of Appendix A.4 and shows that our main
univariate estimate falls at the lower end of these alternative specifications.
Our main estimate maps into a mean value for the elasticity of substitutability be-
tween ecosystem services and human-made goods of 1.28 (95-CI: 1.00 to 1.79).
Table 3: Income elasticity of WTP for aggregate ecosystem services
Sample ln(INCOME) S.E. N Adj. R2
Univariate 0.57*** 0.13 749 0.88
With controls 0.78*** 0.11 749 0.89
Notes: Choice set of controls including the publication year, sample size, income
information (gross/net, individual/household), payment type and elicitation method.
Significance levels: * p<0.1, ** p<0.05, *** p<0.01
Estimates on subsets of aggregate ecosystem services allow us to investigate the
extent of heterogeneity. We consider different sub-types of ecosystem services, and po-
tential differences across continents and time frames. Table 4reports income elasticities
of WTP across different sub-types of ecosystem services: regulating and cultural ser-
vices as well as key sub-categories. Overall, we find little variation in income elasticities,
1990 to 2010 and improves again thereafter. We thus also redo the analysis of growth rates for the time
frame from 2010 to 2016.
11
noting that oftentimes projects valued in CV studies encompass contributions to mul-
tiple services. Only the estimate of the income elasticity for recreation and ecotourism
is slightly lower—the category closest to being rivalrous. We also split the sample into
forest and non-forest ecosystem services as this serves as a key input to our applica-
tion on natural capital accounting in the CWON framework in Section 5. We find that
the income elasticity of forest ecosystem services is slightly higher than the aggregate
estimate. We present the univariate and choice set of controls estimates alongside key
subgroups to be discussed in Figure 1.
Table 4: Heterogeneity of income elasticities of WTP across ecosystem service sub-types
ln(INCOME) S.E. N Adj. R2
Climate regulation 0.78** 0.33 160 0.93
Air quality regulation 0.78*** 0.14 248 0.92
Water regulation 0.84*** 0.14 278 0.89
Erosion regulation 0.84*** 0.12 193 0.86
Regulating Services 0.78*** 0.15 503 0.93
Spiritual and religious values 0.84*** 0.12 119 0.63
Aesthetic values 0.72*** 0.11 386 0.89
Recreation and ecotourism 0.65** 0.26 307 0.84
Biodiversity preservation 0.81*** 0.11 316 0.89
Cultural Services 0.73*** 0.14 464 0.86
Forest ecosystem services 0.81*** 0.12 233 0.86
Non-forest ecosystem services 0.77*** 0.16 516 0.89
Notes: Multivariate regressions. Significance levels: * p<0.1, ** p<0.05, *** p<0.01
We next divide our sample by the continent in which the hypothetical environmental
project to be undertaken was located, and report the results in Table 5. We note that
the estimates are mostly concentrated in Asia, followed by Europe, and relatively few
estimates from the rest of the world. In terms of income elasticities, we find lower
estimates for North America—as well as less consistency—versus the rest of the world.
Table 5: Heterogeneity of income elasticities of WTP across continents
ln(INCOME) S.E. N Adj. R2
North America 0.57* 0.30 86 0.95
South America 0.23 0.30 34 0.96
Africa 0.73*** 0.20 33 0.97
Europe 0.74*** 0.25 244 0.80
Asia 0.73*** 0.13 345 0.92
Notes: Multivariate regressions. Significance levels: * p<0.1, ** p<0.05, *** p<0.01
12
Figure 1: Point estimates of the elasticity of WTP from select models.
Univariate
Multivariate
Regulating svcs
Cultural svcs
Forests
0 .2 .4 .6 .8 1 1.2
Income elasticity of WTP
Notes: Estimates are the coefficients on 𝑙𝑛 (𝐼𝑁𝐶𝑂𝑀𝐸)from the main and control specifications in
Table 3as well as estimates based on subsets of observations on regulating services, and cultural
services, and forests. 95 percent confidence interval estimates are included around the point estimates.
