Research ArticleArticles

Spatial Preference Heterogeneity: A Choice Experiment

Roy Brouwer, Julia Martin-Ortega and Julio Berbel
Land Economics, August 2010, 86 (3) 552-568; DOI: https://doi.org/10.3368/le.86.3.552

Abstract

The main objective is to assess preference heterogeneity related to the spatial distribution of water quality improvements throughout a river basin. In a choice experiment, the river basin’s hydrogeographical units and the levels of water quality improvement are included as attributes in the experimental design. Changes in water quality throughout the river basin are visualized with maps and modeled simultaneously in relation to where respondents live, in a random utility model. Not accounting for spatial preference heterogeneity results, in this case study, in an underestimation of welfare when aggregating willingness-to-pay values from subbasins to the river basin as a whole. (JEL Q25, Q51)

I. Introduction

Choice modeling has become increasingly popular in the economic valuation domain, including water and water quality improvements (e.g., Morrison and Bennett 2004; Hanley, Wright, and Alvarez-Farizo 2006). Choice models (CMs) have a number of distinct advantages compared to contingent valuation (CV), including the “internal” scope test (Hanley, Mourato, and Wright 2001), and superior conditions for benefits transfer (Morrison et al. 2002). Boxall et al. (1996) argue that CMs offer the opportunity to explore in detail trade-offs between substitutes and corresponding preferences. This is where the interest and focus of this paper lies. Preferences and values for water quality improvements at specific locations in a river basin are expected to be, at least in part, determined by substitution possibilities elsewhere in the basin. The availability of substitutes influences scarcity conditions and, hence, willingness to pay (WTP) for a good or service. CMs have the convenient property that marginal rates of substitution constitute an integral part of the estimated random utility model (e.g. Train 2003; Hensher, Rose, and Greene 2005). That is, the marginal rate of substitution reflects the rate of change in one attribute, in this case, for example, water quality at location A, relative to the rate of change in a second attribute, namely, water quality at location B.

This study’s main objective is to assess spatial preference heterogeneity in choice behavior related to water quality improvements throughout a river basin. Johnston, Swallow, and Bauer (2002) observe that stated preference studies rarely incorporate spatial attributes, which are relevant to welfare estimation. Preferences for water quality improvements are measured and modeled here while accounting for changes in water quality in other places than those where respondents live. Spatial preferences refer to the central hypothesis tested in this paper that people living in a river basin are expected to value water quality improvements differently depending on where they live. Spatial choices are defined as choices among particular alternatives at specified locations (e.g., Fotheringham 1988).

In a CV study testing part-whole bias in the context of instream flow protection, Brown and Duffield (1995) show that WTP is sensitive to the information provided about the number of rivers protected. In this study, we look at the value of water quality improvements in a wider river basin context by presenting water quality improvement alternatives to respondents in different parts of the river basin simultaneously. In a choice experiment, these alternatives are represented by maps depicting changes in water quality levels across the whole river basin. The hydrogeographical units making up the river basin and the levels of water quality improvement are included as attributes in the CM.

The approach presented here differs, for example, also from conventional testing of distance-decay effects in stated preference research (e.g., Bateman et al. 2006), in that variation is introduced not only in the spatial distribution of respondents who benefit from the change in environmental good provision in relation to one specific site or area (e.g., lake or river stretch), but in relation to multiple locations at the same time without relying upon unidirectional measurement units such as kilometers. So, we look at the value of water quality improvements in different subbasins (parts) at the same time, which together make up the (whole) river basin, hence enlarging the choice set. To this end, we model choice behavior in relation to alternatives represented in a geographical map, depicting spatially differentiated changes in water quality levels across the river basin.

The relevance of the work presented here is found in the design of appropriate rules of aggregation when aiming to estimate a total economic value for an environmental change such as water quality improvements, where both the environmental good in question, and the population of beneficiaries and their preferences are expected to be nonuniformly distributed over space. In this case study we pay special attention to the influence of an individual’s place of residence (i.e., subbasin) in a river basin in relation to water quality changes inside and outside the individual’s subbasin. The maps in the choice experiment are used as a visual aid to help respondents see where they live in relation to the different locations of the proposed water quality improvements. The policy context of the study is the European Water Framework Directive (WFD). The WFD distinguishes different water quality levels from poor to very good and requires that the economic analysis of water quality improvement be carried out both at riverbasin and water-body scale (e.g., Brouwer 2008).

II. The Choice Experiment

Choice experiments such as the one presented here have their roots in random utility theory (e.g., McFadden 1974; Ben-Akiva and Lerman 1985). The multinomial logit (MNL) model is the most commonly used structure for choice models but is often rather restrictive in practice. The random parameters logit (RPL) or mixed logit model is more flexible and relaxes the assumption of independence of irrelevant alternatives (IIA) as a result of the iid property underlying MNL (Train 2003), and allows for preference heterogeneity, the subject of study in this paper. According to McFadden and Train (2000), any random utility model can be approximated by a mixed logit model. The standard indirect utility function underlying the mixed logit model is

Embedded Image[1]

where Uij refers to the utility of individual i obtained from choice alternative j, Vij is the measurable component of utility measured through a vector of utility coefficients β associated with a vector of observed attribute and individual characteristics Xij, and εij captures the unobserved influences on an individual’s choice with an iid extreme value distribution. The utility coefficients β vary according to individual (hence βi) with density f(β). This density can be a function of any set of parameters and represents in this case the mean and covariance of β in the sample population:

