| Front page | | Contents | | Previous | | Next |

Practical tools for value transfer in Denmark – guidelines and an example

2 National and international value transfer: Methodological issues and validity

2.1 Value transfer methods

There are two main approaches to value transfer (Navrud 2004):

1.     Unit Value Transfer

        i)  Simple unit transfer

ii) Unit Transfer with income adjustments

2.     Function Transfer

i)  Benefit Function Transfer

ii) Meta-analysis

2.2 Unit value transfer

Simple (or naïve) unit transfer assumes that the well being experienced by an average individual at the study site is the same as will be experienced by the average individual at the policy site. Thus, we can directly transfer the benefit estimate, often expressed as mean willingness-to-pay (WTP) per household per year.

For the past few decades this procedure has routinely been used in the United States to estimate the recreational benefits associated with multipurpose reservoir developments and forest management. The selection of these unit values could be based on estimates from only one or a few valuation studies considered to be close to the policy site (both geographically and in terms of the good valued), or based on an average WTP estimate from literature reviews of many studies. Walsh et al. (1992, table 1) present a summary of unit values of a day spent in various recreational activities, obtained from 287 Contingent Valuation (CV) and Travel Cost (TC) estimates. Rosenberger and Loomis (2001) updated this for the USDA Forest Service and provide an annotated bibliography of studies from 1967 to 1998 (covering 21 recreational activities plus a category for wilderness recreation) with a total of 163 individual studies referenced, providing 760 benefit measures. They also provide value transfer guidelines and a decision tree to determine how to obtain values for recreational activities.

The obvious problem with this transfer of unit values for recreational activities is that individuals at the policy site may not value recreational activities the same as the average individual at the study sites. There are two principal reasons for this difference. First, people at the policy site might be different from individuals at the study sites in terms of income, education, religion, ethnic group or other socio-economic characteristics that affect their demand for recreation. Second, even if individuals´ preferences for recreation at the policy and study sites were the same, the recreational opportunities (substitute sites and activities) might not be. Unit values for other use values, e.g. groundwater and surface water for drinking, should be based on WTP/household/year (as this is often paid as an annual water fee, which is also the payment vehicle used in many SP studies).

Unit values for non-use values[1] of e.g. ecosystems from CV studies might be even more difficult to transfer than recreational (use) values for at least two reasons. First, the unit of transfer is more difficult to define. While the obvious choice of unit for use values (typically from Travel Cost studies or Contingent Valuation surveys) is consumer surplus (CS)[2] per activity day (defined as one individual performing the activity for some period of time during one day), there is greater variability in reporting non-use values from CV (and other Stated Preference (SP) methods), both in terms of WTP for whom, and for what time period. WTP is reported as per household or per individual, and as a one-time payment, annually for a limited time period, annually for an indefinite time, or even monthly payments. Second, the WTP is reported for one or more specified discrete changes in environmental quality, and not on a marginal basis. Therefore, the magnitude of the change at the study and policy sites should be close, in order to get valid transfers of estimates of mean, annual WTP per household. Also the initial levels of environmental quality should be close if one should expect non-linearity in the benefit estimate or underlying physical impacts. Since WTP to avoid a loss in environmental quality typically would be valued higher than a similar gain in environmental quality due to loss aversion (and there is no general estimate of the difference in value), the direction of environmental change should be the same at the policy and study sites. If the selected study for transfer values a gain while the policy question involves a loss of similar magnitude, we could still use the transferred estimate as a lower estimate. In the opposite case, we know that the transferred estimate is an overestimate, but unfortunately not by how much.

Mean WTP is the correct welfare measure, but since estimated mean WTP from dichotomous choice (DC) contingent valuation is very sensitive to distributional assumptions, median WTP could be used in a sensitivity analysis as a lower, conservative estimate. Mean WTP from open-ended (OE) questions should be corrected for protest zero answers before being transferred to avoid underestimation of WTP.

Since CV studies value discrete rather than marginal changes, attempts have been made to construct unit value expressed e.g. pr ha of an ecosystem. The results from a CV survey in terms of WTP/household/year would then be multiplied by the number of “affected” households, and then divided by the e.g. area of the ecosystem preserved. The size of the area in question could be calculated manually or using GIS; see e.g. the approach developed by the EcoValue project at the University of Vermont http://ecovalue.uvm.edu. In Choice Experiment (CE) studies similar marginal values are calculated more directly from the CE model based on the choices respondents make, but respondents still choose between discrete changes in the environmental good.  In addition to the uncertainty in conversion from discrete to marginal values, there is a need to develop ways to adjust for differences in population density when transferring unit values pr. ha across sites. Therefore, at the present, unit transfers in terms of WTP/household/year for a specified change in environmental quality seem to be the only defensible way to transfer non-use values.