We note that the broadest comparable meta-analysis on the income elasticity of WTP
(for biodiversity conservation only) was conducted by Jacobsen and Hanley (2009). Their
main result was an income elasticity of WTP estimate of 0.38, but published more than
a decade ago. It is therefore interesting to investigate how our estimate of the income
elasticity of WTP relates in a comparable time frame, as well as in comparison to the
most recent decade. In Table 5we break down the sample by sampling year. We con-
duct this analysis based on our univariate estimation strategy such that year and other
covariates potentially correlated with the two periods are not included. First, we con-
sider estimates from publications based on samples collected up to and including in
year 2010, and find a smaller income elasticity of 0.49. In contrast, the income elasticity
based on articles data collected from 2011 onwards is considerably higher, at 0.69. If
publication year is instead used, the spread in income elasticities is even larger at 0.17
versus 0.79 for pre- and post-2010 groups, respectively. This is a noteworthy increase in
the income elasticity of WTP in recent years. See Table 6.
13
Table 6: Heterogeneity of income elasticities of WTP across decades (publication year)
ln(INCOME) S.E. N Adj. R2
pre-2010 0.49** 0.20 377 0.82
2011-2021 0.69*** 0.12 372 0.91
Notes: Univariate regressions. Significance levels: * p<0.1, ** p<0.05, *** p<0.01
4.2 Growth rates
Table 7reports estimates on the growth rates of ecosystem service categories and their
standard errors. We present these alongside the growth rate of GDP per capita. Each
growth metric is estimated based on data for the years 1993 to 2016. We find substantial
heterogeneity in growth rates. The living planet index and climate regulation metrics
show the largest rates of de-growth, while the change in forest area and air quality
metrics show the lowest rates of change.2Our estimate of aggregate ecosystem service
change is -1.01 percent (CI: -1.34 to -0.68), while GDP per capita has increased by 1.82
percent (CI: 1.78 to 1.86) over the same period. This amounts to a sizable shift in the
relative scarcity of ecosystem services vis-a-vis human-made goods. Ecosystem services
have become relatively scarcer by 2.83 percent per year (best to worst case scenarios of
2.46 to 3.20 percent).
Table 7: Good-Specific Growth Rates
Indicator Growth rate (S.E.)
Forest area -0.11% (0.00%)
Living planet index -2.84% (0.06%)
Red list index -0.42% (0.01%)
Air quality (PM2.5) -0.16% (0.17%)
Climate regulation -1.50% (0.14%)
Aggregate Ecosystem Services -1.01% (0.17%)
GDP per capita 1.82% (0.02%)
4.3 Relative price changes of ecosystem services
We can now combine the two critical pieces of information – income elasticity and
growth rate estimates – to compute relative price changes (RPC) of ecosystem services.
Table 8reports our estimates of relative price changes both in the aggregate and for
different ecosystem service categories. Our central estimate for the relative price change
2Results are qualitatively similar when constraining the analysis to the most recent trend data, except
for air quality regulation which shows a positive development in the current trend data (2010 to 2016),
improving by 1.78% per year. In contrast, the decline rate for climate regulation is more strongly negative.
Overall, we find a somewhat smaller rate of de-growth of -0.73 percent for the time period 2010 to 2016.
14
of aggregate ecosystem services is 2.21 percent (CI: 1.83 to 2.60) That is, the value of
ecosystem services is increasing by around 2.2 percent per year relative to human-made
goods. This is 2.4 times (and potentially over 4.6 times) the estimate of 0.91 ±0.35
percent in Baumg¨
artner et al. (2015). The RPC estimate for regulating services is only
slightly higher than that for cultural ecosystem services, which is qualitative similar
to what Heckenhahn and Drupp (2022) find for a German case study. While the in-
come elasticity for forest ecosystem services is higher than for ecosystem services on
aggregate, the rate of decline of forest area is considerably smaller; in combination, the
relative price change of forest ecosystem services (1.57 percent) is smaller than that of
aggregate ecosystem services (2.21 percent).
Table 8: Relative Price Changes (RPC) of Ecosystem Services
Sample 𝜉=1/𝜎(S.E.) 𝑔𝐶−𝑔𝐸(S.E.) 𝑅𝑃𝐶 (C.I.)