Embedded Image[2]

Mixed logit models assume heterogeneity to be continuous over the interval spanned by the assumed distribution for the taste parameters (Scarpa, Willis, and Acutt 2005). Alternatively, a finite mixture approach can be used, such as latent class (LC) models (e.g., Nylund, Asparouhov, and Muthen 2007). Here, preferences are assumed to be homogenous within a limited number of classes, but preferences vary between classes. Hence, the utility of individual i attached to choice alternative j depends on class membership s:

Embedded Image[3]

Besides the choice alternative equation, the LC model also consists of a classmembership equation, estimated through the same underlying MNL process as the choice equation, consisting of respondent instead of attribute characteristics. Here, we will compare results using both models. For a more detailed comparison of mixed logit and LC models, see, for example, work by Green and Hensher (2003), Provencher and Bishop (2004), Scarpa, Willis, and Acutt (2005), or Hynes, Hanley, and Scarpa (2008).

In this case study, alternatives are defined in terms of water quality improvements in specific subbasins throughout a river basin. The water quality improvements can be realized at a certain price to be paid through the household water bill and are presented as possible policy scenarios between which people are asked to choose compared to the baseline scenario of expected lower future water quality levels throughout the whole basin and doing nothing extra to improve water quality. The cost in this baseline alternative is therefore zero. Based on the choice design used in this case study, equation [1] can be rewritten as (suppressing the notation of random terms for the sake of simplicity)

Embedded Image[4]

where β0 is the alternative specific constant, βqs the vector of coefficients attached to the water quality attributes Q in subbasin s, βp the price vector, and βy the vector of coefficients related to the individual’s socio-economic characteristics Yi (e.g., income). The inclusion of a monetary attribute (price) in the choice model allows for the estimation of monetary Hicksian welfare measures for different water policy scenarios and changes in individual components of these scenarios (e.g., Hensher, Rose, and Greene 2005). The marginal rate of substitution for a change in the water quality attribute Q in a specific subbasin s (Qs) is estimated as follows:

Embedded Image[5]

where H is the compensating variation for a marginal change in water quality in subbasin s, and the price attribute is interpreted as the marginal utility of income.

The central hypothesis tested in this study is that people value water quality improvements more highly if the improvement occurs in the area where they live, namely, in their own subbasin compared to the same improvement in another subbasin (keeping all else constant). More formally, this translates to the following test of spatial preferences:

Embedded Image[6]

where Vijt is the utility of respondent i living in subbasin τ associated with choice alternative j, Qτ the change of water quality in subbasin τ, and Qv the same change of water quality in subbasin v. Note that we do not expect the value attached by respondent i living in subbasin τ to a water quality improvement in subbasin v to be zero. People are expected to value improvements in water quality outside their own subbasin, too, for example, because they visit recreation sites outside the area where they live, or value water quality improvements elsewhere for non-use reasons.

The implication of the central hypothesis tested in this paper is that we expect the economic value for a water quality improvement across a river basin not simply to be equal to the sum of the value attached to the parts by local residents. Accounting for spatial preferences and distinguishing between local and nonlocal residents results in the hypothesis

Embedded Image[7]

where HΩ,s refers to the compensating surplus of a water quality change across the entire set of subbasins in the river basin, and HΩ,t to the same water quality change in subbasin t (t being a subset of the whole set of subbasins s in the river basin). Note that this hypothesis is related to, but not the same as, traditional testing of part-whole bias in the stated preference literature (e.g., Carson, Flores, and Meade 2001). Although water quality changes are varied across the subbasins, respondents are always asked to value the whole river basin. Furthermore, the parts are not valued separately, except that we analyze the value attached to subbasins by residents and nonresidents.

III. Case Study Description

The case study is carried out in the Guadalquivir River basin (GRB). The Guadalquivir River is the longest river in the south of Spain, with a length of around 650 km. It originates in the southeast in the Sierra de Cazorla and empties into the Atlantic Ocean (Gulf of Cadiz). The GRB is an interesting case study because it is perceived and referred to by the population living in the basin as a clearly defined geographical unit (Moyano et al. 2004; Martin-Ortega 2008). The basin has been extensively studied in the context of the WFD (e.g., Martin-Ortega, Gutierrez Martin, and Berbel 2008) and covers an area of 57,527 km2 with a population of over 4 million people. The basin has a Mediterranean climate with a heterogeneous precipitation distribution. Annual average temperature is 16.8uC and average precipitation 630 mm. Land cover and land use consist of forests (49.1%), agriculture (47.2%), urban areas (1.9%), and wetlands (1.8%). Natural annual flow levels are 6,900 Hm3 for surface water and 2,576 Hm3 for groundwater (CHG 2008). About half of these water flows are used for human consumption, mainly in agriculture (.80%). Per capita water consumption in the GRB in 2005 was 1,600 m3. Water consumption is expected to increase by 5% in the coming years (Gutierrez, Martin-Ortega, and Berbel 2008).

The variability in water resource availability, the increasing demand from different water users, and the recurrent droughts lead to cyclical scarcity episodes. Local and seasonal droughts result in aquifer salinization and environmental stress. Water quality is a significant problem throughout the river basin. The main sources of pollution include urban and industrial wastewater discharge, erosion, and nutrient and pesticide runoff from agricultural land. Concentration levels of nitrogen, phosphorus, heavy metals, and organic pollutants in surface- and groundwaters are expected to increase by about 30% in the near future (CHG 2008).