Although some economists have raised awareness of the need to pay attention to the spatial and ecological characteristics of sites in relation to transfers (Bateman et al 2002; Eade & Moran 1996; Lovett et al 1997; Ruijgrok 2001), practitioners in the field have not yet effectively standardized the decomposition of transfers into spatially homogeneous units, which are widely recognized as being similar at different locations. Since ecologists have developed such classifications (i.e., land cover types), it is useful to explore whether it is possible to determine the economic values for the ecological goods and services provided by similar ecosystem types, and then transfer those values from one location to another using basic ecological principles (de Groot et al. 2002; Farber et al. 2002) [3]. The valuation step is often a part of a damage function approach (DFA; see figure 1). Therefore, a linkage has to be developed between the units of the endpoints of dose-response functions, and the unit of the economic estimates. This has been done successfully for e.g. visibility changes at US national parks (measured as percentage change in miles of visibility; see Smith and Osborne 1996[4]), health impacts (Alberini et al. 2004, Ready et al. 2004, EC DG Environment 2000), and annoyance from road transportation noise (Navrud 2002; EC DG Environment 2002).

Figure 1. Damage Function Approach (DFA) applied to water quality

Figure 1. Damage Function Approach (DFA) applied to water quality

For water quality a co-ordinated international classification of eutrophication of surface water taking into consideration different natural conditions and based on the suitability for drinking water and different recreational activities is now being developed. However, contingent valuation studies of national classifications of eutrophication levels in terms of the four levels: ”Drinkable”, ”Swimmable”, Fishable”, and ”Boatable” have already been conducted in e.g. Norway, Germany, UK, Poland and the Czech Republic; see appendix I for the classification used in CV-studies in Norway. These classifications are based directly on the levels of biological and chemical oxygen demand (BOD and COD). These levels correspond to the endpoints (resulting maximum and minimum levels of BOD and COD) when dose-response functions between increased concentration of nitrates and phosphates in the water bodies from increased depositions of these nutrients, and the resulting level of BOD and COD, are applied for a relevant range of deposition levels.

A similar type, five-level classification system has been developed for general ecological status of water quality within the EU Water Framework Directive (WFD). The concept of ”ecological status" and its five levels have not been defined in detail yet, but it is clear that it represents a wider set of parameters than the chemical and biological measures of water quality (e.g. euthrophication indices like Biological and Chemical Oxygen Demand). Recently, Danmarks Miljøundersøgelser has developed a similar five-level system for the state of terrestrial ecosystems. If these classification systems for the state of aquatic and terrestrial ecosystems can be linked to pollutants and other environmental pressures through dose-response functions (or expert assessments), and people understand the differences between the levels in the classification system, they can be used in Stated Preference studies and value transfer exercises in CBAs of measures to achieve improved ecological status. Hanley and Wright (2003) present an application to aquatic ecosystems using this classification system. They performed a Choice Experiment to value improvements to the ecology of two rivers in the UK, chosen as broadly representative of the kind of water bodies in the UK where moderate improvements in water quality are likely to be needed in order to meet ”Good Ecological Status” according to the WFD.

The simple unit value transfer approach should not be used for transfer between countries with different income levels and costs of living. Therefore, unit transfer with income adjustments has been applied.

The adjusted value estimate Vp' at the policy site can be calculated as

Vp' = Vs (Yp / Ys) β                                                                                   (1)

where Vs is the primary value estimate from the study site, Ys and Yp are the income levels at the study and policy site, respectively, and β is the income elasticity of demand for the environmental good in question. There is, however, little empirical evidence on how the income elasticity of demand β for different environmental goods and health impacts varies with income.

The primary assumption in adjusting WTP values in proportion to some measure of income is that the income elasticity of WTP is 1.0. However, there is no reason to think that WTP for environmental quality varies proportionally with income.

When we lack data on the income levels of the affected populations at the policy and study sites, Gross Domestic Product (GDP) per capita figures have been used as proxies for income in international benefit transfers. However, Barton (1999) clearly shows how this approach could give wrong results in international value transfers when income levels at the study and/or policy site deviates from the average income level in the countries.

