Abstract
We demonstrate how choice experiment survey methods can be used to guide ecosystem restoration efforts. We use a choice experiment survey to estimate willingness to pay (WTP) for different attributes of restored grassland ecosystems. We find that the presence of nearby grasslands increases a respondent’s WTP and that species richness, population density, and the presence of endangered species are all significant factors that contribute to the respondent’s WTP. This implies that all these conservation success measures should be taken into account in planning and in research. To our knowledge this is the first study to calculate the WTP for grasslands. (JEL Q51, Q57)
I. Introduction
Nonmarket valuation techniques have been developed by economists to estimate the value to society of goods not sold in the marketplace, such as environmental quality and mortality risk reduction. In environmental economics, these value estimates have been used primarily as inputs to cost-benefit analyses and to estimate damages for which firms can be held liable after events such as oil spills. In this paper, we demonstrate how a tool in the environmental-economics valuation toolkit—choice experiment survey methods—can also be used for another important purpose: guiding complex decisions about how best to carry out and manage ecosystem restoration projects. We do this by estimating consumer preferences over multiple conservation attributes of restored ecosystems, illustrated using the restoration of grasslands, an ecosystem that has not been widely studied in the economic valuation literature.
Large-scale conversion of many natural habitats has put pressure on rare and endangered species and decreased the flows of many ecosystem services. In response, conservation organizations seek to protect and restore land with high conservation and biodiversity values; this has led to much research on optimal protected area planning (e.g., Primack 1993; Ando et al. 1998; Margules and Pressey 2000) and restoration (Loomis et al. 2000; Milon and Scrogin 2006; Meyerhoff and Dehnhardt 2007). Most of that research uses productionside factors—the locations of endangered species, the cost of land, the threat posed to natural areas by development—to guide decisions about where to locate dedicated natural areas and what features those areas should have. However, Ando and Shah (2010) show that conservation activity can yield higher social benefits if decision makers consider the preferences of people when they plan their network of natural areas.
Two features of consumer preferences are important for deciding how best to invest social resources in restoration projects. First, optimal positioning of a restored area in the landscape depends on how the value people derive from an area varies with proximity and with features of the landscape around it. Competing economic theories yield diverse predictions about how the existing quantity of an environmental public good (an existing natural area) affects willingness to pay (WTP) for providing more of that good (restoring more of that ecosystem). Decreasing marginal WTP predicts that marginal WTP for an increase in a public good will be lower for consumers who already have access to a relatively large quantity of that good. On the other hand, endogenous preferences or experience can lead to the opposite effect (Cameron and Englin 1997; Bowles 1998; Zizzo 2003; Gowdy 2004). We evaluate these competing predictions by analyzing how the WTP to restore a new grassland is affected by the presence of grassland areas nearby. We also estimate how consumer WTP for a restored area varies with how far the consumers live from it, contributing more evidence to the growing body of work on this subject (e.g., Bateman et al. 2006).
Second, the structure of preferences over multiple attributes of a given restoration project affects the nature of the value-maximizing bundle of attributes. Most existing nonmarket valuation research that identify values for restoration use contingent valuation (CV), which does not allow relationships in the values of multiple attributes to be analyzed. The studies of restoration values that use choice experiment (CE) surveys (Carlsson, Frykblom, and Liljenstolpe 2003; Birol, Karousakis, and Koundouri 2006; Christie et al. 2006) do not use attribute interaction terms; the standard econometric specification of that research implicitly assumes consumers have linear indifference curves between pairs of attributes that comprise the good. This paper uses CE valuation techniques to estimate the values of and the nature of substitutability between multiple facets of a restored ecosystem by including interaction terms between attributes. This allows the testing of whether the marginal value of any one measure of conservation success— species richness, population density, and the presence of endangered species—is affected by the levels of the other two.
We carry out our research on the structure of consumer preferences over restoration projects in a setting that has been neglected by the valuation literature: grassland ecosystems. Though there have been many CV (and more recently CE) studies estimating the values of conserving and restoring ecosystems such as wetlands and forests, economic valuation efforts have not attempted to estimate the social value of grassland ecosystems. Massive conversion of grassland in North America to urban and agricultural use has stressed wildlife and cut ecosystem service provision in large swaths of the continent. This problem can be addressed with grassland restoration activities, but such projects are costly and require difficult and seemingly arbitrary choices to be made about the exact nature of the grasslands created. The restoration ecologists who carry out grassland restoration have no guidance from the economic valuation literature about the preferences people have over the characteristics of restored grasslands. In this paper we meet that need for knowledge by using a CE survey of Illinois residents to analyze WTP for grassland habitat restoration.