In this study, the GRB is divided into four distinct subbasins (Figure 1): Sierra Morena and Alto Guadalquivir in the north (hereafter “Alto”), Valle del Guadalquivir in the center of the basin (hereafter “Valle”), Campiña in the south and the Doñana National Park at the mouth of the Guadalquivir river in the southwest (hereafter “Doñana”). The subbasins differ in terms of water quality levels, but also with regards to the homogeneity of the landscape, land use, and population characteristics. Alto can be characterized as a mountainous, sparsely populated, extensively agricultural area. Valle is the valley through which the main Guadalquivir river stream flows and is highly fertile with intensive agricultural land use and the highest population density, concentrated in the cities Sevilla, Cordoba, and Granada, and some of the biggest industrial areas in the region. Campiña has a low-land agricultural landscape with a number of medium-sized cities. Doñana is home to a wide variety of rare protected species such as the Iberian lynx and the imperial eagle.

Figure 1.
Figure 1.

Current Water Quality Levels in the Guadalquivir River Basin

Using the categorization of water quality levels in the WFD, the current situation in the GRB is presented in Figure 1. The map is based on the WFD Article = report for the GRB and was developed in collaboration with the Confederation Hidrografica del Guadalquivir (CHG). Only the northern part of the basin (Alto Guadalquivir) is currently in good condition. The central, southeastern, and western parts are in moderate and poor condition.

IV. Survey Design

In order to obtain the necessary data to estimate the CM, more than 600 face-to-face interviews were conducted by a local marketing company in October 2006 throughout the GRB, targeting a random sample ofthe urban and rural population in nine different municipalities more or less equally distributed across the three populated subbasins Alto (n = 210), Valle (n = 259), and Campiña (n = 150). The questionnaire was pretested at length over a period of = months based on two focus groups and 100 face-to-face interviews. The information provided in the questionnaire, including the maps displaying current water quality levels throughout the basin and the water quality ladder (see below), was developed and tested in collaboration with the water experts of the GRB authority CHG.

The questionnaire used in the survey consists of three main parts:

  • Respondent awareness of water-related problems and professional and recreational water experiences, respondent perception and opinion related to water quality and water quality improvements aimed at testing the internal consistency of stated preferences based on the theory of reasoned action (Fishbein and Ajzen 1975; Ajzen and Fishbein 1980)

  • Respondent preferences and values toward water quality improvements in the GRB and its subbasins elicited through the choice experiment

  • Respondent demographic and socioeconomic characteristics, aimed at testing preference heterogeneity

The primary interest here is the choice experiment. In the choice experiment respondents were presented with four choice tasks using four different cards, each card showing three maps ofthe GRBreferringto situation A, B, and the current situation with different water quality levels in each subbasin, and a cost price over and above the household current water bill (Figure 2). The water quality levels (poor, moderate, good, very good) are based on the categorization of water bodies in the WFD1 and described along the lines of the U.S. Environmental Protection Agency water quality ladder (e.g., Carson and Mitchell 1993) in terms of their consequences for different types ofwater use and environmental risk. This approach was understood best by the lay public. During the pretests, a technical description of water quality levels based on physical indicators appeared too hard to understand for the general public, resulting in nonsignificance of the attributes, as described by Hanley, Wright, and Alvarez-Farizo (2006). Moderate water quality is described as suitable for sprinkling gardens and irrigation, good water quality as suitable for recreation such as swimming and fishing, and very good water quality as suitable for drinking water, reflecting at the same time a high ecological status of the water environment. The water quality improvements in Doñana mainly benefit the environment and wildlife and were also presented as such in the experiment. As in the original water quality ladder applied by the U.S. Environmental Protection Agency, each level includes the quality characteristics of the level below.

Figure 2.
Figure 2.

Example of a Choice Card

The choice experiment is based on an orthogonal fractional factorial representation of a resolution III main effects design (Addelman 1962) consisting of 36 choice sets blocked in nine versions offour cards.2 Situation A and B represent improvements with respect to the baseline situation; that is, water quality in each subbasin is equal to or better than the baseline situation in each subbasin. This results in a one-level improvement for Alto (from good to very good), two levels of improvement in the case of Valle (from moderate to good and from moderate to very good), three levels of improvement for Campiña (from poor to moderate, from poor to good, and from poor to very good), and two levels of improvement in the case of Doñana (from moderate to good and from moderate to very good). The cost price has six different levels (€10, €25, €50, €75, €100, €150).

V. Results

Sample Characteristics

The main demographic and socioeconomic characteristics of the sample (n = 619) are presented in Table 1. The sample provides a cross section of the total population in the GRB, as can be seen from the river basin population statistics.

View this table:
Table 1

Sample and Population Characteristics

Average household income is higher (10%) in the sample than in the population from which it was drawn. The same applies for the number of people with a higher education level, suggesting some degree of self-selection. In all other respects the sample is considered representative for the river basin population as a whole. overall, around a quarter of all respondents are related to agriculture, either directly through employment or indirectly through family and relatives. This finding is consistent with the rural nature of large parts of the river basin, and important for this study in view of the relevance of water in agriculture in this part of Spain (e.g., Moyano et al. 2004). Almost 60% of the sample population is employed in trades and the service sector, and just over 15% in public administration. Although a majority of 60% claim to be interested in the environment in general, only 1% are an active member or donator to an environmental protection organization.

