Valuing wetland attributes: an application of choice experiments

Transcription

1 Ecological Economics 47 (2003) ANALYSIS Valuing wetland attributes: an application of choice experiments Fredrik Carlsson a, Peter Frykblom b, *, Carolina Liljenstolpe c a Department of Economics, Gothenburg University, Sweden b Department of Economics, Appalachian State University, Boone, NC , USA c Department of Economics, Swedish University of Agricultural Economics, Sweden Received 4 December 2001; received in revised form 18 June 2002; accepted 2 September Abstract The interest for wetlands is increasing, not only because of the possibility of a cost-efficient uptake of nutrients, but also because wetlands can be designed to provide other services. What values that are supplied depend largely on the design. There are numerous different design options, and different actors may promote different alternatives. Whether we want to design a wetland for nutrient retention alone, or one that also serves other interests, policy makers need information about the value of different options. Conducting a choice experiment, we are able to identify attributes that increase and decrease citizens perceived value of wetlands. Using a random parameter model we find that biodiversity and walking facilities are the two greatest contributors to welfare, while a fenced waterline and introduction of crayfish decrease welfare. D 2003 Elsevier B.V. All rights reserved. Keywords: Wetland; Valuation; Choice experiment; Random parameter model 1. Introduction There is an increasing interest in restoration and construction of wetlands, as a means to increase the retention of nutrients. In southern Sweden, more than 90% of the original wetlands have been eradicated as cities, roads and agriculture have expanded. Combined with heavy loads of nutrients, the region suffers from severe eutrophication damages to coastal and groundwater. Several projects to protect the water are being carried out, where construction of wetlands is * Corresponding author. Tel.: ; fax: address: frykblomlp@appstate.edu (P. Frykblom). one measure to reduce the run-off of nutrients. Studies by Gren (1993) and Byström (1999) show that wetlands can be a cost-efficient means of nitrogen reduction. While the nitrogen reduction is an important characteristic, wetlands can further contribute to other goods that are valuable to society, such as biodiversity, recreation, diversity in the landscape, etc. The cost-efficiency studies by Gren (1993) and Byström (1999) are enough to answer a question of why wetlands are important from a social perspective, if we assume that the sum of the value of other services from wetlands is at least non-zero or even positive. Such an assumption has been supported by studies where the value of wetland services has been discussed, and in some cases also estimated (see Gren, 1995; Stevens et al., 1995; Oglethorp and Miliadou, /$ - see front matter D 2003 Elsevier B.V. All rights reserved. doi: /j.ecolecon

2 96 F. Carlsson et al. / Ecological Economics 47 (2003) ). However, this is not to say that the other characteristics of a wetland should be disregarded when constructing the areas. In this paper we try to identify what characteristics of wetlands, besides retention of nutrients, that individuals think are important. More precisely, we try to estimate individuals marginal willingness-to-pay (WTP) for different attributes of a wetland. This information can thus be used as input when deciding what type of wetland that should be constructed, resulting in both increased social welfare and a wider acceptance of wetlands. Individuals marginal WTP is estimated through a choice experiment, in which individuals are asked to choose between different wetland alternatives with different characteristics. In the choice experiment we identify several important attributes. Biodiversity and walking facilities are the highest valued attributes, while a fenced waterline and introduction of crayfish are regarded as something negative. 2. Design of wetlands Apart from being important for retention of nutrients, a wetland area can be designed to provide different additional services and values. A wetland can for example be designed with surrounding walking facilities, tree plantations and certain water depths to attract birds or fish. There are a number of different design options, where different actors may promote different attributes. No matter whether we want to design a wetland mainly for nutrient retention or one that also serves other interests, policy makers need information about the valuation of the different attributes. There are a number of valuation studies of wetlands, see Heimlich et al. (1998) for an extensive overview of valuation studies. This overview includes a broad variety of valuation techniques, such as the contingent valuation method, hedonic price, replacement value, damage avoided and production value. These studies focus on estimating values of different types of wetlands or single services provided by wetlands. It may be difficult to estimate the marginal value of different attributes within a wetland though, because of lack of variation in the data. A choice experiment can solve this problem. The basic idea behind a choice experiment is to create a hypothetical market situation and elicit individuals preferences for the attributes by asking them to make choices between certain alternatives. The roots of choice experiments can be traced back to Lancaster (1966) and the theory of demand for characteristics, and random utility theory (Thurstone, 1927; McFadden, 1974). In recent years there has also been an increased use of choice experiments in non-market valuation (see e.g. Adamowicz et al., 1994; Layton and Brown, 2000; Alpizar et al., 2001). The method is particularly suitable for estimating marginal rates of substitution between different attributes of for example a wetland. Morrison et al. (1999) is the only study applying a choice experiment to the valuation of wetlands, as far as we know. These authors estimate non-use environmental values of a wetland area in Australia. In particular they investigate the trade-off between nonuse values in job losses and environmental quality. The objective of these wetland valuation studies in general has been to estimate the WTP for a wetland area itself, in particular the use and non-use values of improved environmental quality. Our objective is instead to value the different attributes, including biodiversity, of a wetland. 