We find that species richness, population density, presence of endangered species, presence of wildflowers, and distance from an individual's home are all significant factors that affect consumers' WTP to restore an endangered ecosystem. This challenges the common practice used in economic valuation and protected area planning studies of using just one measure, such as species richness, as a stand-alone indicator of conservation success. We also find that respondents with existing grasslands nearby have a higher WTP for restoring a new grassland; this result can possibly be explained by endogenous preferences or locational sorting. We find some evidence that the marginal value respondents place on species richness is lower if the level of population density is high (and vice versa). This finding implies that households may favor restoration projects with either a very diverse set of species or very large numbers of individual birds in the area, rather than a balanced portfolio of those two features of the conservation outcome.
II. Literature Review
There is a fairly extensive literature on using nonmarket valuation to obtain values for restoring ecosystems. Examples include CV studies of the values of restoring an impaired river basin (Loomis et al. 2000), total economic value of restoring ecosystem services in the Ejina region in China (Zhongmin et al. 2003), benefits of woodland restoration in native forests in the United Kingdom (Macmillan and Duff 1998), benefits of riparian wetland restoration focused on the river Elbe in Germany (Meyerhoff and Dehnhardt 2007), and factors that lead to community participation in mangrove restoration in India (Stone et al. 2008) and CE studies of preferences over river restoration (Weber and Stewart 2009), the socioeconomic factors and psychometric measures that effect wetland restoration (Milon and Scrogin 2006), and WTP for the conversion of cropland to forest and grassland program in northwestern China (Wang et al. 2007). Much of this literature uses CV methodology and therefore is unable to identify the structure of consumer preferences over various facets of ecosystem services.
A single measure of conservation success, such as species richness or the number of endangered species, has been used in many ecological and protected areas selection studies (Csuti et al. 1997; Ando et al. 1998; Haight, Revelle, and Snyder 2000; (Onal 2004; Cabeza et al. 2004; (Onal and Briers 2005; Pressey et al. 2007; Kharouba and Kerr 2010). Studies such as those by Loomis and Larson (1994) and Fletcher and Koford (2002) demonstrate that wildlife population density is also an important variable affecting the public's WTP for habitats. Christie et al. (2006) study public preferences and WTP for biodiversity in general, and Meyerhoff, Liebe, and Hartje (2009) find that the species richness is a significant attribute that determines the WTP for forest conservation. However, neither of the above studies includes wildlife population density as an attribute, making it difficult to understand the role that each of these attributes plays in determining the WTP for restoration projects.
Much of the nonmarket valuation literature on conservation and restoration has focused on wetland preservation and restoration (Heimlich et al. 1998; Woodward and Wui 2001; Boyer and Polasky 2004), forest preservation and restoration (Adger et al. 1995; Lehtonen et al. 2003), the protection of individual endangered bird species (Loomis and Ekstrand 1997; Bowles 1998), or recreation and hunting (Boxall et al. 1996; Roe, Boyle, and Teisl 1996; Hanley, Wright, and Koop 2002; Horne and Petajisto 2003). To our knowledge, no economic valuation study to date has analyzed preferences for grassland ecosystems. The closest study is a paper by Earnhart (2006) that estimates the aesthetic benefits generated by open space adjacent to residential locations, where the open space is denoted by prairie, but this paper does not analyze the preferences for characteristics of grassland ecosystems nor the WTP to restore grasslands.
Identifying whether existing and new environmental public goods act as substitutes or complements, especially with regard to restoration of ecosystems and natural habitats, will enable conservation organizations to better target conservation efforts. Carson, Flores, and Meade (2001) discuss how the WTP will decrease as more of a public good is provided. This follows from the idea of a downwardsloping demand function. At the same time, the presence of an environmental public good can lead to learning, experience, and appreciation such that agents who currently experience high levels of the public good may have a higher WTP for more of that good (Cameron and Englin 1997; O'Hara and Stagl 2002). This can be explained using endogenous preference theory, which argues that consumers who are familiar with a good may be willing to pay more than consumers who are unfamiliar with the good (Bowles 1998; O'Hara and Stagl 2002; Zizzo 2003; Gowdy 2004). Cameron and Englin (1997) show that experience can lead to higher resource values, using a CV study of WTP for trout fishing. They find that experience, measured by the number of years in which the respondent has gone fishing, has a significant positive impact on the WTP. A related theory of planned behavior proposed by Ajzen (1991) states that WTP is expected to increase with a more favorable attitude toward paying for a good (Liebe, Preisendorfer, and Meyerhoff 2011). Therefore, if a favorable attitude toward grasslands can arise from opportunities to experience existing nearby grasslands, respondents with grasslands nearby may have a higher WTP to restore a new grassland.
III. Background on Grassland Ecosystems
Grasslands are open land areas where grasses and various species of wildflowers are the main vegetation. In North America there are three main types of grassland ecosystems. The short-grass ecosystem predominantly occurs on the western and more arid side of the Great Plains. The mixed-grass ecosystem is located farther to the east. The tall-grass ecosystem occurs on the eastern side of the Great Plains. Tall grass can grow 4 to 6 feet. Figure 1 presents the distribution of grassland ecosystems in North America.