Water is a very important issue for most sample respondents. over 90 percent consider water issues one of the most important problems facing Spain. Sixty percent have suffered water cuts in the past. Forty percent perceive water quality in the GRB as poor. Less than 25% of the sample population recreate in the Guadalquivir river and its tributaries, for example, fishing and swimming. Forty percent are well aware of their current water bill, stating an average value for the current water bill (€250/household/year), which is not too far off the average household payments reported by the Spanish Ministry for the Environment (Ministerio de Medio Ambiente 2007).

Small but statistically significant positive correlations are found between some of the awareness, attitude, and perception variables, such as respondent environmental concern and awareness of the current water quality situation in the GRB (the more concerned, the more aware), or experience with water cuts and respondent perception of the water problems (the more experience, the more water is perceived as a problem). As expected, a higher score is found for the degree of familiarity with the area’s environmental characteristics, including water quality levels, if the area refers to the subbasin where the respondent lives and works. Forty-three percent of the sample population have visited the Doñana National Park, and more than 80% believe they will visit the park in the future. A majority of 92% of the sample considers the future protection of the Doñana important. We now turn to the question of to what extent respondents are also willing to financially commit themselves to the improvement of water quality throughout the GRB.

Estimated Choice Model

Three different random utility choice models were estimated, the results of which are presented in this section.3 The overall fit ofthe models as measured through McFadden’s R-squared is good (see, e.g., Hensher and Johnson 1981). The first RPL model simply examines the effect of the attributes on choice behavior, while the second and third are more extended models that account for preference heterogeneity in the valuation of the water quality improvements throughout the river basin (Table 2). The second RPL model is the same as the first RPL model but accounts for spatial preference heterogeneity through the inclusions ofinteraction terms between attributes for water quality levels in specific subbasins and respondents’ place of residence. The RPL models are estimated using 100 halton draws (Bhat 2001) and assuming normal distributed random parameters.4 The third model is the LC model, where a respondent’s place of residence is included as a possible factor driving segmentation. The number of appropriate classes was identified based on the Akaike information criterion (AIC) and the Bayesian information criterion (BIC). The former is lowest for five classes and the latter in the case of three classes (Table A1). Following the recommendation by Scarpa and Thiene (2005), we therefore also examined the significance of the attributes across classes and found that the four-class model is in that case the preferred model. Class membership is highest for Class 2 (71%), followed by Classes 1 and 3 (both 12%) and Class 4 (5%).

View this table:
Table 2

Estimated Random Utility Choice Models

All water quality-related attributes in the first basic “attributes only” RPL model are, as expected, positive and statistically significant at the 10% level, except for the change in Doñana from the current moderate state to a good state. The cost price has the expected negative sign, implying that an alternative is less likely to be chosen if the cost is higher. The significant positive constant term indicates that respondents prefer a change in water quality in the GRB compared to the baseline situation.5 Except for good water quality in Valle and moderate water quality in Campiña, the standard deviations of the attributes are statistically significant at the 5% level, suggesting the presence of preference heterogeneity. The results show that the marginal utilities related to improvements in water quality across the subbasins increase as the change in water quality increases, reflecting sensitivity to scope. The parameter estimates for different water quality levels in a subbasin are statistically significant based on the Wald test, except for the difference between moderate and good water quality in Campiña (Table A2).

The highest marginal value is found for the change from a poor to a very good state in Campiña, and the lowest value for the change from a good to a very good state in Alto. This is an interesting finding, as it confirms that the biggest improvement in water quality (from poor to very good) has the highest value, and the smallest improvement (from good to very good) the lowest value. An important observation from these findings is, furthermore, that the value of water quality improvements is spatially not uniformly distributed, partly due to differences in baseline conditions, but also because of significant preference heterogeneity. This is what the results related to the change from a moderate to a very good state in Doñana and Valle, for example, suggest. The baseline conditions are the same in these two subbasins, while the estimated values are significantly different (Table A2).

In the second and third extended RPL and LC models, we test to what extent it matters where respondents live as to what value they attach to water quality improvements across the basin. Although the outcome of the Ben-Akiva and Swait (1986) test for nonnested choice models indicates that the LC model outperforms the extended second RPL model,6 class membership can hardly be explained by observed respondent characteristics, such as the subbasin where respondents live or their household income level.7 The RPL models are therefore more instructive than the LC model and will be used here to test the study’s central hypothesis. More specifically, we are interested in finding out whether respondents value their own subbasin more than the other subbasins making up the GRB. Besides the subbasin where respondents live, household income is the only other respondent characteristic that appeared to be a significant determinant of choice behavior. None of the other demographic respondent characteristics (e.g., gender, age, household size, education level) or respondent attitude characteristics has a significant impact on water quality preferences. Given the fact that no people live inside Doñana, water quality levels in Doñana were interacted with the variable whether or not someone ever visited or plans to visit the Doñana in the near future, but these interaction terms did not significantly influence choices either. This suggests that the value attached to Doñana does not merely concern current or future use value, but, as expected, largely non-use values. For consistency and comparability purposes between the extended RPL and LC models, the variable whether or not someone ever visited the Doñana in the past is presented in Table 2 as an interaction term with the alternative specific constant.8

The parameter estimates for the water quality changes in the second extended RPL model are slightly lower than in the first RPL model, except for the quality change to very good in Alto and Doñana. The coefficient for very good water quality in Doñana in the extended model is twice as large as the same coefficient in the basic attributes-only model. The differences between the parameter estimates for the water quality levels in a subbasin are less pronounced and statistically not significant, except for the difference between good and very good water quality in Doñana (Table A2). As in the first RPL model, only very good water quality in Doñana is considered worth paying for. Part-worth utilities remain highest for water quality improvements in Campiña and lowest in Alto. No major changes occur in the random part of the estimated utility function.