2.3 Function transfer

Transferring the entire benefit function is conceptually more appealing than just transferring unit values because more information is effectively taken into account in the transfer. The benefit relationship to be transferred from the study site(s) to the policy site could be estimated using either revealed preference (RP) approaches like TC and HP methods or stated preferences (SP) approaches like the CV method and Choice Experiments (CE). For a CV study, the benefit function can be written as:

WTPij  = b0 + b1Gj + b2 Hij + e                                                               (2)

where WTPij = the willingness-to-pay of household i at site j, Gj = the set of characteristics of the environmental good at site j, and Hij  = the set of characteristics of household i at site j, and b0 , b1  and  b2   are sets of parameters and e is the random error.

To implement this approach the analyst would have to find a study in the existing literature with estimates of the constant b0 and the sets of parameters b1 and b2. Then the analyst would have to collect data on the two groups of independent variables G and H at the policy site, insert their mean values in equation (1), and calculate households´ WTP at the policy site.

The main problem with the benefit function approach is due to the exclusion of relevant variables in the WTP (or bid) function estimated in a single study. When the estimation is based on observations from a single study of one or a small number of recreational sites or a particular change in environmental quality, a lack of variation in some of the independent variables usually prohibits inclusion of these variables. For domestic benefit transfers researchers tackle this problem by choosing the study site to be as similar as possible to the policy site.

Benefit functions enable the calibration of the function to differences between the study site for which the function was developed and the policy site to which the function is applied (Loomis 1992; Parsons and Kealy 1994; Bergland et al. 1995; Kirchhoff et al. 1997 (for the bird-watching model only)). Although theoretically superior to unit transfer, the validity transfer tests (see appendix G for a review) suggest that in practice function transfers do not perform better than unit transfers. This is mostly due to the fact that very often benefit functions have low explanatory power due to omitted variables. This is especially true from Stated Preference methods, and less so for Travel Cost models for recreational activities and Hedonic Price models. Thus, unit transfer might work just as well, or even better, for both use and non-use values from Stated Preferences studies

Instead of transferring the benefit function from one selected valuation study, results from several valuation studies could be combined in a meta-analysis to estimate one common benefit function. Meta-analysis has been used to synthesize research findings and improve the quality of literature reviews of valuation studies in order to come up with adjusted unit values. In a meta-analysis, several primary studies are analysed as a group, where the result from each study is treated as a single observation in a regression analysis. If multiple results from each study are used, various meta-regression specifications can be used to account for such panel effects.

The meta-analysis allows us to evaluate the influence of a wider range in characteristics of the environmental good, the features of the samples used in each analysis (including characteristics of the population affected by the change in environmental quality), and the modelling assumptions. The resulting regression equations explaining variations in unit values can then be used together with data collected on the independent variables in the model that describes the policy site to construct an adjusted unit value. The regression from a meta-analysis would look similar to equation 2), but with one added independent variable; Cs = characteristics of the study s (and the dependent variable would be WTPs = mean willingness-to-pay from study s).

Smith and Kaoru’s (1990) and Walsh et al.´s (1990, 1992) meta-analyses of TC recreation demand models using both summary of TC and CV studies for the USDA Forest Service’s resource planning program were the first attempts to apply meta-analysis to environmental valuation. There have also been applications to CV studies of both use and non-use values of water quality improvements (Magnussen 1993), CV studies of groundwater protection (Boyle et al 1994), TC studies of freshwater fishing (Sturtevant et al. (1995), CV studies of visibility changes at national parks (Smith and Osborne 1996), CV studies of endangered species (Loomis and White 1996), CV studies of environmental functions of wetlands (Brouwer et al. 1997), CV studies of landscape changes (Santos 1998), CV studies of WTP for waste water treatment in coastal areas (Barton 1999), and outdoor recreation (Shrestha and Loomis 2001).

Many of these meta-analyses of relatively homogenous environmental goods are not particularly useful for value transfer even within the US where most of these analyses have been conducted, because they focus mostly on methodological differences. Methodological variables like "payment vehicle", "elicitation format", and "response rates" (as a general indicator of quality of mail surveys) in CV studies, and model assumptions, specifications and estimators in TC and HP studies, are not particularly useful in predicting values for specified change in environmental quality at the policy site. This focus on variables describing the methodological choices is partly due to the fact that some of these analyses were not constructed for benefit transfer (e.g. Smith and Karou 1990, Smith and Huang 1993, and Smith and Osborne 1996). Another reason is that insufficient and/or inadequate information was reported in the published studies with regard to characteristics of the study site, the change in environmental quality valued, and income and other socio-economic characteristics of the sampled population.  Particularly, the last class of variables would be necessary in international value transfer, assuming cross-country heterogeneity in preferences for environmental goods.