Grassland Types in North America
Source: Modified from image obtained from Wikimedia Commons (http://commons.wikimedia.org/wiki/File:US_Great_Plains_Map.svg).
The loss of grassland in North America is attributed to deforestation in the eastern United States, fragmentation and replacement of prairie vegetation with a modern agricultural landscape, and large-scale deterioration of western U.S. rangelands (Brennan and Kuvlesky 2005).1 The loss of grassland ecosystems in most areas of North America has exceeded 80% since the mid-1800s (Knopf 1994; Noss, LaRoe, and Scott 1995; Brennan and Kuvlesky 2005). As depicted in Figure 2, Illinois has lost 99.9% of its original prairie since the early 1800s.
Grasslands in Illinois: Map of Historic Prairies from 1820 (left); Map of Rural Grassland from 2000 (right) Source, left: Adapted from Anderson 1970; Source, right: Data from IDA 2003.
North American prairies are a major priority in biodiversity conservation (Samson and Knopf 1994). The loss of grasslands has contributed to a widespread and ongoing decline of bird populations that have affinities for grassland and grass shrub habitats (Vickery and Herkert 1999; Askins and Zickefoose 2002; Brennan and Kuvlesky 2005). An analysis of the Breeding Bird Survey routes between 1966 and 2002 showed that only 3 of 28 species of grassland specialists increased significantly, while 17 species decreased significantly (Sauer, Hines, and Fallon 2003). In Illinois, which has lost 99.9% of it's original grasslands, grassland songbird populations declined between 75% and 95% during the 25-year period ending in 1984 (Heaton 2000). Vickery and Herkert (1999) state that given the extent of the decrease in grassland habitat, widespread restoration of grasslands throughout the United States is the most effective approach to restoring bird populations.
To address these growing concerns, ecologists and conservation biologists are engaged in restoring grassland habitats to protect endangered flora and fauna. Restoration ecologists have the ability to structure the restoration to emphasize certain attributes in restored ecosystems, but such restoration projects are currently informed by knowledge only from the physical, biological, and ecological sciences (Hatch et al. 1999; Howe and Brown 1999; Fletcher and Koford 2002; Martin, Moloney, and Wilsey 2005; Martin and Wilsey 2006). One exception is Murdoch et al.'s (2010) work, which presents an economic model that incorporates costs and benefits, but it does not take public values into consideration. Restoration planners must make choices about exactly how and where to carry out ecological restoration, and those choices entail physical trade-offs between the exact types of restored ecosystems that result, the kinds of animals and plants that inhabit the restored areas, the variety of species supported by the project, the density of wildlife populations that will be present, and the types of management tools used to maintain these areas. These choices must currently be made in an absence of knowledge about public preferences regarding the characteristics of grassland restoration projects.
IV. Methodology
CE Surveys
CE surveys are used by economists to elicit public preferences for environmental goods and policies that are typically not related to existing markets (Boxall et al. 1996; Louviere, Hensher, and Swait 2000). CE surveys are based on Lancaster's (1966) consumer theory that consumers obtain utility from the characteristics of goods rather than the good itself. Therefore, CEs can be considered the analog of hedonic analysis for stated preference valuation methods. Though CE surveys are more complex to analyze and implement than CV studies, they allow the researcher to gain a detailed understanding of the respondents' preferences for the policy or scenario being analyzed. Unlike CV surveys, CE surveys allow the calculation of part worth utilities for the attributes of a good, which is necessary to answer the research questions in this paper. Hanley, Mourato, and Wright (2001), Hensher, Rose, and Greene (2005), and Hoyos (2010) provide reviews of the CE methodology.
In a typical CE survey, the respondent repeatedly chooses the best option from several hypothetical choices that have varying values for important attributes. Experiment design techniques are used to identify combinations of attributes and levels to create the question sets appearing on each survey.
Survey Instrument
The survey for this particular study presents respondents with opportunities to express preferences over pairs of hypothetical restored grasslands that have the following attributes: species richness, wildlife population density, number of endangered species, frequency of prescribed burning, prevalence of wildflowers, distance to the site from the respondent's house, and cost. Some attributes were motivated by our intent to explore preferences regarding common measures of conservation success. The exact list of grassland attributes was refined after studying the grassland restoration literature and non-market valuation literature and analyzing the results from the focus groups.
Providing information about the presence of birds has been found to have significant effects on WTP for urban green space and native species restoration (Caula 2009; Kaval and Roskruge 2009). Therefore, we include information about bird species in the survey. Gourlay and Slee (1998) find that wildflowers are valued “highly” or “very highly” in landscapes; since wildflowers are an integral part of grassland ecosystems, we include the area covered by wildflowers as an attribute. Historically, fire has been a natural component of grassland ecosystems, and many grassland restoration efforts require management by fire to prevent woody succession and to eliminate invasive species (Vogl 1979; Schramm 1990; Howe 1995; Copeland, Sluis, and Howe 2002). At the same time, smoke and ash from prescribed burns can be hazardous to motorists and become a problem for local residents. We include the use of prescribed burns as an attribute in the survey to see which effect dominates preferences.