Regulating Services 0.78 (0.15) 2.83% (0.17%) 2.21%
(1.64% to 3.00%)
Cultural Services 0.73 (0.14) 2.95% (0.09%) 2.15%
(1.89% to 2.42%)
Aggregate E.S. 0.78 (0.11) 2.83% (0.17%) 2.21%
(1.83% to 2.60%)
Forest E.S. 0.81 (0.12) 1.94% (0.07%) 1.57%
(1.40% to 1.74%)
Notes: RPC 95% confidence interval estimates based on 𝜉(𝑔𝐶−𝑔𝐸) ± 1.96 ×r𝑆.𝐸.(𝜉)
𝜉2+𝑆.𝐸.(𝑔𝐶−𝑔𝐸)
𝑔𝐶−𝑔𝐸2.
5 Application to Environmental-Economic Accounting
Relative price adjustments of ecosystem services are relevant for both environmental-
economic appraisal and accounting. Here, we explore implications for environmental-
economic accounting, considering the Changing Wealth of Nations (CWON) 2021 report
by the World Bank (2021) as a prominent case study.
CWON, like most measures of comprehensive wealth, only features natural capital
to a limited degree, with the bulk of natural capital components relating to fossil energy
resources and other provisioning services that are traded on markets. CWON, however,
also considers non-timber forest benefits as part of its natural capital accounting. Non-
timber forest benefits are currently estimated to be around 12 percent of the total value
of natural capital World Bank (2021). Non-timber ecosystem service values in the year
2018, in WTP per hectare, are based on a meta-regression analysis based on 270 estimates
from non-market valuation studies of non-timber forest benefits by (Siikam¨
aki et al.,
2021). Per-hectare values are assumed to be constant over time and only adjusted for
15
inflation by using country-specific GDP deflators (World Bank,2021). The capitalized
value of non-timber ecosystem services is calculated as the present value of annual
services, discounted over a 100 year time horizon at a constant discount rate of 4 percent.
This implies that no adjustment for relative price changes is factored in despite forest
de-growth, particularly in comparison to GDP per capita. Implicitly, this carries the
assumption that WTP does not increase with income and—in the setting of our model—
that ecosystem services are considered perfect substitutes to human-made goods.3
We perform relative price change adjustments to the yearly WTP estimates for non-
timber forest ecosystem services based on our estimates and compare the adjusted nat-
ural capital value to the current CWON approach. We show the results in Figure 2,
which depicts the estimated increase in CWON’s non-timber forest natural capital value
(in %), relative the CWON’s current estimate, as a function of the degree of comple-
mentarity between forest ecosystem services and human-made goods, measured by the
income elasticity of WTP for forest ecosystem services. For instance, Cobb-Douglas
substitutability, as the knife-edge case between substitutability and complementarity
(𝜎=𝜉=1), would imply uplifting the public natural capital value of non-timber forest
ecosystem services by 72 percent. In comparison, a prominent assumption in the in-
tegrated assessment modelling literature by Sterner and Persson (2008) of an elasticity
of substitution of 0.5, and thus a degree of complementarity of 2, would translate into
uplifting the public natural capital value by 270 percent.
For our central estimate of relative price changes of forest ecosystem services, we
find that the value of non-timber forest natural capital should be uplifted by 52 per-
cent, with a 95 percentile confidence interval around the central estimate of the income
elasticity resulting in a range of uplift-factors of 32 to 78 percent (see Figure 2). For com-
parison, using the relative price change of aggregate ecosystem services, which builds
on a slightly lower income elasticity but a much larger difference in growth rates, yields
a central uplift-factor for public natural capital of 88 percent.
Considering the limited degree of substitutability and the shifts in relative scarcity
by performing relative price adjustment of the annual WTP values in computing the
natural capital value of non-timber forest services makes a material difference to natural
capital accounting in CWON. The 52 percent increase in non-timber forest value would
lead to an increase of the overall natural capital value in CWON of around 6 percent (CI:
3.8 to 8.8 percent). In such a revised natural capital value, the adjusted non-timber forest
value would have an elevated contribution of around 17 percent (CI: 15 to 19 percent)
instead of the current 12 percent.
3Siikam¨
aki et al. (2021) report positive and significant GDP elasticities of WTP for recreation and habi-
tat/species conservation, for instance, but these are not considered in the CWON natural capital valuation.