In most of the meta-analyses secondary information was collected on at least some of these initially omitted site and population characteristics variables, or for some proxy for them. These variables make it possible to value impacts outside the domain of a single valuation study, which is a main advantage of meta-analysis over the benefit function transfer approach. However, often the use of secondary data and/or proxy variables introduces added uncertainty. An example is a meta-analysis of recreational value of fishing using income data for the overall regional population in lack of income data for fishermen at the study site. On the other hand, the secondary data are more readily available at the policy site without having to do a survey of the fishermen at the policy site.

Most meta-analyses caution against using them for adjusting unit values due to potential biases from omitted variables and specification/measurement error of included variables. To increase the applicability of meta-analysis for value transfer, one should limit the scope of the analysis by selecting studies that are as similar as possible with regards to methodology, and thus be able to single out the effects of site and population characteristics on the value estimates. However, it is a problem that there are usually so few valuation studies of a specific environmental good that one cannot to do a statistically sound analysis.

2.4 Validity of spatial value transfer

Transfer errors arise when estimates from study sites are adapted to policy sites. These errors are inversely related to the degree of correspondence between factors  describing the study site and the policy site. The degree that any of these factors affects value transfer accuracy is an empirical question. However, the greater the correspondence (or similarity) of the policy site with the study site, the smaller the expected error (Boyle and Bergstrom 1992; Desvousges et al. 1992).

In the value transfer validity tests, two or more parallel valuation studies are conducted at different sites. Then an imaginary transfer is conducted from a study site (or a pooled data set from several study sites) to a policy site where we have also performed an original study. The transferred value, WTPT, is then compared to the value estimated in the primary valuation study at the policy site, WTPP.  The transfer error (TE) is calculated as the percent difference between the transferred estimate and the policy site primary estimate:

formel                                                                     (3)

Ready et al. (2004) show in their transfer tests of CV estimates of respiratory illnesses in five European countries that even if the distribution of WTP had been  the same in all countries, they would have measured an average transfer error of 16%.  Thus, they point out the average transfer error of 38 % they did find between should be assessed relative to this background level of random sampling error.

Much academic work has taken place in the past 10 years, testing the validity in of alternative value transfer methods for different environmental goods. However, even fairly small transfer errors can be rejected using the classical statistical tests (usually t-tests with null hypotheses of transferred value being equal to the original value). Bergland et al. (1995, 2004) rejected value transfers statistically in cases of average transfer errors of less than 20 % in two CV studies performed simultaneously of similar water quality improvements in two closely located and similar lakes. However, the standards of accuracy required in academic work may exceed those viewed as tolerable by policy-makers, especially in cost-benefit analyses like those likely to be performed at the national level e.g. in relation to the the EU Water Framework Directive to prioritize alternative investments in water quality and to show disproportionate costs. Kristofersson and Navrud (2005) suggest the use of equivalence testing as more appropriate and a clear compliment to the shortcomings of the classical tests. Equivalence tests test the null hypothesis of difference between the original and transferred value estimates (which is in most cases what we would expect rather than similar values). Equivalence tests also combine the concepts of statistical significance and policy significance into one test, by defining an acceptable transfer error prior to the validity test[5]. For applications of these tests see Kristofersson and Navrud (2003) and Muthke and Holm-Mueller (2004).

Appendix G shows that errors in individual transfers vary a lot, both within and between different validity tests and for all transfer methods. Since some of the transfer validity tests are performed under ideal conditions (i.e. same SP survey instrument used on a similar good in a nearby location at the same point in time; e.g. Bergland et al. 1995) they might underestimate transfer errors in practical transfer exercises. However, surprisingly many of these validity tests are performed under less than ideal conditions, and probably reflect quite well the transfer errors in practical value transfers. Brouwer (2000) surveyed seven of these value transfer studies and found that the average transfer error is around 20-40% for unit value transfers, and as high as 225% for benefit function transfers. Ready et al. (2004), however, found an average transfer error of 38% in a multi-country transfer test both for unit and function transfer. Shrestha and Loomis (2001) found an average transfer error of 28% in a meta-analysis model of 131 US recreation studies. Santos (1999) found that an international meta-analysis of CV estimates of landscape features could obtain less than 30% transfer error in 26% of cases; and less than 50% transfer error in 44% of cases.