In an effort to keep the number of attributes and the experimental design manageable, we do not include characteristics of grasslands that are not directly important to the research questions as variable attributes. Instead we keep these attributes fixed for all choice sets and inform the respondents of this in the survey instructions. The first such feature is size. In discussions we had with them, grassland managers and ecologists emphasized that a natural grassland ecosystem requires a grassland that is approximately 100 acres or larger. Given that the focus of this work is not on the scale and scope effects, we chose not to make size a variable attribute. Instead, we state that every grassland restoration project takes 100 acres (about 92 football fields) of unused marginal lands in the general description. The second fixed feature is access. Recreation is an important consideration when respondents think about the value of protected areas; we state that every restored grassland will be a place where people can go hiking, biking, and bird watching and that the grasslands will not allow hunting and fishing within the grassland. The third fixed feature is labeling. The amount of information present at a site can influence respondents' willingness to pay; therefore we state that every hypothetical grassland will have signs with information about the plants and birds that will live in the grassland area (but will not have a staffed visitor center).
Once an initial list of attributes was developed, we conducted informal focus groups with potential survey respondents and discussed the survey with ecologists and land managers at grasslands. Formal pre-tests of the survey were then conducted at the University of Illinois. The final survey instrument contains background information about grasslands, a description of the attributes and the levels (Table 1), seven sets of binary choice question sets, and a small demographic questionnaire. The Appendix contains an example of one choice question. For each of the binary choice sets the respondents choose between the two given alternatives and the status quo option. The choices will contain different features of the restored area and specific values for these features.
Attributes and Levels for the Survey Instrument
The demographic questionnaire has two questions regarding the presence of nearby grasslands and nongrassland nature areas.2 The presence of a grassland or a nongrassland nature area allows us to identify which respondents are aware of existing natural areas. The answers to these questions are used to test whether the presence of near by grasslands and nature areas has a significant impact on WTP to provide a new grassland. We use self-reported presence of natural areas as a proxy for respondent experience with and knowledge of grasslands and natural areas.
The survey was mailed to a random sample of 2,000 addresses in Illinois, stratified according to population density. The addresses were obtained from the Survey Research Lab at the University of Illinois. The addresses were oversampled from two counties known to have existing grasslands and two counties known not to have any existing grasslands. One dollar bills were included in half of the surveys to increase the survey response rate.
Empirical Design
We followed standard practice in the choice modeling literature (Adamowicz et al. 1997, 1998; Louviere, Hensher, and Swait 2000) and created an efficient experiment design that will allow both main effects and interaction effects to be estimated. Given that we are interested in studying the interaction effects between different indicators of conservation success, the design incorporates pairwise interactions between species richness, population density, and number of endangered species.
The design for the seven attributes was generated following Kuhfeld (2010).3 The design achieves a 99.57% D-efficiency and can be implemented with 54 choice profiles4 (Vermeulen, Goos, and Vandebroek 2008; Kuhfeld 2010). We created a block design where the 54 choice sets were separated into blocks of six choice profiles, giving nine unique surveys with six questions each. Carlsson, Mork-bak, and Olsen (2010) test for learning and ordering effects in CE surveys and show that dropping the first choice question can decrease the error variance of estimates. Therefore, we added an additional choice question before the six choice questions and dropped the first choice question when conducting the analyses to account for possible learning effects. In order to account for possible ordering effects, we reversed the order of the questions in half the surveys and obtained 18 unique versions of the survey.
Model and Estimation
Though the standard multinomial logit model has been used in many valuation studies of environmental goods, it assumes that the respondents are homogeneous with regard to their preferences (the βs are identical for all respondents). This is a strong and often invalid assumption. Therefore, we use a mixed multinomial logit (MMNL) model5 that incorporates heterogeneity of preferences (Hensher and Greene 2003; Carlsson, Frykblom, and Liljenstolpe 2003). The MMNL model is derived as follows (Dissanayake 2014). Assuming a linear utility, the utility gained by person q from alternative i in choice situation t is given by
[1]
where Xqit is a vector of nonstochastic explanatory variables. The parameter αqi represents an intrinsic preference for the alternative (also called the alternative specific constant). Following standard practice for logit models, we assume that εqit is independent and identically distributed extreme value type I.
We assume the density of βq is given by f(β\ Ω), where the true parameter of the distribution is given by Ω. The conditional choice probability of alternative i for individual q in choice situation t is logit6 and given by
[2]
The unconditional choice probability for individual q is given by
[3]
The above form allows for the utility coefficients to vary among individuals while remaining constant among the choice situations for each individual (Train 2003; Carlsson, Frykblom, and Liljenstolpe 2003; Hensher, Rose, and Greene 2005). There is no closed form for the above integral; therefore P q needs to be simulated. The unconditional choice probability can be simulated by drawing R drawings of β, βr, from f (β I Ω)f(β I Ω)7 and then averaging the results to get
[4]
In the CE questions, option A and option B are both restoration options that can be viewed as being closer substitutes with each other than with option C, the status quo option (Haaijer, Kamakura, and Wedel 2001; Blaeij, Nunes, and van den Bergh 2007). One method to incorporate this difference in substitution between options is to use an econometric specification for the MMNL model that contains an alternative specific constant that differentiates between the status quo option and choices that represent deviations from the status quo. We do so by using a constant that is equal to one for option A or option B.