16
Figure 2: Increase in public natural capital value along the degree of complementarity.
32%
52%
78%
050 100 150
Increase in non-wood forest natural capital value (in %)
0.25 .5 .75 11.25 1.5
Degree of complementarity (income elasticity of WTP)
Notes: The red line in this figure shows the estimated increase in The Changin Wealth of Nations’
(CWON) non-timber forest natural capital value (in %), relative the CWON’s current estimate, as a
function of the degree of complementarity between forest ecosystem services and human-made goods,
measured by the income elasticity of WTP for forest ecosystem services. The vertical black line
indicates the central estimate of the income elasticity of WTP for forest ecosystem services while the
grey-shaded area indicates its 95 percentile confidence interval. Horizontal, dashed helplines indicate
the corresponding increase in the public natural capital value (in percent).
6 Discussion
We discuss our empirical approaches and assumptions in turn, focusing on the estima-
tion of the income elasticity of WTP, as a proxy for the degree of limited substitutability,
and the aggregation of ecosystem services and the computation of growth rates.
We identify the degree of complementarity via the income elasticity of WTP for
ecosystem services based on a meta-analysis of the peer-reviewed literature. For this, we
consider 749 unique (mean) income-WTP pairs across studies and geographical contexts
as well as across a 20 year time frame. A number of remarks are in order.
First, our analysis is subject to concerns on the underlying data quality of contingent
valuation studies, including hypothetical bias etc., which has been discussed at length
in the literature (e.g., Kling et al.,2012). Schl¨
apfer (2008), for instance, argues that (too)
small income effects in contingent valuation studies may be an artefact of anchoring
biases, but we are not aware of a clear empirical test of this hypothesis. If this were
17
the case, we might capture too small income elasticities and thus undererstimate the
degree of complementarity. This implies that our estimates of the appropriate upward-
adjustment of natural capital values may represent a conservative estimate.
Second, besides contingent valuation specific concerns, our approach to identifying
the (aggregate) income elasticity of WTP—while building on the state of the art in the
literature—is very coarse, and rests on a very heterogeneous, imbalanced panel. Broadly
speaking, our new sample contains studies that reflect both methodological refinements
that have been introduced over time that have arguably deflated WTP estimates Barrio
and Loureiro (2010), and an increasing share of studies from Asia and lower-income
countries over time. Ideally, we’d like to identify the income elasticity of WTP based
on a sample that is not subject to methodological revisions or major changes in its geo-
graphical composition. While a few test-retest investigations exist that draw repeatedly
from the same sample (see Skourtos et al.,2010, for an overview), these typically concern
shorter time frames and have not been designed to investigate income effects. Evidence
to date suggests that mean WTP estimates are relatively constant over time frames of up
to 5 years, but that this is not the case for longer time frames (Skourtos et al.,2010).
Third, our approach of relying on a direct relationship between the income elasticity
of WTP and the elasticity of substitution holds under a very common but still very
specific assumption on preferences, specifically that preferences are represented by a
constant-elasticity-of-substitution (CES) utility function (e.g., Ebert,2003;Baumg¨
artner
et al.,2017). We are not aware of systematic studies trying to test the relative goodness-
of-fit of CES versus other utility specifications, but note that extensions exist in the
applied theoretical literature. One interesting case is an extension of preferences that
consider critical thresholds in the form of subsistence needs (Baumg¨
artner et al.,2017;
Drupp,2018;Heal,2009). If there exists some critical level of ecosystem services, 𝐸>0,
then the degree of substitutability becomes endogenous to the level of the ecosystem
service over and above the critical level, and the equation for the relative price change
of ecosystem services is adjusted to (cf. Drupp,2018):4
𝑅𝑃𝐶𝑡=𝜉𝑔𝐶𝑡−𝑔𝐸𝑡
𝐸𝑡
𝐸𝑡−𝐸. (10)
Such an extension implies slightly higher relative price changes that increase substan-
tially as one gets close to the critical basic need threshold (Drupp,2018). It would lead
to an upward revision of the natural capital values adjustment discussed in Section 5.