Several of the studies listed in Appendix G support the hypothesis that the greater the correspondence, or similarity, between the study site and the policy site, the smaller the expected error in benefit transfers. Lower transfer errors resulted from in-state transfers than from across-state transfers (Loomis 1992; Van den Berg et al. 2001). This is potentially due to lower socioeconomic, socio-political, and socio-cultural differences for transfers within states, or political regions, than across states. In the Loomis et al. (1995) study, their Arkansas and Tennessee multi-site lake recreation models performed better in benefit transfers between the two regions (percent errors ranging from 1% to 25% with a nonlinear least squares models and 5% to 74% with the Heckman models) than either one when transferred to California (percent errors ranged from 106% to 475% for the nonlinear least squares models and from 1% to 113% for the Heckman models). This suggests that the similarity between the eastern models implicitly accounted for site characteristic effects.

Van den Berg et al. (2001) show accuracy gains when they transfer values and functions within communities that have shared experiences of groundwater contamination than transferring across states, within states, or to previously unaffected communities.

Brouwer (2000) suggests that if non-use values are motivated by overall commitment to environmental causes, they may tend to be relatively constant across populations and contexts. In a contingent valuation survey of the national populations in all Nordic countries Kristofersson and Navrud (2005) found that transfer errors are consistently smaller for the non-use value of a preservation plan for Nordic freshwater fish stocks. The results for a non-use value scenario by non-anglers in Norway and Sweden produced average transfer errors below 20 %. Use values for anglers showed higher transfer errors.  It may be that non-use value in these two countries is motivated by similar factors and is relatively context independent, while use value is more context specific. Clearly, this could be different for other environmental goods, particularly if the good has higher cultural significance in one country (or part of a country).

To summarize, the transfer validity studies conducted to date show that the average transfer error for spatial value transfers both within and across countries tends to be in the range of 25% - 40%. However, individual transfers could have errors as high as 100 - 200%. Function transfer does not seem to perform better than unit value transfer. Meta-analyses could also produce high transfer errors, and only those with a limited scope in terms of similar type environmental goods and similar type, state-of-the-art methodology, should be used. The validity tests also support the hypothesis that it is preferable to find a study site located close to the policy site of interest.  The closer the study site is to the policy site, the more likely that both the good being valued and the user population affected will be similar, and therefore the transfer errors would be lower. Transfer validity tests also suggest that transfer errors are smaller if people have had experience with the environmental good in question. Current evidence from CV studies, however, is not clear whether transfer errors for use values are lower than for non-use values, at least not between countries. [6]

2.5 Currency Conversion

The first issue that must be addressed when conducting international benefit transfer is the conversion to a common currency.  As will be shown, even in situations where the same currency is used in more than one country, there is still an issue related to currency conversion between countries.

Using the official exchange rates to convert the currency at the study to the currency of the policy site does not reflect the true purchasing power of currencies, since the official exchange rates reflect political and macroeconomic risk factors. If a currency is weak on the international market (partly because it is not fully convertible), people tend to buy domestically produced goods and services that are readily available locally. This enhances the purchasing powers of such currencies on local markets. To reflect the true underlying purchasing power of international currencies, the World Bank´s International Comparison Program (ICP) has developed measures of real Gross Domestic Product (GDP) on an internationally comparable scale. The transformation factors are called Purchasing Power Parities (PPPs); see also OECD’s list of PPPs at http://www.oecd.org/dataoecd/61/56/1876133.xls

Consider two individuals living in two different countries with the same preference structure over consumption of market goods, x, and the level of public goods available, Q.  Under what circumstances would we expect these two individuals to have the same willingness-to-pay (WTP) for a change in the level of the public good?  The individual in Country A has WTP for a change from Q0 to Q1 defined by the indirect utility function V:

V(IA, pA, Q0) = V(IA-WTPA, pA, Q1)                                                  (4)

where IA is income in Country A and pA is the price of market goods in Country A.  If the exchange rate between the currency in Country A and the Currency in Country B is given by β, what do we know about WTPB relative to WTPA?

