Bio-based and Applied Economics 9(3): 283-304, 2020 ISSN 2280-6180 (print) © Firenze University Press ISSN 2280-6172 (online) www.fupress.com/bae Full Research Article DOI: 10.13128/bae-7764 Does the place of residence affect land use preferences? Evidence from a choice experiment in Germany Julian Sagebiel1,*, KlauS glenK2, Jürgen Meyerhoff3 1 Swedish University of Agricultural Sciences 2 SRUC, Land Economy and Environment Research Group 3 Technische Universität Berlin Abstract. Discrete choice experiments can be used to inform policy makers on people’s preferences for landscapes and cultural ecosystem services. Recent studies have shown that the spatial context influences preferences and related willingness to pay values. In this paper we investigate the effect of the landscape surrounding people’s places of residence on their willingness to pay using data from a discrete choice experiment on local land-use changes and cultural ecosystem services throughout Germany. For anal- ysis, we apply a latent class logit model and include landscape categories as explanatory variables for class membership. We find that the different landscapes people live in are correlated with preferences. Especially people from urban areas and farm- and grass- land landscapes have larger willingness to pay values for improvements in cultural eco- system services than people from forest landscapes and cultural landscapes. The results are important for policy makers as different willingness to pay values in different land- scapes imply different welfare effects for land use changes. Taking this information into account can help in reaching more efficient resource allocations. Keywords. Landscape preferences, latent class model, spatial heterogeneity, willing- ness to pay. JEL Codes. Q51, Q57. 1. Introduction Policy makers at different scales initiate land use changes to conform with subordi- nated laws and guidelines. Decisions should balance social and private costs and benefits for different stakeholder groups and the local population. Rigorous cost-benefit analysis is often difficult to conduct, as most regulating and cultural ecosystem services that are pro- duced by landscapes are not traded in markets, making it impossible to directly observe societal demand for them. Benefit estimates of changes in ecosystem service provision need to be inferred through the use of non-market valuation techniques; in particular stated *Corresponding author. E-mail: julian.sagebiel@slu.se Editor: Meri Raggi. 284 Julian Sagebiel, Klaus Glenk, Jürgen Meyerhoff preference methods, which allow estimation of willingness to pay through direct elicitation of preferences in hypothetical markets. In Europe, several non-market valuation studies assessing preferences for components and management of agrarian landscapes have been conducted, but they rarely accounted for spatial differences in preferences (Zanten et al. 2014; Glenk et al. 2019). The few studies that considered spatial heterogeneity in preferenc- es found that the place of residence of respondents in stated preference surveys influences willingness to pay estimates (Campbell, Scarpa, and Hutchinson 2008; Campbell, Hutchin- son, and Scarpa 2009; Brouwer, Martin-Ortega, and Berbel 2010; Broch et al. 2013; Garrod et al. 2012; Johnston and Ramachandran 2014). As land-use changes are often conducted locally, such information can significantly impact the results of cost-benefit analyses and may reveal insights on where a land-use change offers the largest benefits. This paper contributes to the literature on spatial preference heterogeneity by investi- gating how preferences for policy-relevant landscape attributes differ across respondents residing in different landscapes. Spatially-driven differences in preferences for changes in landscape attributes can occur for two main reasons. First, it is well-established that individual preferences are affected by the current level of endowment (Glenk 2011; Hess, Rose, and Hensher 2008; Tversky and Kahneman 1981). Therefore, an increase or decrease in a good is valued relative to this status quo situation. Because the marginal value of a good or service may not be constant over levels of provision, individuals with different status quo situations may value additional changes in provision differently. In particular, economic theory suggests that the utility or value that is attributed to an additional unit of a good or service is higher if its scarcity increases. The concept of diminishing marginal utility suggests, for example, that people residing in a forest landscape are willing to pay less for additional forest area created than individuals living in farm- and grassland land- scapes with little forest cover (Sagebiel, Glenk, and Meyerhoff 2017). Diminishing mar- ginal utility may apply if more of a good or service is always preferred over less; however, this may not always apply to landscape attributes, where optimal shares of certain land use shares and landscape elements may exist. That is, an increase in land use share may be perceived beneficially up to a threshold, beyond which utility for an additional increase in provision decreases (Schmitz, Schmitz, and Wronka 2003). Second, the overall composition of a landscape has a unique value that is qualitatively different from other landscapes and that is difficult if not impossible to describe in terms of separate landscape attributes. That is, residents have different perceptions of landscapes and of the role that specific elements play in achieving uniqueness. Consequently, pref- erences for changes in landscape attributes may differ across landscape types, either in a systematic fashion in case that subjective perceptions of landscape amenity and value are similar across individuals living in a particular landscape type, or in an unpredictable way if there is considerable heterogeneity in perceptions. For example, those individuals living in forest landscapes may have a systematically greater demand for enhancing biodiversity, whereas people living in farm- and grassland landscapes may prefer additional structural elements. Similarly, some people living