Finally, we assume that preferences elicited primarily on small scale projects aimed
at improving ecosystem service conditions scale up to the global level. However, ser-
vices may be perceived as complements (substitutes) at the local level, but as substitutes
(complements) at a global scale. This issue may be more pronounced when the focus is
4WTP estimates are typically assumed to be a function of the ecosystem service level themselves
(Baumg¨
artner et al.,2017). Empirical evidence, however, is mixed—Barrio and Loureiro (2010) and oth-
ers find, for instance, that WTPs decrease with forest cover, while Taye et al. (2021) find that WTPs increase
with forest cover—as it’s often challenging to isolate the pure effect of the level of the ecosystem service.
18
relatively more on local public goods as compared to global public goods. We cannot
directly test for this, but a comparison of the income elasticity of WTP for recreational
services versus other services may serve as a proxy for this idea. Indeed, we find that
the income elasticity for recreational services is smaller than the estimate for the other
ecosystem services, but also that there is more variation around the income elasticity of
WTP for recreational services.
We have updated and extended the “Herculean task” (Baumg¨
artner et al.,2015, p. 278)
of assembling a proxy for the aggregate growth rates of ecosystem services. There exists
no accepted standard for how to aggregate various measures of environmental quality,
and also the data sources we draw on have to be considered imperfect proxies them-
selves. We have followed Baumg¨
artner et al. (2015) in using the unweighted arithmetic
mean of the growth rates for the different types of ecosystem services. This assumes
that the elasticity of substitution between different ecosystem services is equal to one
(Cobb-Douglas), which implies that WTPs would be the same for all types of ecosystem
services if their quantities were similar, an assumption we cannot properly test. We
note that there are other conceivable means of aggregation, using different weightings
to different degrees of substitutability. We leave a systematic exploration of this issue
to future work; the same holds for exploring the role of uncertainty around projecting
past growth estimates into the future (Gollier,2010) as well as the potential convergence
of ecosystem service and human-made goods growth rates, as the scarcity and limited
substituability of ecosystem services as intermediate inputs to production may manifest
itself as a drag on growth (Zhu et al.,2019).
7 Conclusion
We present the largest global database to estimate the degree of complementarity of
ecosystem services vis-a-vis human-made goods, via the income elasticities of WTP for
ecosystem services, in order to compute relative price changes of ecosystem services.
We find relative price changes of ecosystem services of around 2.2 percent per year on
aggregate. Relative price changes are smaller (1.57 percent) for forest ecosystem services
as these show a slower rate of de-growth as compared to other ecosystem service com-
ponents. These estimates can be employed to adjust WTP estimates for project appraisal
or environmental-economic accounting. In an application on natural capital valuation,
taking the Changing Wealth of Nations (CWON) 2021 report by the World Bank (2021)
as a case study, we show that adjusting natural capital estimates for non-timber ecosys-
tem services for relative price changes results in uplifting the present value over a 100
year time period by around 50 percent (CI: 32 to 78 percent), materially elevating the role
of public natural capital. This echoes work on the importance of limited substitutability
in climate policy appraisal (Bastien-Olvera and Moore,2021;Drupp and H¨
ansel,2021;
Sterner and Persson,2008).
The adjustment techniques we present are generally applicable for environmental-
economic appraisal and accounting, while the specific numerical inputs, such as on
19
growth rates, need to be adjusted on a case-by-case basis. Our results suggest that the
case for making relative price adjustments is reasonably robust and that more countries
and institutions than present (Groom et al.,2022) may consider making such adjust-
ments to correct the current mis-valuation of non-market goods in appraisal and of
public natural capital values in (comprehensive) wealth accounting.
20
References
Barrio, M. and M. L. Loureiro (2010). A meta-analysis of contingent valuation forest
studies. Ecological Economics 69(5), 1023–1030.
Bastien-Olvera, B. A. and F. C. Moore (2021). Use and non-use value of nature and the
social cost of carbon. Nature Sustainability 4(2), 101–108.
Baumg¨
artner, S., M. A. Drupp, J. N. Meya, J. M. Munz, and M. F. Quaas (2017). In-
come inequality and willingness to pay for environmental public goods. Journal of
Environmental Economics and Management 85, 35–61.