An alternative approach is to use a nested logit model to account for difference in substitutability between pairs of options (Haaijer, Kamakura, and Wedel 2001; Blaeij, Nunes, and van den Bergh 2007). Thus, we test the robustness of the results by using a nested logit model in addition to a mixed multinomial model. The nested logit models are derived as follows (Dissanayake 2014). Assume that the I alternatives are divided into N nests (or groups) of similar alternatives, where each nest has In alternatives. Following Manski and McFadden (1981) and Haaijer, Kamakura, and Wedel (2001), the probability that choice i is selected from nest n is given by
[5]
where
[6]
and
[7]
Equation [6] is the standard multinomial logit probability within a nest, and equation [7] is the probability of choosing nest n. The variable λ is the dissimilarity coefficient or the scaling parameter that measures the degree of dissimilarity between the alternatives in the nest. The inclusive value, Vn, can be thought of as the weight attributed to the particular nest and is defined as
[8]
Since the status quo option is a nest with one alternative, in the econometric estimation we set P(n,1)=1 for that nest.
The coefficient estimates for the MMNL model and the nested logit model cannot be interpreted directly. Therefore, we calculate average marginal WTP for a change in each attribute i by dividing the coefficient estimate for each attribute with the coefficient estimate for the cost term, as given in [9]:
[9]
Econometric Specification
We analyze two econometric specifications for both the MMNL and the nested logit models.
The main effects specification is
[10]
The interaction effects specification is
[11]
The specification in equation [11] includes three variables that are pairwise interactions between the conservation success attributes. A significant and positive coefficient on any conservation success interaction term implies that the respondent has higher marginal utility for increases in one conservation success measure when the levels of the other conservation success terms are high. This would imply convex indifference curves between conservation attributes. A significant and negative coefficient on the interaction terms implies the opposite, with concave indifference curves between attributes and utility-maximizing bundles tending to have corner solutions. If the coefficient is insignificant, then the contours are linear; this is the standard implicit assumption of most CE econometric specifications.
Equation [11] also includes terms that interact the cost attribute with person-specific dummy variables that indicate the presence of grasslands and the presence of nongrassland natural areas nearby. These interaction terms allow us to analyze the impact of existing natural areas on the WTP for a new hypothetical grassland. If the coefficient is positive (negative) and significant, this implies that respondents who have a nearby natural area are willing to pay more (less) to restore a new grassland.
The MMNL models were estimated both with and without an alternative specific constant. The nested logit model was estimated without an alternative specific constant because the structure of the model does not require one. For nested logit models with a degenerate nest, the level of the normalization of the scaling parameter can affect the results (Hensher, Rose, and Greene 2005). The nested logit model results were robust to normalizing at the top level and the bottom level; we present the results normalized at the bottom level. All the econometric models were estimated using NLogit 4 (Greene 2007).
Though we oversampled people living in some parts of the state, the sampling probabilities are an exogenous function of the presence of grasslands. Because the presence of grasslands is controlled for in the regression as an exogenous variable, unweighted regressions are not biased (Wooldridge 1999; Solon, Haider, and Wooldridge 2013). As weighting is not necessary to eliminate bias and may reduce the precision of the estimates, we therefore do not use weights in our econometric analysis.
V. Results and Discussion
Out of the 2,000 surveys that were mailed out, 48 were undeliverable. Of those that were delivered, 316 surveys were returned, out of which 275 were complete. The process yielded 1,650 choice question observations, with an overall response rate of 16.19%. Each of the 18 different survey versions was returned at least 10 times.8 This ensures that each of the 54 choice profiles was represented in the final analysis. Of the 316 surveys that were returned, 197 were surveys that included the dollar bill; including the dollar bill increased the response rate by 65%.
Table 2 compares the demographic characteristics of the state and the respondents; the first column presents the mean values for the Illinois population and the second column presents the mean values for the sample, with the standard deviations (where available) indicated within parenthesis. All of the state averages fall within one standard deviation of the sample means, showing our sample to be reasonably representative of adults in the state.