Because indirect utility functions are homogeneous of degree 0, we know that:

V(β*IA, β*pA, Q0) = V(β*IA- β*WTPA, β*pA, Q1)                            (5)

Therefore, the individual in Country B will have willingness to pay WTPB= β*WTPA only if he has income IB= β*IA and faces prices pB= β*pA.  This last point is critical.  Identical individuals using different currencies will have the same real WTP only if they have the same real income and face the same real prices.  Thus, the appropriate exchange rate for converting values into a common currency is the exchange rate that equalizes market prices (see also Pattanayak et al. 2002).

This type of exchange rate is called a purchasing power parity (PPP) adjusted exchange rate.  The Penn World Table includes a list of PPP-adjusted exchange rates for 168 countries, based on price surveys conducted by the OECD and the World Bank.  PPP-adjusted exchange rates can differ markedly from financial exchange rates (the conversion rates offered in international financial markets).  This issue has not disappeared as a result of currency unification with the EU. Even though e.g. both the Netherlands and Portugal now use the euro, there remain differences in market prices between the two countries.  An individual living in the Netherlands with an annual income of 50,000 euros has a very different standard of living than an individual with identical preferences with the same income in Portugal, and will likely have different WTP for public goods.

When the policy site is smaller than an entire country, the analyst may need to worry about differences in prices even within a country; e.g. prices for market goods in Lisbon were 45% higher than the national average for Portugal.  When city or regional PPP indices are available, those should be used to account for local differences in prevailing prices.  This is true whether the benefit transfer is being conducted between countries or within a country.  In the U.S., the American Chamber of Commerce Research Association (www.accra.org) calculates cost of living indices for more than 300 cities.

A more difficult issue is differences among countries in in-kind income.  In many countries, health care, college tuition and retirement income are provided free of charge to all residents.  These represent a supplement to the real income of the citizens in those countries.  Citizens of these countries need to save less money for college expenses and retirement needs, and consequently can afford to pay more for public goods.  The challenge is to quantify these types of in-kind income, so that total income can be measured in consistent ways across different countries.

To conclude, even if PPP adjusted exchange rates can be used to adjust for differences in income and cost of living in different countries, this will not  correct for differences in individual preferences, cultural and institutional conditions between countries (or even within different regions of a country). An alternative is of course to restrict international benefit transfer to countries that are relatively similar with regards to in-kind income and most of the institutional setting; e.g. the Nordic countries.

2.6 Differences in income between countries

The most striking issue in international value transfer is differences in the level of incomes across countries.  Even within the EU, average per capita Gross Domestic Product (GDP) measured in PPP terms varies by over a factor of five between the richest and the poorest countries. Because most existing valuation studies were conducted in the U.S. or Western Europe, international benefit transfer often involves transfer of a value from a high-income country to a low-income country.  One common, simple approach to dealing with income differences between the study country and the policy country is to multiply unit values by the ratio of income in the policy country to income in the study country (or per capita GDP).  This approach assumes that WTP varies proportionally with income, an assumption that is typically not found to hold within individual studies. Studies conducted within one country more typically show that WTP for environmental goods increases with income, but at less than a proportionate rate.  Thus, using the income ratio as an adjustment will tend to undercorrect for income differences when the policy country is much richer than the study country.

A conceptually better approach is to apply a benefit function, i.e. WTP as a function of income.  In order to estimate such a function, variability in income is needed in the source data.  This variability typically comes from variation within the sample of users surveyed at the study site.  For example, we may discover that WTP for a public good valued at a study site is higher for users with higher income.  We use this variation to estimate a benefit function.  If the average income at a policy site is higher or lower than that at the study site, the benefit function adjusts for that difference.  A simple benefit function might include only income as the explanatory variable, and assume constant income elasticity of WTP.  However, the benefit function could account for other measurable characteristics, and could take other forms (see chapter 2.4).

The value function approach is probably defensible when the difference in measurable characteristics between the study site and the policy site is small, so that the average at the policy site falls well within the range of observations at the study site.  However, when conducting international benefits transfer, this may not always be the case.  A valuation survey conducted in Northern Europe will include respondents with varying levels of income.  However, few respondents will have incomes as low as those found in some developing countries.  Simply plugging the average user characteristics from a low-income policy site into a value function estimated in a high-income country can lead to serious problems:

First, there is the familiar problem of extrapolating outside the range of the data.  Particularly in socialized economies, the range of income within which most of the respondents fall may be fairly narrow.  The variability in the data may not be sufficient to identify curvature in the relationship between income and WTP.  But small changes in curvature have big implications when transferring the value function to a policy country where average incomes may be one tenth those of the study country.