Baumg¨
artner, S., M. A. Drupp, and M. F. Quaas (2017). Subsistence, substitutability and
sustainability in consumption. Environmental and Resource Economics 67(1), 47–66.
Baumg¨
artner, S., A. M. Klein, D. Thiel, and K. Winkler (2015). Ramsey discounting of
ecosystem services. Environmental and Resource Economics 61(2), 273–296.
Butchart, S. H., M. Walpole, B. Collen, A. Van Strien, J. P. Scharlemann, R. E. Almond,
J. E. Baillie, B. Bomhard, C. Brown, J. Bruno, et al. (2010). Global biodiversity: indica-
tors of recent declines. Science 328(5982), 1164–1168.
Cohen, F., C. J. Hepburn, and A. Teytelboym (2019). Is natural capital really substi-
tutable? Annual Review of Environment and Resources 44, 425–448.
Dasgupta, P. (2021). The economics of biodiversity and ecosystem services: The das-
gupta review.
Drupp, M. A. (2018). Limits to substitution between ecosystem services and manu-
factured goods and implications for social discounting. Environmental and Resource
Economics 69(1), 135–158.
Drupp, M. A. and M. C. H¨
ansel (2021). Relative prices and climate policy: How the
scarcity of nonmarket goods drives policy evaluation. American Economic Journal: Eco-
nomic Policy 13(1), 168–201.
Ebert, U. (2003). Environmental goods and the distribution of income. Environmental
and Resource Economics 25(4), 435–459.
Gollier, C. (2010). Ecological discounting. Journal of economic theory 145(2), 812–829.
Groom, B., M. Drupp, M. C. Freeman, and F. Nesje (2022). The future, now: A review
of social discounting. Annual Review of Resource Economics.
Groom, B. and C. Hepburn (2017). Reflections—looking back at social discounting pol-
icy: the influence of papers, presentations, political preconditions, and personalities.
Review of Environmental Economics and Policy 11(2), 336–356.
21
Hanley, N., L. Dupuy, and E. McLaughlin (2015). Genuine savings and sustainability.
Journal of Economic Surveys 29(4), 779–806.
Heal, G. (2009). Climate economics: a meta-review and some suggestions for future
research. Review of Environmental economics and Policy.
Heckenhahn, J. and M. A. Drupp (2022). Relative price changes of ecosystem services:
Evidence from germany. CESifo Working Paper (No. 9656).
Hoel, M. and T. Sterner (2007). Discounting and relative prices. Climatic Change 84(3),
265–280.
Jacobsen, J. B. and N. Hanley (2009). Are there income effects on global willingness to
pay for biodiversity conservation? Environmental and Resource Economics 43(2), 137–
160.
Kling, C. L., D. J. Phaneuf, and J. Zhao (2012). From exxon to bp: Has some number
become better than no number? Journal of Economic Perspectives 26(4), 3–26.
MEA (2005). Millennium ecosystem assessment. Millennium Ecosystem Assessment.
Neumayer, E. (2003). Weak versus strong sustainability: exploring the limits of two opposing
paradigms. Edward Elgar Publishing.
Richardson, L. and J. Loomis (2009). The total economic value of threatened, endangered
and rare species: an updated meta-analysis. Ecological economics 68(5), 1535–1548.
Rouhi Rad, M., W. Adamowicz, A. Entem, E. P. Fenichel, and P. Lloyd-Smith (2021).
Complementarity (not substitution) between natural and produced capital: Evidence
from the panama canal expansion. Journal of the Association of Environmental and Re-
source Economists 8(6), 1115–1146.
Schl¨
apfer, F. (2008). Contingent valuation: a new perspective. Ecological economics 64(4),
729–740.
Siikam¨
aki, J., M. Piaggio, N. da Silva, I. ´
Alvarez, and Z. Chu (2021). Global assessment
of non-wood forest ecosystem services: A revision of a spatially explicit meta-analysis
and benefit transfer.
Siikam¨
aki, J., F. J. Santiago- ´
Avila, and P. Vail (2015). Global assessment of non-wood
forest ecosystem services. Spatially Explicit Meta-Analysis and Benefit Transfer to Improve
the World Bank’s Forest Weatlh Database, 1–97.