Comparison of State Population and Sample
Table 3 presents results for the main effects specifications for the MMNL with the alternative specific constant (ASC), MMNL without ASC, and nested logit models; Table 4 presents results for the specifications with interactions effects for those three econometric models. The nested logit model has a better fit than the MMNL model for both specifications (as indicated by higher McFadden's pseudo R2 and the lower Akaike information criterion [AIC] values). In both of the nested logit specifications we cannot reject a null hypothesis that the inclusive value parameter is equal to one. Such a finding can indicate that the nested logit model collapses to a standard MNL model or that the nesting structure is incorrect (Hensher, Rose, and Greene 2005). Nonetheless, we continue to present the nested logit results because the current nesting structure is the only possible one for this particular choice question and because the nested logit model appears to be a better fit for the data than the MMNL model.
Regression Results for the Main Effects Specification
Results for the specification with Interaction Terms
In Table 3 the third column of each of the MMNL results indicates that individual heterogeneity is significant in the MNL setting for some attributes. The coefficient results are robust across the three econometric models. The three conservation attributes and wildflowers have positive and significant coefficients; all three conservation measures should be considered in conservation planning and research. Distance and cost are negative and significant; people prefer restoration projects that are inexpensive and close to home. Burning is insignificant in all the specifications. This could be causedby a subset ofthe sample identifying burning as having a positive effect and a subset of the sample identifying burning as having a negative impact.
The main effects regressions are not able to answer questions about substitutability of conservation goods and the impact of existing grasslands on WTP. To consider those issues, we turn to Table 4, which presents the results of three regressions using the specification with interaction effects: the MMNL with ASC, MMNL without ASC, and nested logit. The specification with interaction terms has an unambiguously better fit than the main effects model for the MMNL without ASC and the nested logit specification (as indicated by higher McFadden's pseudo R2 and the lower AIC). For the MMNL with ASC, the specification with interaction terms has a higher McFadden's pseudo R2 and similar AIC values (1.962 vs. 1.961) compared to the main effects specification. The coefficient for the interaction of the cost and the grassland near variables is positive. This implies that respondents who live near existing grassland areas have a higher marginal WTP for each of the attributes. This finding could be evidence of endogenous preferences: individuals who consume and experience a good can have a higher WTP for the good than individuals who have not experienced the good. It could alternatively be argued that this result is caused by locational sorting wherein respondents who have an inherent preference for grasslands choose to live close to them. We note that people with intrinsically high values for grasslands may also have relatively high values for other natural areas, but the interaction effect for nongrassland natural areas being nearby is not significant in the regression; this might imply that the positive coefficient on the interaction of cost with the grassland near dummy is more likely to be caused by endogenous preferences than by sorting.
The two-way interaction term is significant for the species richness × population density variable, though only for two of the three specifications. In the MMNL regression with the ASC, the ASC itself is not significant, but its inclusion increases the standard errors on many of the other variables including the interaction between richness and density; thus, lack of significance of the interaction variable in that regression may just be a result of multicollinearity. It should be noted that interaction models in discrete choice analysis require a high sample size for statistical power. The 275 returned surveys give a total of 1,650 choice scenarios.
The significant and negative interaction term implies that the marginal value of species richness is lower when the level of population density is high and vice versa. This has implications for the structure of the grassland that maximizes consumer utility. Figure 3 shows total WTP as a function of species richness for different levels of population density. Total WTP increases as the value of species richness increases. The three lines in Figure 3 correspond to different levels of population density. As population density increases, the total WTP at each level of species richness increases. With a significant interaction between species richness and population density, the slope of the total WTP-species richness line decreases as the level of population density increases.9 An increase in species richness has a smaller impact on total WTP at high levels of population density than at low levels of population density.
Total Willingness to Pay (WTP) as a Function of Species Richness for Varied Density Levels, Generated Using Regressions from the Nested Logit Model in Table 4
In addition, the significant interaction term between species richness and population density implies that the isovalue curves (a total WTP level set, drawn on the space of conservation attributes, similar to an indifference curve) are nonlinear, as depicted in Figures 4 and 5, which depict the isovalue curve in species richness and population density space (similar to a utility function in two-good space). Figures 4 and 5 contain isovalue curves for a total WTP of $100. The contours show the combination of species richness and population density that yield a total WTP of $100. When the interaction terms are not included in the econometric specification, the resulting isovalue curve is linear, indicating a fixed marginal rate of substitution. When the interaction terms are included, the isovalue curve is concave, indicative of an increasing marginal rate of substitution. Figure 4 is drawn using the full specification, equation [11]. Since all three interaction terms are considered in this specification, the intercepts of the isovalue curves are different. For example the y-axis intercept in Figure 4 corresponds to a point where the species richness value is zero but the values for both the population density and the endangered species attributes are positive. This corresponds to equation [11b]:
Isovalue Curves (Total Willingness to Pay), Generated Using Regression Results from the Nested Logit Model in Table 4 for a Total WTP of $100
As equation [11b] and equation [10] (the specification without interaction terms, the solid line) are not identical (because of the interaction term between density and endangered species), the end points for the two curves are not the same. The curve corresponding to the specification with the interactions terms, which are negative, has the higher end point at the intercepts. This is because a higher amount of population density (species richness) is needed to achieve a total WTP of $100 at the y-axis (x-axis) intercept compared to the curve for the specification without interaction terms. Instead, if we include only the interaction term between the attributes on the axes, by setting the interactions terms with the endangered species variable to zero (Figure 5), the intercepts of the two isovalue curves are the same.