3. The wetland choice experiment 3.1. The survey The choice experiment concerns a wetland area in Staffanstorp, southern Sweden. The municipality of Staffanstorp plans to develop a wetland in this area, and therefore it is highly suitable for our choice experiment. The survey was conducted in collaboration with the municipality of Staffanstorp, where there has been a public discussion about the location and design of the wetland. The respondents were the local population, members of individuals living in Staffanstorp. The questionnaire consisted of two parts. The first part contained the choice experiment and the second part questions regarding the respondent s socio-economic status. The questionnaire and the attributes used in the choice experiment were developed in cooperation with researchers specialized on wetlands from Linköping, Lund and Uppsala University. Several focus group discussions and one pilot study were

3 F. Carlsson et al. / Ecological Economics 47 (2003) Table 1 Attributes and attribute levels Attribute Description Levels Total cost The total cost for the individual SEK 200, 400, 700, 850 if the alternative was chosen. Surrounding vegetation Forest or meadow-land. Forest, Meadow Biodiversity a The design of the wetland area can Low, Medium, High promote the plant, animal and insect life so that the wetland contains different numbers of both rare and more common species. Fish The design of the wetland area can improve the No, Yes conditions for species such as bass, pike, roach, etc. Fenced waterline It is possible to surround the water with a 1-m fence in No, Yes order to prevent drowning accidents. Crayfish It is possible to introduce Swedish crayfish and allow fishing. No, Yes Walking facilities It is possible to construct the wetland area for outdoor life with construction of walking tracks and information signs about the plant and animal life. The tracks are suitable for walking and jogging. No, Yes a Biodiversity was described as the number of rare species that would be found in the wetland (none, few or many). This choice of description was driven by a difficulty to explain the complexity in a meaningful and understandable way. conducted in the process of designing a working mail questionnaire. 1 In the introduction to the choice experiment, the purpose of construction of wetlands was briefly explained. The respondents were then informed about the particular wetland area in Staffanstorp that is about to be developed, and told that we were interested in their view on the best possible wetland. Next the attributes used in the choice experiment were explained. The respondents were provided with a separate fact-sheet describing the attributes. In the choice experiment, each respondent answered four choice sets. In each choice set they were asked to choose between three alternatives. The first alternative was always the base alternative, in which there would be no improvements to the wetland area, at no cost. The two other alternatives implied a number of improvements to the wetland area. The attributes and the levels of the attributes are briefly presented in Table 1, and an example of a choice situation is presented in Appendix A. 1 In the focus groups, the subjects filled out the questionnaire, and were later asked written questions on their perceptions. This was followed up by an open group discussion led by a conductor. After revisions in the questionnaire were made, the procedure was repeated with new subjects. Having achieved a questionnaire that worked in our focus groups, a pilot study was undertaken. The questionnaire was sent out to 130 randomly chosen individuals. Every fifth respondent and nonrespondent was later telephone interviewed about the questionnaire. The choice sets were created using the OPTEX procedure in SAS, which is a linear D-optimal design procedure (see Kuhfeld, 2001). The design is selected from the collective factorial, where collective factorial is an L AC factorial, where C is the number of alternatives and where each alternative has A attributes with L levels. Using the OPTEX procedure we created 60 choice sets, these were then blocked into 15 versions each containing four choice sets. We also test a hypothesis of stable preferences during the choice experiment by incorporating a simple test into the design. Since the respondent faces all choice situations sequentially in the short time required to perform the survey, the assumption of stable preferences seems realistic, but nevertheless it is often mentioned as a potential problem of choice experiments. Half of the respondents received the choice sets in the order {1,2,3,4} and half of them in the order {4,2,3,1}. We can then test for stability by comparing the estimated preferences for the two different orders. This can be done in several ways: using all choice sets, using only the first and last three, or only the first and last. Based on the conditional logit model, the hypothesis of stable preferences cannot be rejected for any of the tests. Both Carlsson and Martinsson (2001) and Layton and Brown (2000) perform similar tests and fail to reject the stability hypothesis.