Skourtos, M., A. Kontogianni, and P. Harrison (2010). Reviewing the dynamics of eco-
nomic values and preferences for ecosystem goods and services. Biodiversity and con-
servation 19(10), 2855–2872.
22
Smith, V. K. (2023). Accounting for income inequality in benefit transfers: The impor-
tance of the income elasticity of wtp. Journal of Environmental Economics and Manage-
ment, 102781.
Smulders, S. and D. van Soest (2023). Natural capital substitution: Implications for
growth, shadow prices, and natural capital accounting. Mimeo, Tilburg University.
Sterner, T. and U. M. Persson (2008). An even sterner review: Introducing relative prices
into the discounting debate. Review of Environmental Economics and Policy.
Subroy, V., A. Gunawardena, M. Polyakov, R. Pandit, and D. J. Pannell (2019). The
worth of wildlife: A meta-analysis of global non-market values of threatened species.
Ecological Economics 164, 106374.
Taye, F. A., M. V. Folkersen, C. M. Fleming, A. Buckwell, B. Mackey, K. Diwakar, D. Le,
S. Hasan, and C. Saint Ange (2021). The economic values of global forest ecosystem
services: A meta-analysis. Ecological Economics 189, 107145.
Traeger, C. P. (2011). Sustainability, limited substitutability, and non-constant social
discount rates. Journal of Environmental Economics and Management 62(2), 215–228.
Weikard, H.-P. and X. Zhu (2005). Discounting and environmental quality: When should
dual rates be used? Economic Modelling 22(5), 868–878.
World Bank, T. (2021). The Changing Wealth of Nations 2021: Managing Assets for the Future.
The World Bank.
Zhu, X., S. Smulders, and A. de Zeeuw (2019). Discounting in the presence of scarce
ecosystem services. Journal of Environmental Economics and Management 98, 102–272.
23
Appendix A Selection of relevant valuation studies
A.1 Search string
Our focus is on values for regulating ecosystem services and cultural ecosystem ser-
vices (not provisioning services) that have been elicited using the contingent valuation
method. The search string has three components (1) focus on ecosystem services, (2)
focus on WTP estimates, (3) focus on the contingent valuation method.
( TITLE-ABS-KEY ( environment* OR natur* OR ecosystem OR biodiversity OR bi-
ologic* OR ecologic* OR habitat* OR forest* OR species OR protected OR conserv* OR
endangered OR ”national park*” OR landscape* OR terrestrial OR pollination OR tree*
OR tropic* OR vegetation OR peatland* OR grassland* OR dryland* OR pastoral OR
soil OR animal* OR bird* OR wild* OR air OR water OR aquatic OR marine OR coast*
OR water* OR fish* OR wetland* OR mangrove* OR reef* OR marsh* OR floodplain*
OR river* OR climate OR storm* OR erosion OR pest* OR hazard* OR recreat* OR
touris* OR “urban green” OR sacred OR spirit* OR sanctuary OR “natural heritage” OR
aesthetic*)
AND TITLE-ABS-KEY ( wtp OR willingness-to-pay OR ”willingness to pay*” OR
”willing to pay*” OR ”shadow price*” OR ”shadow value*” OR ”implicit price*” OR
”implicit value*”)
AND TITLE-ABS-KEY ( ”contingent valuation*” OR cvm OR ”contingent choice*”) )
AND ( LIMIT-TO ( SRCTYPE , ”j” ) ) AND ( LIMIT-TO ( DOCTYPE , ”ar” ) ) AND
( LIMIT-TO ( PUBYEAR , 2021 ) OR LIMIT-TO ( PUBYEAR , 2020 ) OR LIMIT-TO (
PUBYEAR , 2019 ) OR LIMIT-TO ( PUBYEAR , 2018 ) OR LIMIT-TO ( PUBYEAR , 2017
) OR LIMIT-TO ( PUBYEAR , 2016 ) OR LIMIT-TO ( PUBYEAR , 2015 ) OR LIMIT-TO (
PUBYEAR , 2014 ) OR LIMIT-TO ( PUBYEAR , 2013 ) OR LIMIT-TO ( PUBYEAR , 2012
) OR LIMIT-TO ( PUBYEAR , 2011 ) OR LIMIT-TO ( PUBYEAR , 2010 ) OR LIMIT-TO (
PUBYEAR , 2009 ) OR LIMIT-TO ( PUBYEAR , 2008 ) OR LIMIT-TO ( PUBYEAR , 2007
) OR LIMIT-TO ( PUBYEAR , 2006 ) OR LIMIT-TO ( PUBYEAR , 2005 ) OR LIMIT-TO (
PUBYEAR , 2004 ) OR LIMIT-TO ( PUBYEAR , 2003 ) OR LIMIT-TO ( PUBYEAR , 2002 )
OR LIMIT-TO ( PUBYEAR , 2001 ) OR LIMIT-TO ( PUBYEAR , 2000 ) ) AND ( LIMIT-TO
( LANGUAGE , ”English” ) )