Isovalue Curves When Interaction Terms between Endangered Species and Other Conservation Success Terms Were Set to Zero (Generated as for Figure 4)
Including the interaction terms causes an outward shift in the WTP curve as the negative interaction term leads to the marginal rate of substitution to increase, requiring more of both population density and species richness to achieve the total WTP of $100 as the values for both population density and species richness increase (compared with the specification without interactions terms, which has a constant marginal rate of substitution and therefore a straight total WTP curve).
Marginal WTP values are shown in Table 5.10 The marginal value estimates should not be compared to each other directly since the units for each attribute differ. All three of the conservation success measures (species richness, population density, and endangered species) have significant per household values. A respondent who lives near a grassland is willing to pay between $1.11 and $1.13 each year to have an additional bird species present in the grassland, and between $7.72 and $10.22 for an additional endangered species; while households do value diversity per se, people place greater value on restorations that promote species that are endangered. Increasing the population density of birds in grassland by an additional bird per acre is worth between $1.50 and $1.94. These marginal WTP values decrease by around 30% for respondents without a grassland nearby. An alternative way to present the marginal WTP values is to calculate the amount of each attribute necessary for a WTP of $1. We show this information in Table 6 for the MMNL results.
Marginal Willingness to Pay
Marginal Change per Attribute for Which Consumers Are Willing to Pay $1
Finally, we calculate the total WTP for a hypothetical grassland with realistic attribute values (Table 7). We estimate the total WTP for a hundred-acre hypothetical grassland with 30 different bird species, 15 individual birds per acres, six endangered species, 60% wildflower coverage, and controlled burning once every year; the calculations are for a household that does not have a nongrassland nature area nearby. For a respondent without a grassland nearby the total value estimated ranges between $75 and $105 per household per year, depending on the distance to the grassland. The results indicate that as the distance to the restored grassland increases from 10 miles to 100 miles, total WTP decreases by as much as 26%. For a respondent that currently has a grassland nearby, the values increase by more than 40% to between $98 to $150. Total value estimates are reasonably robust to the econometric methodology used in the regression, especially if the hypothetical restoration project is located close by.
VI. Conclusion
We analyze the structure of public WTP for different attributes of grassland ecosystems using a CE survey. This work yields several findings that have broad implications for conservation planning and environmental valuation. First, we find that several features of an ecosystem that are used as measures of conservation success—species richness, population density, and presence of endangered species—have large positive marginal values. Much of the work on optimal protected-area planning and design uses a single measure of conservation success as the objective to be maximized. Our results imply that when there are physical trade-offs between conservation outcomes (e.g., one can increase the population of a single species such as pheasant, but in doing so one might lower species richness) planners should be careful to consider all conservation success measures in order to maximize the social welfare obtained from conservation and restoration efforts.
Second, we find that respondents who live near existing grassland areas have a higher marginal WTP for restoring additional grasslands. This result contradicts what would be expected if the marginal value of a good declined with its total quantity. This result may reflect the existence of endogenous preferences wherein individuals who consume and experience a good learn to appreciate and enjoy it and can therefore have a higher WTP than individuals who have not experienced the good. The finding could instead be caused by locational sorting, but other features of the regression results lend some support to the former explanation. If this finding is robust to other ecosystems and other locations, it has implications for conservation planning in terms of locating new conservation areas; for example, the welfare maximizing conservation strategy may be to have similar ecosystem types partially clustered in the landscape. This finding also complements findings from the location choice modeling literature, in particular work by Wu (2006), which finds that both development patterns and community characteristics are strongly influenced by the spatial distribution of environmental amenities; Warziniack (2010), which models the endogenous location of open space; Irwin and Bockstael (2004), which finds that smart growth policies can concentrate development and create open space; and Carricin-Flores and Irwin (2004), which finds that the location of new residential development is influenced by preferences for lower density. We are currently unable to answer the question of why respondents who have grasslands nearby are willing to pay more to restore additional grasslands. Future work should analyze whether this result is robust to other ecosystems and other locations, and should develop an empirical strategy to identify the source of this phenomenon.
Total Willingness to Pay as a Function of Distance ($/year)
Third, in an effort to analyze the structure of the preferences for the conservation success attributes in more detail, we use a specification that contains pairwise interactions of the conservation success terms. We find that the interaction term between species richness and population density is significant for two of the three specifications. This implies that respondents value population density less when species richness is high (and vice versa). Thus, the value to society of a project that maximizes species richness will vary across sites that have different levels of wildlife population density (and vice versa). Furthermore, households with such preferences are likely to prefer restoration outcomes that have large numbers of different species or have a large number of individual animals (as opposed to a balanced bundle of both species diversity and abundance of individual animals).