Turning to the interaction terms in Table 2, an interesting finding is that none of the interaction terms between place of residence and the change to moderate or good water quality are statistically significant at the 10% level. This implies that the change to moderate or good water quality is valued the same by all respondents throughout the GRB, irrespective of the subbasin where respondents live. On the other hand, place of residence is statistically significant at the 10% level for the change to very good water quality, but only if residents live in the subbasin in which this specific water quality improvement occurs. This leads us to accept the central hypothesis that people living in a river basin value water quality improvements differently depending on where they live. None of the interaction terms between a water quality improvement in one subbasin (e.g., Valle) and place of residence in another subbasin (e.g., Campiña) are significant in the extended RPL model. Hence, residents in one subbasin value water quality improvements in another subbasin, but no more than in their own subbasin. The negative interaction term between very good water quality in Doñana and residents living in Valle indicates that the latter value the improvement of water quality in Doñana significantly less than Alto residents.

So, throughout the entire river basin the change to very good quality is valued in a significant positive way in all subbasins, but local residents place an additional value on this change in the specific subbasin where they live compared to nonresidents. That is, local residents in the subbasin Valle put an extra value on the change of water quality in their subbasin from moderate to very good, and the same applies to the local residents in the subbasin Campiña when they value the change from poor to very good in their subbasin. In conclusion, irrespective ofthe subbasin where people live, the improvement of water quality to a good state is valued equally by local and nonlocal residents, but when improving water quality one level further to very good quality, one has to account for the fact that local residents place an additional value on this improvement if it occurs in their own subbasin. This has important implications for the welfare estimation procedure at river basin scale, as will be shown in the next section.

Estimation of Compensating Surplus Measures

In order to obtain a total economic value, namely, the compensating surplus (CS), for the whole river basin, the marginal values for improving water quality, as per equation [5], are added in a linear way:

Embedded Image[8]

Marginal values and the estimated compensating surplus measures are presented in Table 3. Confidence intervals are calculated based on the Krinsky and Robb (1986) bootstrapping procedure using 1,000 draws for each of the models. The complete combinatorial approach proposed by Poe, Giraud, and Loomis (2005) is subsequently used to test whether the WTP differences across models are significant or not. The null hypothesis of equality of marginal WTP cannot be rejected in most cases when comparing the first and second RPL model. Only the value of the change from moderate to very good water quality in Doñana is significantly different between the two RPL models. The presented WTP values for the LC model are the weighted averages for the four different classes where the probability of belonging to a class is used as the weighting factor. No significant difference exists between the second RPL and LC model results (Table A3), and the focus remains, therefore, as before on differences between the first and second RPL models.

View this table:
Table 3.

Estimated WTP Values (€/Household/Year) Based on the Three Random Utility Choice Models

Of most interest here are the estimated CS measures based on the two RPL models, where the second model includes the spatial interaction terms and, hence, accounts for the fact that respondents living in different parts ofthe basin value the parts making up the whole differently. The 95% confidence interval around the CS for “good” water quality across the whole GRB obtained from the first basic attributes-only RPL model is €83 to €123 per household per year, with mean WTP equal to €102 per household per year. Comparing this value with mean WTP obtained from the second extended RPL model, we find that the difference (32%) is statistically not significant at the 5% level. In the case of a water quality improvement to a “very good” level, we have to take into account the fact that local residents put an extra value on this water quality level when realized in their own subbasin, and add this into the aggregation procedure. This yields a 95% confidence interval around the CS for very good water quality in the whole GRB of €169 to €257 per household per year, with a significantly higher mean WTP value for very good water quality than good water quality of €212 per household per year. We find that when comparing mean WTP from the extended RPL model, where we account for spatial preference heterogeneity, with the value from the first attributes-only model, where we do not account for spatial preference heterogeneity, the former is significantly higher than the latter at the 5% level. Not accounting for spatial preference heterogeneity results in an underestimation of the estimated economic value for improved water quality in the whole river basin of 33%.

VI. Discussion and Conclusions

In this study we apply a novel choice experiment design using maps to elicit nonmarket welfare measures for water quality improvements in the Guadalquivir River basin in the south of Spain. We introduced the concept ofspatial preference heterogeneity, meaning that respondents are expected to value changes in environmental good provision differently depending on where the welfare change takes place in relation to their place of residence. Spatial aspects of environmental change are hardly ever accounted for in the choice experiment literature. The approach used in this study allows us to account for the differential spatial distribution of the improvement of water services and their beneficiaries. Respondents living in different parts of the river basin are asked to value simultaneous water quality changes across different parts ofthe basin, enlarging the choice set in a spatially explicit way and, hence, enabling to more adequately capture possible substitution effects. This is expected to be particularly relevant for the implementation of the WFD, where baseline conditions and the population of beneficiaries vary across European river basins. The WFD furthermore offers the opportunity to lower water quality objectives for all water bodies or delay their realization in time based on the concept of disproportionate costs. For this, a comparison of the expected costs and benefits and their distribution across water bodies and water quality levels is needed. Even though the measures to reach the WFD objectives only have to be cost-effective, insight into the return value of the foreseen large investment programs in water quality improvements across European water bodies helps to prioritize limited available budgets for WFD implementation across water bodies within and between river basins in the next decades.