4 98 F. Carlsson et al. / Ecological Economics 47 (2003) Econometric specification In the analysis of the responses we apply a general type of model, a random parameter logit model. In such a model, taste variation among individuals is explicitly treated (see e.g. Train, 1998). A random parameter logit model is a generalization of a standard multinomial logit. The advantages of a random parameter logit model are that (i) the alternatives are not independent, i.e. the model does not exhibit the independence of irrelevant alternatives property and (ii) there is an explicit account for unobserved heterogeneity. Define a latent utility function of alternative i for individual q, at choice situation t, consisting of a systematic and a stochastic part, U iqt ¼ a iq þ c i s q þ b q x iqt þ e iqt ð1þ where x iq is a vector of attributes and s q is a vector of socio-economic characteristics. The alternative specific intercept a iq captures an intrinsic preference for the alternative and c i s q captures systematic preference heterogeneity as a function of individual characteristics. The coefficient vector b q varies among the population with density f (b h), where h is a vector of the true parameters of the taste distribution. If the e s are IID type I extreme value we have a random parameter logit, or a mixed logit, model (Train, 1998). The conditional probability of alternative i for individual q in choice situation t is then P q ðit b q Þ¼ expða iq þ c i s q þ b q x iqt Þ X jaa t expða jq þ c j s q þ b q x jqt Þ ; ð2þ where A t ={A 1,..., A N } is the choice set. The conditional probability of observing a sequence of choices, denoted j( q, t), from the choice sets is the product of the conditional probabilities S q ðb q Þ¼ Y t Pðjðq; tþtjb q Þ: ð3þ In the choice experiment, the sequence of choices is the number of hypothetical choices each respondent makes in the survey. The unconditional probability for a sequence of choices for individual q is then the integral of the conditional probability in Eq. (3) over all values of b: Z P q ðhþ ¼ S q ðbþf ðbjhþdb: ð4þ In this simple form, the utility coefficients vary among individuals, but are constant among the choice situations for each individual. This reflects an underlying assumption of stable preference structures for all individuals (Train, 1999a). In general the integral in Eq. (4) cannot be evaluated analytically, and we have to rely on a simulation method for the probabilities. Here we will use a simulated maximum likelihood estimator, using Halton draws, to estimate the models (see Revelt and Train, in preparation; Train, 1999b). It is also necessary to make an assumption regarding the distribution of each of the random coefficients. In principle any distribution could be applied. However, the choice is often limited by difficulty of model estimation and availability econometric software. The two main alternative assumptions are a normal and a log-normal distribution. Applying a log-normal distribution means that we restrict all respondents to have the same sign of each coefficient. In our case this is not desirable, since we expect different respondents to have positive and negative preferences for the different attributes of the wetland area. Therefore we assume a normal distribution of all the attributes of the choice experiment. It is also reasonable to expect that there is a correlation between the randomly distributed parameters. Therefore we estimate the full preference variance covariance matrix. We then have that the vector of random parameters is normally distributed with variance covariance matrix X, i.e. b q fn(b, X). The random part of the coefficient vector can then be written b q =b +Cg q, where g q is a vector of independent standard normal deviations. We then estimate b and C, where C is the lower triangular Cholesky factor of the preference variance covariance matrix X, i.e. X=CC V. This type of random parameter model is less restrictive than standard conditional logit models. However, we believe that researchers should be cautious in applying these less restrictive models. Apart from being more difficult to estimate, our experience is that the results can be rather sensitive to the distributional assumptions and, for example, the number of draws applied in the simulation. Although specification tests exist (McFadden and Train,