A.2 Exclusion criteria
1. Citations: We excluded all studies that had not been cited (in SCOPUS).
2. Abstract screening: We excluded non-topical publications based on abstract-screening
that do not report new primary WTP estimates. Specifically, we excluded: Theory,
reviews, comments, non-primary valuation (such as benefit transfer), as well as
WTPs for non-environmental goods, WTPs for provisioning services, WTPs de-
rived from valuation approaches other than CV
3. PDFs obtainable: We excluded studies where we could not access the PDFs.
24
4. Paper screening: We excluded non-topical publications based on abstract-screening
that do not report new primary WTP estimates. Specifically, we excluded: Theory,
reviews, comments, non-primary valuation (such as benefit transfer), as well as
WTPs for non-environmental goods, WTPs for provisioning services, WTPs de-
rived from valuation approaches other than CV PLUS XYZ
5. WTP data screening: We excluded papers that reported a median WTP instead of
mean WTP.
A.3 Inflation and currency conversion
All monetary values were converted to 2020 US Dollar by first inflating the respective
national consumer price index and then applying purchasing-power-parity (PPP) con-
version. The relevant year for the inflation of the values was the year of study data
collection. When the authors did not provide the study year, we estimate the average
lag between study and publication years based on the studies where both pieces of in-
formation is available. The difference is approximately 4.0 years on average. We use this
to estimate the study year when missing.
A.4 Alternative specification results
This section presents a specification graph that suggests the robustness of our results
to the inclusion or exclusion of covariates. We also present results based on alternative
statistical models to suggest the robust of our results to model selection.
25
Figure 3: Income elasticity of WTP estimates based on alternative model specifications.
.2
.3
.4
.5
.6
.7
.8
.9
1
1.1
1.2
Income elasticity of WTP
Main specification Point estimates 95% CI
Notes: Estimates are the result of 213 =8, 192 alternative specifications of Equation 9. The main
specification is based on Equation 9which is at the 59.4th percentile ranking of our income elasticity
coefficient estimates from smallest to largest. The 95 percent confidence interval estimates are included
and results are plotted from smallest (0.43) to largest (0.84) coefficient estimate on 𝑙𝑛(𝐼𝑁𝐶𝑂 𝑀 𝐸).
26
Figure 4: Income elasticity of WTP estimates based on alternative statistical models.
weighted FE
unweighted FE
unweighted RE
unweighted OLS
weighted OLS
0 .2 .4 .6 .8 1 1.2
Income elasticity of WTP
Notes: The main result is based on a fixed-effects (FE) model at the study level and weighted by the
square root of the sample size. Some frequent alternatives to this approach include unweighted fixed
effects and random effects models and weighted and unweighted OLS estimates. While a Hausman
test suggests FE model is most appropriate, we provide these alternative estimates.
27
Figure 5: Income elasticity of WTP estimates by weight selection.
sqrt(sample size)
sample size
1/sqrt(sample size)
1/sample size
0 .2 .4 .6 .8 1 1.2
Income elasticity of WTP
Notes: The main result is derived with weights based on the square root of the sample size. Some
alternatives that are more or less reasonable are to use the sample size, inverse of the sample size, and
inverse of the square root of the sample size. Inverse sample sizes will tend to place more weight on
studies with smaller sample sizes and squared sample size weights will tend to bias estimates toward
studies with substantially larger samples.
28