Our approach supports the derivation of economic welfare measures under circumstances where the physical characteristics of the ecosystem change (i.e., water quality changes in a river flow within a basin), the associated change in water services, and the population of beneficiaries are spatially distributed in a nonuniform way. We find that the marginal values attached to water quality improvements are significantly different in different parts of the river basin. This is partly explained by the existing variation in baseline conditions, and partly by significant spatial preference heterogeneity. Accounting for spatial preference heterogeneity, in other words, where inhabitants in one subbasin also hold values for water quality improvements in other subbasins, these values can simply be added in order to obtain a total economic value for “good” water quality conditions throughout the river basin, but not in the case ofthe highest water quality level (very good), which is what the WFD ultimately aims for. In the latter case we have to include the additional value local residents attach to reaching very good water quality in their own subbasin, yielding a significantly higher aggregate welfare measure of around 30% across the individual subbasins.

Our interpretation of this result is that respondents have preferences for water quality improvements to acceptable levels throughout the entire river basin but are not willing to pay extra to reach a more than good condition elsewhere, only in their own subbasin. This is supported by the fact that we find no significant WTP by nonlocal residents for the improvement of water quality in Alto, the only subbasin in the GRB where water quality levels are already in a good state. On the other hand, in the case of Doñana only very good water quality is good enough. We suspect that this is due to the high ecological value attached to the national park, irrespective of the question whether respondents have visited the national park before or intend to visit it in the future.

Appendix

View this table:
Table A1

Tests of Latent Class Numbers

View this table:
Table A2

Wald Tests of Random Parameters Logit (RPL) Model Coefficient Equality

View this table:
Table A3

Poe Test Results for WTP Values Derived from the Extended RPL and LC Model

Footnotes

  • The authors are, respectively, professor, Department of Environmental Economics, Institute for Environmental Studies, VU University, Amsterdam; postdoctoral researcher, Basque Centre for Climate Change, University of Cordoba; and professor, Department of Agricultural Economics, University of Cordoba. This study was conducted as part of the EU DG Research funded project AquaMoney (SSPI-022723) (www.aquamoney.org). The survey was cofunded by the CICYT project 300083-AGR2006 financed by the Spanish Ministry for Science and Technology. The work of Dr. Martin-Ortega was financed by the Andalusian Research Plan, for which she received the Centre of Andalusian Studies 2008 Ph.D. Award. We are grateful to Anna Alberini, Wolfgang Haider, Sebastiaan Hess, Vincent Linderhof, and Marije Schaafsma for useful discussions about the experimental design and estimated statistical models. The usual disclaimer applies.

  • ↵1 Ecological status is used in the WFD to indicate the quality of water bodies. Annex V of the directive classifies ecological status as (1) high if there are no or only very minor anthropogenic alterations to the chemical, hydromorphological and biological quality conditions of the water body; (2) good if the environmental and biophysical conditions show low levels of distortion due to human activities; (3) moderate if the quality conditions deviate moderately from those normally associated with an undisturbed water status; and (4) poor conditions for water bodies in a less than moderate state.

  • ↵2 A main effects design was used as this typically accounts for 70% to 90% of the observed variation (Louviere, Hensher, and Swait 2000). The efficiency of the design was tested using the conjoint choice software Sawtooth (Sawtooth Software 2008), based on the relative magnitude of the expected standard errors of the attribute levels. This produced an average efficiency for the attribute levels of 89%. No a priori information was available about the parameter values that would have allowed us to test the efficiency of alternative design procedures as, for example, reported by Ferrini and Scarpa (2007).

  • ↵3 The assumption of IIA was rejected based on the Hausman test (Hausman 1978). Excluding the first policy alternative yields a chi-square of 23.35 (p = 0.009), and excluding the second policy alternative a chi-square of 22.01 (p = 0.015). This rules out the use of a conditional (multinomial) logit model.

  • ↵4 Different distributions were tested for the random parameters. A uniform distribution with a (0,1) bound is usually suggested in case of dummy variables (Hensher, Rose, and Greene 2005). In this study, imposing a uniform distribution on the random parameters produced the same results as a normal distribution.

  • ↵5 The baseline situation was chosen in 14% of all choice occasions. Alternative A was chosen in 42% of the cases and alternative B in 44% of the cases.

  • ↵6 The probability that the goodness of fit measure of the RPL model is greater than that of the LC model is p # W(2360.51), which is virtually zero, with W() being the standard normal cumulative distribution function.

  • ↵7 As for the RPL model, different respondent characteristics were included as possible explanatory factors to find out which factors have a significant impact on class membership.

  • ↵8 The results are the same when including this variable as an interaction term with the two Doñana water quality levels.