5 F. Carlsson et al. / Ecological Economics 47 (2003) ), the approach of finding a correct specification of the random part of the model is tedious. Furthermore, the gain in terms of precision of the estimates of willingness to pay is unclear. We therefore also report the estimates of conditional logit model in which all coefficients are treated as fixed. 4. Results Table 2 Descriptive statistics observations included in final estimations Variable Description Mean Std. Min Max Age Respondent age Male =1 if respondent is a male Kid =1 if household has children Responses in choice experiment Number Alt 1. No 408 improvement Alt Alt The population that the sample was chosen from was defined as those between 18 and 75 years living in Staffanstorp; in all 13,000 citizens. A random sample of 1200 individuals was selected from the Swedish census register. A mail survey was conducted in May 2001, a reminder was sent out 2 weeks afterwards to those that still had not replied. In total 580 (48%) individuals returned the questionnaire, of which 468 were available for analyses due to nonresponses to various items. Not all of these answered all four choice sets, however we still chose to include these individuals in our estimations. In Table 2 below, the descriptive statistics of the sample used in the estimations are presented. Any self-administred survey runs the risk of selfselection bias. Undertaking a comparison of average age and gender, with data from official records on citizens in Staffanstorp municipality, a null hypothesis of equality could not be rejected. Using Limdep 7 we estimate the random parameter logit models with simulated maximum likelihood using Halton draws with 250 replications. For comparison we also estimate a standard conditional logit model. We include one common alternative-specific intercept for the two alternatives that imply changes in the design of the wetland area, i.e. the non-base alternatives, since these were presented in a generic form. We let the cost variable be fixed, and not randomly distributed, for two reasons: (i) the distribution of the marginal willingness-to-pay for an attribute is then simply the distribution of that attribute s coefficient, and (ii) we wish to restrict the price variable to be non-positive for all individuals. The non-price attributes are all randomly distributed with a normal distribution, with the exception Surrounding vegetation. This variable was insignificant in the conditional logit model, and in the random model both the mean and standard deviation were insignificant. We, therefore treat the variable Surrounding vegetation as fixed in the random model. In addition, a number of individual characteristics are included as fixed coefficients. These characteristics interact with the alternative-specific intercept. The results of the estimations for both the conditional and the random parameter logit model are presented in Table 3. The significance of the estimated standard deviations is a sign of heterogeneity in preferences among the respondents. There is also a correlation in the heterogeneity of preferences between attributes. This together with the substantial increase in the likelihood ratio index indicates the advantage of applying the random parameter model instead of the conditional logit model. All attributes except for Surrounding vegetation are significant in the conditional logit model. This attribute is also insignificant in the random parameter logit model. All other attributes and their standard deviations are significant in the random parameter model, except for the mean coefficient for Crayfish. This implies that there is heterogeneity in preferences for these attributes. Furthermore, the relative magnitude of the standard deviations implies that there is a probability that people have the reverse preference for a particular attribute. This can also be seen from column (5) in Table 3, which reports the probability that the coefficient will have the reverse sign, compared to the mean estimate. The mean coefficient is negative for both Fenced waterline and Crayfish, so it is more likely that the respondents dislike these attributes. All other attributes are significant and have a positive coefficient estimate. Even so,

6 100 F. Carlsson et al. / Ecological Economics 47 (2003) Table 3 Conditional logit and random parameter logit estimations, p-values in parentheses Logit Random parameter logit Coefficient ( p-value) Coefficient ( p-value) Coeff std. ( p-value) High Biodiversity, b (0.00) (0.00) (0.00) 0.17 Medium Biodiversity, b (0.00) (0.00) (0.00) 0.28 Fish, b (0.00) (0.00) (0.00) 0.36 Fenced waterline, b (0.01) (0.01) (0.00) 0.39 Crayfish, b (0.03) (0.33) (0.00) 0.46 Walking facilities, b (0.00) (0.00) (0.00) 0.26 Intercept (0.00) (0.01) Surrounding vegetation is meadow land (0.40) (0.88) Cost (0.00) (0.00) Male (0.02) (0.67) Age (0.00) (0.03) Kid (0.06) (0.87) Log-likelihood Likelihood ratio index No. of respondents 490 No. of observations 1717 Prob. reversed sign Correlation matrix for random parameters b 1 b 2 b 3 b 4 b 5 b 6 b 1 1 b b b b b the estimated standard deviations are high, and even for these attributes there is a non-negligible probability that the respondents dislike the attributes. From the correlation matrix we see that there is a strong positive correlation between a positive preference for biodiversity and the other attributes, and that there is a positive correlation between a positive preference for Fish and Crayfish. 