Finally, this study is the first to generate estimates of the total benefits to households of efforts to restore grasslands, an ecosystem type that is disappearing throughout North America. It needs to be emphasized that the survey asked respondents to make choices regarding a grassland of 100 acres. The results are not necessarily linearly scalable; it is important to be aware of scale and scope effects with regard to applying these values to larger grasslands.11 This study provides valuable information to conservation planners and ecologists engaged in restoring and conserving ecosystems regarding the values placed on grasslands by the public. The results allow policy makers to calculate the total WTP for a grassland with varied characteristics. For the plausible grassland described in the results section, the annual value per household ranges between $75 and $150, depending on the distance to the restored grassland and on whether the respondent currently has a grassland nearby.12 This information is especially important in places like Illinois, where some lands could potentially be restored as wetland, tallgrass prairie, or forest with different restoration and management techniques.13 Total WTP estimates can help land managers weigh the benefits of restoring a grassland against the other benefits that society could gain from an area of land.
Acknowledgments
The authors express their gratitude to John B. Braden, Robert Johnston, Dan Lafave, Jeff Brawn, James R. Miller, Jürgen Meyerhoff, Catalina Londoño, Taro Mieno, participants at the AERE Annual Summer Workshop, the AAEA annual meeting, the NAREA annual meeting and the PERE workshop at University of Illinois, and two anonymous referees for comments and suggestions. Funding for this research was provided by the Research Board at the University of Illinois and the Robert Ferber Dissertation Award awarded by the Survey Research Lab at the University of Illinois. This paper is also based in part on work supported by the U.S. Department of Agriculture National Institute of Food and Agriculture, Hatch project #ILLU 05-0305.
Appendix
Sample Choice Question
Footnotes
The authors are, respectively, assistant professor, Department of Economics, Colby College, Waterville, Maine, and Department of Economics, Portland State University, Portland, Oregon; and professor, Department of Agricultural and Consumer Economics, University of Illinois at Urbana-Champaign, Urbana.
↵1 Though this research is analyzing values to restore grasslands in North America, the loss of grasslands is occurring throughout the world and there are restoration efforts that are being undertaken globally (Muller et al. 1998; Pywell et al. 2002; Walker et al. 2004; Murdoch et al. 2010; Torok et al. 2011).
↵2 Question 4: Is there a grassland nearby that you can visitŒ (Yes, No, Not Sure); Question 5: Is there another nature reserve that is not a grassland (such as a forest) nearby that you can visitŒ (Yes, No, Not Sure).
↵3 The experiment design was conducted using the SAS experiment design macro (Kuhfeld 2010).
↵4 D-efficiency is the most common criterion for evaluating linear designs. D-efficiency minimizes the generalized variance of the parameter estimates given by D = det [V(X, ß)1/•] where V(X, ß) is the variance-covariance matrix and k is the number of parameters. Huber and Zwerina (1996) identify four criteria (orthogonality, level balance, minimum overlap, and utility balance) that are required for a D-efficient experiment design.
↵5 This approach is also referred to as the mixed logit, hybrid logit, random parameter logit, and random coefficient logit model.
↵6 The remaining error term is the iid extreme value.
↵7 Typically, f(βIΩ) is assumed to be either normal or log-normal, but it needs to be noted that the results are sensitive to the choice of the distribution.
↵8 On average each survey version was returned 16.6 times, with a standard deviation of 0.88, a minimum of 10, and a maximum of 24.
↵9 With no interaction terms, the increase in total WTP caused by higher population density is the same at every species richness level (the lines are parallel).
↵10 The marginal values for each of the conservation success variables are dependent on the values of the other conservation success variables (due to the interaction terms). The marginal values were calculated at the mean values of the conservation success terms (species richness = 30, population density = 7, endangered species = 3).
↵11 For a discussion and literature review of scale and score issues in valuation we refer the reader to Czajkowski and Hanley (2009) and Desvousges, Mathews, and Train (2012).
↵12 For a 100 acre hypothetical grassland with 30 different bird species, 15 individual birds per acres, six endangered species, 60% wildflower coverage, and controlled burning once every year.
↵13 To put those values in context, we list here value estimates that have been obtained for other ecosystems. Boyer and Polasky (2004) give examples of stated preference surveys that yield WTP for wetlands in the range of $15 (1987 dollars) to $87 (1998 dollars) per hectare per year. Brander et al. (2006) conduct a comprehensive summary of stated preference studies on wetlands and find the median WTP is approximately $200 (1995 dollars) per hectare per year.
References
In this issue
Jump to section
Related Articles
Cited By...
- Early Exposure to Nature and Willingness to Pay for It: The Value of Tallgrass Prairie Grassland Restoration
- Early Exposure to Nature and Willingness-To-Pay for It: The Value of Tallgrass Prairie Grassland Restoration
- Quality over Quantity: Nonmarket Values of Restoring Coastal Dunes in the U.S. Pacific Northwest