References

  1. ↵
    1. Addelman Sidney
    . 1962. “Symmetrical and Asymmetrical Fractional Factorial Design Plans.” Technometrics 4 (1): 47–58.
    OpenUrlCrossRef
  2. ↵
    1. Ajzen Icek,
    2. Fishbein Martin
    . 1980. Understanding Attitudes and Predicting Social Behaviour. Englewood Cliffs, NJ: Prentice-Hall.
  3. ↵
    1. Bateman Ian J.,
    2. Day Brett H.,
    3. Georgiou Stavros,
    4. Lake Iain
    . 2006. “The Aggregation of Environmental Benefit Values: Welfare Measures, Distance Decay and Total WTP.” Ecological Economics 60 (2): 450–60.
    OpenUrlCrossRef
  4. ↵
    1. Ben-Akiva Moshe E.,
    2. Lerman Steven R.
    1985. Discrete Choice Analysis: Theory and Application to Travel Demand. Cambridge, MA: MIT Press.
  5. ↵
    1. Ben-Akiva Moshe E.,
    2. Swait Joffre D.
    1986. “The Akaike Likelihood Ratio Index.” Transportation Science 20 (2): 133–36.
    OpenUrlCrossRef
  6. ↵
    1. Bhat Chandra R.
    2001. “Quasi-random Maximum Simulated Likelihood Estimation of the Mixed Multinomial Logit Model.” Transportation Research Part B 35 (7): 677–93.
    OpenUrlCrossRef
  7. ↵
    1. Boxall Peter C.,
    2. Adamowicz Wiktor L.,
    3. Swait Joffre,
    4. Williams Michael,
    5. Louviere Jordan
    . 1996. “A Comparison of Stated Preference Methods for Environmental Valuation.” Ecological Economics 18 (3): 243–53.
    OpenUrlCrossRef
  8. ↵
    1. Brouwer Roy
    . 2008. “The Potential Role of Stated Preference Methods in the Water Framework Directive to Assess Disproportionate Costs.” Journal of Environmental Planning and Management 51 (5): 597–614.
    OpenUrlCrossRef
  9. ↵
    1. Brown Thomas C.,
    2. Duffield John W.
    1995. “Testing Part-Whole Valuation Effects in Contingent Valuation of Instream Flow Protection.” Water Resources Research 31 (9): 2341–51.
    OpenUrlCrossRef
  10. ↵
    1. Carson Richard T.,
    2. Flores Nicholas E.,
    3. Meade Norman F.
    2001. “Contingent Valuation: Controversies and Evidence.” Environmental and Resource Economics 19 (2): 173–210.
    OpenUrlCrossRef
  11. ↵
    1. Carson Richard T.,
    2. Mitchell Robert Cameron
    . 1993. “The Value of Clean Water: The Public’s Willingness to Pay for Boatable, Fishable and Swimmable Quality Water.” Water Resources Research 29 (7): 2445–54.
    OpenUrlCrossRef
  12. ↵
    1. Confederación Hidrográfica del Guadalquivir (CHG)
    . 2008. Ministerio de Medio Ambiente, y Medio Rural y Marino. Government of Spain. Available at www.chguadalquivir.es.
  13. ↵
    1. Ferrini Silvia,
    2. Scarpa Riccardo
    . 2007. “Designs with A Priori Information for Nonmarket Valuation with Choice Experiments: A Monte Carlo Study.” Journal of Environmental Economics and Management 53 (3): 342–63.
    OpenUrlCrossRef
  14. ↵
    1. Fishbein Martin,
    2. Ajzen Icek
    . 1975. Belief, Attitude, Intention and Behaviour: An Introduction to Theory and Research. Wokingham, UK: Addison-Wesley.
  15. ↵
    1. Fotheringham A. Stewart
    . 1988. “Consumer Store Choice and Choice Set Definition.” Marketing Science 7 (3): 299–310.
    OpenUrlCrossRef
  16. ↵
    1. Greene William H.,
    2. Hensher David A.
    2003. “A Latent Class Model for Discrete Choice Analysis: Contrasts with Mixed Logit.” Transportation Research, Part B 37 (8): 681–98.
    OpenUrlCrossRef
  17. ↵
    1. Gutiérrez Carlos,
    2. Martin-Ortega Julia,
    3. Berbel Julio
    . 2008. “Situación y Tendencias del Uso Agrícola del Agua en la Cuenca del Guadalquivir.” Revista Española de Estudios Agrosociales y Pesqueros 220 (4): 163–78.
    OpenUrl
  18. ↵
    1. Hanley Nick,
    2. Mourato Susana,
    3. Wright Robert E.
    2001. “Choice Modelling Approaches: A Superior Alternative for Environmental Valuation?” Journal of Economic Surveys 15 (3): 435–62.
    OpenUrlCrossRef
  19. ↵
    1. Hanley Nick,
    2. Wright Robert E.,
    3. Alvarez-Farizo Begoña
    . 2006. “Estimating the Economic Value of Improvements in River Ecology Using Choice Experiments: An Application to the Water Framework Directive.” Journal of Environmental Management 78 (2): 183–93.
    OpenUrlCrossRefPubMed
  20. ↵
    1. Hausman Jerry A.
    1978. “Specification Tests in Econometrics.” Econometrica 46 (6): 1251–71.
    OpenUrlCrossRef
  21. ↵
    1. Hensher David A.,
    2. Johnson Lester W.
    1981. Applied Discrete Choice Modelling. New York: Wiley.
  22. ↵
    1. Hensher David A.,
    2. Rose John M.,
    3. Greene William H.
    2005.Applied Choice Analysis. A Primer. Cambridge: Cambridge University Press.
  23. ↵
    1. Hynes Stephen,
    2. Hanley Nick,
    3. Scarpa Ricardo