2 Among the socio-economic characteristics, only Age is significant. The negative sign indicates that elder respondents are less likely to choose an improved, and more costly wetland. The interpretation of the coefficient values is not straightforward, except for the significance and relative size. We therefore calculate the marginal rates of substitution between the attributes using the coefficient for cost as numeraire. This implies that we can 2 Most of the elements of Cholesky matrix were significant, indicating a significant correlation between the attributes. interpret the ratios as average marginal WTP for a change in each attribute, as argued by Hanemann (1984). The results presented in Table 4, then provide relevant input for a policy maker when designing the wetland area. Since all attributes are normally distributed, and the cost coefficient is fixed, marginal WTP will also be normally distributed. The distribution of the marginal willingness to pay is obtained with the Krinsky-Robb method (Krinsky and Robb, 1986) using 5000 replications. 3 3 Using this method we randomly draw the coefficients a number of times from the asymptotic normal distribution of the parameter estimates, and calculate the fare equivalents for each of these draws. An alternative method would be bootstrapping where we create a number of new data sets using the estimated residuals, and re-estimate the function. The Krinsky-Robb method is less computationally burdensome. Furthermore, Kling (1991) and Chen and Cosslett (1998) among others find that the two procedures give quite similar standard deviations.

7 F. Carlsson et al. / Ecological Economics 47 (2003) Table 4 Marginal willingness to pay for attributes, 90% confidence interval Logit Random parameter logit Surrounding vegetation ( ) 6.53 ( ) is meadow land High biodiversity ( ) ( ) Medium biodiversity ( ) ( ) Fish ( ) ( ) Fenced waterline ( ) ( ) Crayfish ( ) ( ) Walking facilities ( ) ( ) Surrounding vegetation has an insignificant WTP in both models, and in addition the WTP for Crayfish is insignificant in the random parameter model. The marginal WTP is negative for the two attributes Fenced waterline and Crayfish, i.e. the two variables decrease the average utility derived from a wetland area. The marginal WTP is highest for High biodiversity and Walking facilities, followed by Medium Biodiversity. The other attributes are all of similar magnitude. There is no general pattern in the differences in marginal WTP between the random parameter model and the conditional logit model. Consequently, we cannot say that one method in general gives different results compared to the other. Furthermore, the differences in WTP between the two models are not very large, perhaps with the exception of Crayfish. Consequently, although the conditional logit model can be rejected in favor of the random parameter model, there is no clear pattern in the gain of the WTP estimates. The important additional information that the random parameter model gives is perhaps mainly that there is a strong heterogeneity in the preferences for the attributes. 5. Conclusions What are wetlands good for, besides the uptake of nutrients? By the use of a choice experiment, we have identified a number of attributes that either increase or decrease the utility derived from a wetland area. The results are contextual, i.e. they are the result of a certain study conducted in a specific community. The southwest part of Sweden is an area relatively densely populated with few recreation areas. These circumstances probably affect the result, and the appropriateness of a transfer of the results to other areas. The possibilities of transfers are finally an empirical question, i.e. it can be tested by further studies in similar and different areas. Keeping this in mind, we can still learn something by analysis of the obtained results. First, a comparison of the conditional logit specification with a random parameter model shows that the less restrictive latter model can provide us with information that cannot be shown by the standard model. There are heterogeneous preferences for several of the attributes, as all the coefficients of the random attributes have significant standard deviations and a high probability of a reversed sign. Second, a negative mean WTP was found for the three attributes Meadow land, Fenced waterline and Crayfish. Following, an inclusion of these attributes will decrease social welfare. However, the WTP for Meadow land was insignificant in both models. Third, Biodiversity and Walking facilities have the highest marginal WTP in our study. A natural extension of this survey is to estimate the marginal cost of providing the different attributes of a wetland. In that way, the results can be used for constructing a socially efficient design of the wetland. Acknowledgements We are grateful for help from Rob Hart, Knut Per Hasund, Lars Leonardson, Elisabeth Lundqvist, Jan Lundström, Tore Söderqvist, Stefan Weisner and three anonymous referees. The study was financed by Vastra.