The second issue is that the source of the variability, variation among individuals within the study country, is different from variation among countries.  The implicit assumption is that two individuals in different countries will have the same WTP if they have the same income (appropriately converted).  It is not clear, however, that a very wealthy individual in a poor country will necessarily have the same WTP for the public good as a poor individual in a wealthy country, if those two individuals have the same absolute real income.  Relative income may matter as well.  This is an issue that can only be adequately addressed by comparing value functions estimated from wealthy countries to value functions estimated from poorer countries. A related type of problem arises when transferring values from countries with large variations in income level, e.g.. USA, to Denmark and other Nordic countries with much less variation in income (and social security systems securing everybody a certain minimum level of income). This could easily make transfer from the US to Denmark more uncertain than transfers from other Nordic countries and European Countries with a more equal income distribution than the USA.

2.7 Differences in culture between countries

Not all factors that are important in determining values are measurable. Cultural heritage, shared values and shared experiences can also affect values for public goods. Values for cultural heritage goods and landscapes are probably especially sensitive to culture and shared experience.  A highly-valued traditional landscape in Denmark may not evoke similar values in France and vice versa. A restored ecosystem in Norway would probably be valued lower than a similar restoration project in Denmark, due to both differences in cultural heritage and preferences for undisturbed nature and the higher availability of relatively undisturbed wilderness areas in Norway.

Even differences in attitudes might be captured by additional questions (for example Likert-type agree/disagree questions) in the original study, though the same problem mentioned earlier in the context of income differences, i.e. the use of variability within a country to predict differences in values between countries, is likely to arise.

More difficult to deal with are differences among countries in characteristics that do not vary within each country, e.g. health care quality does not vary much across individuals (the U.S. being a notable exception).  A survey of e.g. WTP to avoid health impacts from poor drinking water quality conducted in Denmark, where everyone has access to care of similar quality, cannot reveal how individuals will value health in a country with better or poorer access or quality of care.  It is perhaps possible to take advantage of variation in health care quality across countries, and estimate a value function from values estimated in multiple countries.  However, the data need to do this is large.  When considering factors that vary by country but that do not vary within each country, surveys in many countries are needed in order to get sufficient degrees of freedom to work with to include the factors that vary by country and affect WTP for health.

2.8 Should values be the same in Denmark as EU average values?

One issue that has not received much attention in the value transfer literature is the issue of whether values should be adjusted when transferring from one jurisdiction to another, or from one population of users to another.  Should we use a recommended EU average unit value (e.g. as suggested by DG Environment of the European Commission for Value of a Statistical Life (VSL) and noise), in Denmark directly, or should we adjust it?

The answer depends on the purpose of the analysis.  First imagine that Denmark is considering an investment in water quality control in Denmark.  Obviously, Denmark should use its own unit values in evaluating that investment, rather than adopting an EU average value.

However, imagine that the EU was deciding which investments it should make in e.g. Denmark and Greece. If it uses Denmark`s values to value improved water quality in Denmark, and uses Greece’s values for improved water quality in Greece, it would direct more resources to improving Danish water quality than to Greek water quality.  It is politically (and probably morally) more defensible to use a common value for the same water quality change in all member countries.

2.9 Extent of the market

Much of the focus in the value transfer literature has been on the validity of transferring per household or per user estimates of value.  At least as important is the issue of how many households the value should be applied for to get aggregate costs or benefits of a new project or policy. Whenever a benefit transfer is conducted, it is critically important to determine the geographic extent of the population holding values for the environmental good.  Consider a typical local environmental good such as a beach, lake or park.  We would usually observe distance decay in value for such a site. If the population of users surveyed at the study site lives close to the site, then the transferred value will be valid only for the population living close to the policy site.  In contrast, if the good has national significance (e.g. a national park), and the transferred value estimate comes from a national sample, then the relevant population at the study site is the entire nation.

When conducting international benefit transfer, however, there are added complexities.  A single type of landscape, park, or river may have national significance in a geographically small country that has few such assets, while a similar good in a larger country has only regional or local significance.  This could make it difficult to transfer use and non-use values for forest areas from the other Nordic countries to Denmark, where forest areas are scarce. Further, the issue of which goods are of local importance and which are of national importance is often cultural as much as geographic.  The analyst needs to be careful when determining the extent of the market for the good.