    . 2008. “Effects on Welfare Measures of Alternative Means of Accounting for Preference Heterogeneity in Recreational Demand Models.” American Journal of Agricultural Economics 90 (4): 1011–27.
    OpenUrlCrossRef
  24. ↵
    1. Johnston Robert J.,
    2. Swallow Stephen K.,
    3. Bauer Dana Marie
    . 2002. “Spatial Factors and Stated Preference Values for Public Goods: Considerations for Rural Land Use.” Land Economics 78 (4): 481–500.
    OpenUrlAbstract/FREE Full Text
  25. ↵
    1. Krinsky Itzhak,
    2. Robb A. Leslie
    . 1986. “On Approximating the Statistical Properties of Elasticities.” Review of Economics and Statistics 68 (4): 715–19.
    OpenUrlCrossRef
  26. ↵
    1. Louviere Jordan J.,
    2. Hensher David A.,
    3. Swait Joffre D.
    2000. Stated Choice Methods: Analysis and Application. Cambridge: Cambridge University Press.
  27. ↵
    1. Martin-Ortega Julia
    . 2008. “Assessment of Non-Market Environmental Benefits of the Implementation of the Water Framework Directive: A Spatial Approach to the Guadalquivir River Basin [in Spanish].” Ph.D. diss. University of Córdoba.
  28. ↵
    1. Martin-Ortega Julia,
    2. Martin Carlos Gutiérrez,
    3. Vecino Julio Berbel
    . 2008. “Caracterización de los Usos del Agua en la Demarcación del Guadalquivir en Aplicación de la Directiva Marco de Aguas.” Revista de Estudios Regionales 81 (1): 45–76.
    OpenUrl
  29. ↵
    1. McFadden Daniel
    . 1974. “Conditional Logit Analysis of Qualitative Choice Behaviour.” In Frontiers in Econometrics, ed. Zarembka Paul, 105–42. New York: Academic Press.
  30. ↵
    1. McFadden Daniel,
    2. Train Kenneth
    . 2000. “Mixed MNL Models for Discrete Response.” Journal of Applied Econometrics 15 (5): 447–70.
    OpenUrlCrossRef
  31. ↵
    1. Ministerio de Medio Ambiente
    . 2007. Precios y Costes de los Servicios del Agua en España. Madrid: Informe Integrado, Ministerio de Medio Ambiente.
  32. ↵
    1. Morrison Mark,
    2. Bennett Jeff
    . 2004. “Valuing New South Wales Rivers for Use in Benefits Transfer.” Australian Journal of Agricultural and Resource Economics 48 (4): 591–611.
    OpenUrlCrossRef
  33. ↵
    1. Morrison Mark,
    2. Bennett Jeff,
    3. Blamey Russell,
    4. Louviere Jordan
    . 2002. “Choice Modelling and Tests of Benefit Transfer.” American Journal of Agricultural Economics 84 (1): 161–70.
    OpenUrlCrossRef
  34. ↵
    1. Moyano Eduardo,
    2. Garrido Fernanco,
    3. Navarro Clemente J.,
    4. Buiza Begoña
    . 2004. “La Cuestión Hídrica en Andalucía: Actitudes y Valores de los Andaluces Respecto al Problema del Agua.” Revista de Estudios Regionales 70:85–116.
    OpenUrl
  35. ↵
    1. Nylund Karen L.,
    2. Asparouhov Tihomir,
    3. Muthén Bengt O.
    2007. “Deciding on the Number of Classes in Latent Class Analysis and Growth Mixture Modeling: A Monte Carlo Simulation Study.” Structural Equation Modeling 14 (4): 535–69.
    OpenUrlCrossRef
  36. ↵
    1. Poe Gregory L.,
    2. Giraud Kelly L.,
    3. Loomis John B.
    2005. “Computational Methods for Measuring the Difference of Empirical Distributions.” American Journal of Agricultural Economics 87 (2): 353–65.
    OpenUrlCrossRef
  37. ↵
    1. Provencher Bill,
    2. Bishop Richard C.
    2004. “Does Accounting for Preference Heterogeneity Improve the Forecasting of a Random Utility Model?” Journal of Environmental Economics and Management 48 (1): 793–810.
    OpenUrlCrossRef
  38. ↵
    1. Sawtooth Software
    . 2008. Available at www.sawtoothsoftware.com.
  39. ↵
    1. Scarpa Riccardo,
    2. Thiene Mara
    . 2005. “Destination Choice Models for Rock Climbing in the Northeastern Alps: A Latent-Class Approach Based on Intensity of Preferences.” Land Economics 81 (3): 426–44.
    OpenUrlAbstract/FREE Full Text
  40. ↵
    1. Scarpa Riccardo,
    2. Willis Ken G.,
    3. Acutt Melinda
    . 2005. “Individual-Specific Welfare Measures for Public Goods: A Latent Class Approach to Residential Customers of Yorkshire Water.” In Econometrics Informing Natural Resource Management, ed. Koundouri Phoebe. Cheltenham, UK: Edward Elgar Publishers.
  41. ↵
    1. Train Kenneth
    . 2003. Discrete Choice Methods with Simulation. Cambridge: Cambridge University Press.
Back to top

In this issue

Print
Download PDF
Article Alerts
Email Article
Citation Tools
Spatial Preference Heterogeneity: A Choice Experiment
Roy Brouwer, Julia Martin-Ortega, Julio Berbel
Land Economics Aug 2010, 86 (3) 552-568; DOI: 10.3368/le.86.3.552
Twitter logo Facebook logo Mendeley logo
Bookmark this article

Jump to section