8 102 F. Carlsson et al. / Ecological Economics 47 (2003) Appendix A. Choice situations Before you chose any alternatives, you should now read the fact sheet. Choice 1 Of the three alternatives below, mark the alternative you prefer. Your choice (Mark your choice) Wetland Alternative 1 Simple ponds Alternative 2 Alternative 3 Attributes: Surrounding vegetation 1. Surrounding vegetation Forest Forest Meadow-land Water issues 2. Fish No actions Good conditions No actions 3. Cray fish No introduction Introduction No introduction Other attributes 4. Biodiversity Low Low High 5. Walking facilities No walking facilities No walking facilities Walking facilities 6. Fence No No Fence Total cost per citizen SEK 0 SEK 850 SEK 400 References Adamowicz, W., Louviere, J., Williams, M., Combining revealed and stated preference methods for valuing environmental amenities. Journal of Environmental Economics and Management 26, Alpizar, A., Carlsson, F., Martinsson, P., Using choice experiments for non-market valuation. Working Paper in Economics No. 52, Department of Economics, Gothenburg University. Byström, O., Wetlands as a nitrogen sink estimation of costs in Laholm bay. In: Boman, M., Brännlund, R., Kriström, B. (Eds.), Topics in Environmental Economics Klüwer Academic Publishers, Dordrecht. Carlsson, F., Martinsson, P., Do hypothetical and actual marginal willingness to pay differ in choice experiments? Application to the valuation of the environment. Journal of Environmental Economics and Management 41, Chen, H., Cosslett, S., Environmental quality preference and benefit estimation in multinomial probit models: a simulation approach. American Journal of Agricultural Economics 80, Gren, I-M., Alternative nitrogen reduction policies in the Mälar region, Sweden. Ecological Economics 7, Gren, I.-M., The value of investing in wetlands for nitrogen abatement. European Review of Agricultural Economics 22, Hanemann, M., Welfare evaluations in contingent valuation experiments with discrete responses. American Journal of Agricultural Economics 66, Heimlich,R.E.,Weibe,K.D.,Claassen,R.,Gadsy,D.,House, R.M., Wetlands and Agriculture: private interests and public benefits. Resource Economics Division, E.R.S., USDA, Agricultural Economic Report Kling, C., Estimating the precision of welfare measures. Journal of Environmental Economics and Management 21, Krinsky, I., Robb, A., On approximating the statistical properties of elasticities. Review of Economics and Statistics 68, Kuhfeld, W., Multinomial logit, discrete choice modeling. An introduction to designing choice experiments, and collecting, processing and analyzing choice data with SAS. SAS Institute TS-643. Lancaster, K., A new approach to consumer theory. Journal of Political Economics 74, Layton, D., Brown, G., Heterogeneous preferences regarding global climate change. Review of Economics and Statistics 82, McFadden, D., Conditional logit analysis of qualitative choice behavior. In: Zarembka, P. (Ed.), Frontiers in Econometrics Academic Press, New York. McFadden, D., Train, K., Mixed MNL models for discrete response. Journal of Applied Econometrics 15, Morrison, M., Bennett, J., Blamey, R., Valuing improved wetlands quality using choice modeling. Water and Resource Research 35, Oglethorp, A.R., Miliadou, A., Economic valuation of the non-use attributes of a wetland: a case study for lake Kerkini. Journal of Environmental Planning and Management 43,

9 F. Carlsson et al. / Ecological Economics 47 (2003) Revelt, D., Train, K., Mixed logit with repeated choices: households choices of appliance efficiency level. Review of Economics and Statistics (in preparation). Stevens, T.H., Benin, S., Larson, J.S., Public attitudes and economic values for wetland preservation in New England. Wetlands 15, Thurstone, L., A law of comparative judgement. Psychological Review 4, Train, K., Recreation demand models with taste differences over people. Land Economics 74, Train, K., 1999a. Mixed logit models for recreation demand. In: Herriges, J., Kling, C. (Eds.), Valuing Recreation and the Environment Edward Elgar, Cheltenham. Train, K., 1999b. Halton sequences for mixed logit. Working Paper Department of Economics, University of California, Berkeley.