2.10 Adjusting values over time

The usual procedure for temporal value transfer is to use the national consumer price index (CPI) to update the estimates to e.g. 2005-DKK, which implicitly assumes that individuals´ valuation of environmental goods increases at the same rate as the valuation of the private goods the CPI is based on. If the study is transferred from another country the recommended procedure for temporal transfer (e.g. in the ENVALUE database) is to convert the estimate to DKK using PPP (Purchase Power Parity) – corrected exchange rates of the time of the data collection at the study site, and then to update to 2005-DKK using the CPI of Denmark. However, the value of environmental goods over time could increase relative to private goods due to increased scarcity of environmental goods. Also, with increased income over time people would place a higher relative value on environmental goods compared to the private goods in the CPI. The empirical analysis of Kriström and Riera (1996) indicates that the income elasticity of environmental goods is in fact probably (slightly) smaller than one.

Several tests of the validity of temporal value transfer have been performed. Bergland et al. (1996) found stable preferences over a 4-year period for the recreational value of elk hunting in Oregon, USA.  Brouwer and Bateman (2005) investigated the temporal stability over a 5-year period of WTP responses from two large-scale CV surveys on flood protection and wetland conservation, respectively.  They found a significant difference across this longer period. However, tests of model transferability indicate that simple models, based solely upon variables derived from economic theory, are transferable across this period. This suggests that underlying relationships for such key determinants are stable even across this longer period. However, expanding models by including theoretically unanticipated factors brings ad-hoc and possibly transitory factors into the models, which consequently prove non-transferable.

Zandersen et al. (2005) report a 20-year temporal transfer validity tests for forest recreation in Denmark. They show that updating the transfer model (with data from a relatively inexpensive household survey, showing a decrease in car-borne travel to forests in these 20 years) leads to a reduction in transfer error by a factor of 4, down to a transfer error of 25 %. This shows the importance of updating the transfer model, even for shorter periods of time, in cases there have been significant changes in the determinants of WTP. Brouwer and Bateman (2005) review previous studies considering shorter periods (typically 1-2 years), and they showed no significant difference in real WTP values over time.

These temporal validity tests indicate that the real value of environmental goods is stable if there have been no big shifts in the determinants of WTP. By transferring from studies performed closer in time (i.e. less than 5-10 years), the chances of such a shift is smaller. Also, the probability of these original studies satisfying the current “best practice” criteria is higher when the studies are performed closer in time. If there have been significant shifts in these best practice criteria, one should very carefully assess the quality of the studies performed before this shift, e.g. Contingent Valuation studies performed before the NOAA Panel guidelines (Arrow et al. 1993).

To conclude, the standard procedure of adjusting values by the CPI seems to be appropriate but would probably underestimate the increase in relative value over time if the good becomes scarce, and/or people’s relative income increases and the income elasticity of the environmental good is high. Thus, the standard procedure for temporal transfer will most probably produce conservative (lower) value estimates.

Footnotes

[1]  Non-use value captures option value, existence and bequest value

[2] Consumer surplus (CS) is the difference between individuals´ maximum WTP (as outlined by the demand curve) and what they actually pay (market price). In the TC model this is the difference between the estimated demand for transportation to the recreational site (typically the number of visits per individual per year as a function of travel costs per visit (round trip; and usually assuming everybody arriving by car)) and all other variables thought to influence the frequency of visits.

[3] The challenge is to make value transfer spatially explicit by disaggregating complex landscapes into constituent land cover units and ecosystem service types that can be effectively transferred from one site to another. Although EcoVaue makes an attempt to do this, the list of valuation studies (on their website http://ecovalue.uvm.edu.) that they use for their analyses is very limited and seems to lack many high quality primary valuation studies relevant for transfer.

[4] However, this does not seem  to be a relevant impact in most European countries.

[5] In t-tests, the usual level of acceptable transfer error is set at 5%, while in equivalency tests the researcher is free to set the level of acceptable transfer error, dependent on the decision situation and policy use. However, the tests themselves do not provide evidence for what the acceptable transfer error should be. This has to be analyzed in a decision theoretic framework in each case (see the Guidelines in Chapter 4).

[6] E.g. Kristofersson and Navrud (2002) found smaller transfer errors for non-use than for use values of freshwater fish stocks in Norway, Sweden and Iceland mainly due to different markets/institutions for recreational fishing between Iceland on one hand and Norway and Sweden on the other.

 

| Front page | | Contents | | Previous | | Next | | Top |



Version 1.0 December 2007, © Danish Environmental Protection Agency