in farm- and grassland landscapes may perceive their openness as a cultural heritage characteristic of a particular region, thus objecting structural change. In the paper, we investigate the correlation between residing in different landscape categories (i.e., different status quo situations) and preferences for changes in landscape 285Does the place of residence affect land use preferences? attributes, for example share of forest or levels of biodiversity. We use data from a web- based discrete choice experiment (DCE) survey in Germany to empirically test if differ- ences in willingness to pay for landscape attributes exist; and if the ‘status quo’ landscape serves as a reference point for choices in the DCE with impacts on willingness to pay esti- mates. The results are relevant for policy makers dealing with local land-use changes and researchers considering DCEs to assist cost-benefit analyses. For example, in Germany, there is a discussion about combating climate change by increasing the share of energy crops for renewable energy generation. A policy maker can set spatially varying incentives or other policy tools aiming at increasing or decreasing the share of corn on agricultural fields. Typically, such incentives are based on private benefits and ecological constraints, e.g. where gross margins are high. Social welfare impacts associated with landscape change are often not considered at all, or are not directly compared with private costs and benefits. Additionally, the importance of acceptance of the land use change by the local population is often neglected, and willingness to pay values, distinguished by landscape categories, can help to identify areas where such a land-use change is likely to find local support. 2. Survey and Data 2.1 Data Collection and Discrete Choice Experiment The DCE is part of a German-wide, web-based survey conducted in March 2013. The respondents were recruited from an online panel of a large international market research institute. People 18 years or older who resided in Germany at the time of the study were eligible to participate. The survey consists of the six sub-samples with different DCEs, totalling around 10,000 respondents. The DCEs differ in their attributes and had differ- ent land-use foci. In all samples, the scenario was a local land use change within a radius of 15 km around the respondent’s place of residence. The radius should represent a typi- cal distance for everyday activities. We discussed the radius in focus groups and came up with 15 km being a widely accepted distance. Besides the DCE, the survey includes ques- tions on leisure activities, perceptions and knowledge on land use and climate change as well as socio-demographic variables. Respondents were requested to provide their postal code or to use the integrated geo-tool which supplies the coordinates of the places identi- fied by respondents such as their residence location. In this paper, we use a sub-sample with attributes related to agricultural land-use changes. The DCE comprises five non-monetary attributes each having three levels, with zero indicating the status quo as today. Table 1 gives a description of all attributes of the used sample as well as the dummy codes used in the analysis. The first attribute Forest refers to the share of forest. It takes the values as today, 10% less and 10% more. We assume that an increase in forest area increases utility with a decreasing rate (diminishing marginal utility). That implies that people living in forest rich areas gain less utility from an increase in forest than people living in areas with a low share of forest. The second attribute Fieldsize describes the average size of fields and forests. The levels include as today, half the size of today and double the size of today. 286 Julian Sagebiel, Klaus Glenk, Jürgen Meyerhoff Smaller field sizes imply a less monotonic landscape and more structural elements, which are assumed to be more attractive in terms of visual amenity (Zanten et al. 2014). On the other hand, larger forests can lead to better forest connectivity which may have posi- tive implications for biodiversity and recreation. We therefore have no clear expectation for this attribute. The third attribute Biodiversity is described with a bird indicator as a proxy for biodiversity. Bird indicators are used in several countries as headline indicators for biodiversity (Gregory et al. 2003; Butchart et al. 2010). The bird indicator, developed by the German Federal Agency for Nature Conservation, provides information on the suitability of the area for birds, where 100 points describe the state in the year 1975 in Germany (Doerpinghaus and Ludwig 2005). For Germany as a whole, the bird indicator is currently estimated to lie at about 55 points. The levels used in the DCE are as today (55 Points), slight increase (85 Points) and strong increase (105 Points). We expect that util- ity increases with increasing points, as it has been found in other DCE studies (Shoyama, Managi, and Yamagata 2013). The fourth attribute Cornshare is the share of corn on agri- cultural fields. The levels are as today, 30% and 70% on the agricultural fields in the sur- rounding. In the focus group discussions conducted prior to the survey, corn was often described as having a negative impact on landscape. We expect that a larger share of corn leads to a decrease in utility. Meadowsshare, the fifth attribute, refers to the share of mead- ows and grassland used for grazing. It takes the levels as today, 25% of the area, 50% of the area. In the focus group discussions, most participants linked a high share of meadows to a more natural landscape. We thus expect a utility increase from an increase in the share. Note that some attribute levels imply a reduction in the endowment compared to the sta- tus quo. This is explicit for Forest and Fieldsize and implicit for Cornshare and Mead- owsshare. In the former case, we expect that some respondents have preferences for a Table 1. Attribute description. Attribute Description Levels Dummy Code Forest Share of forest in % as today 10% decrease omitted ForMinus10 10% increase ForPlus10 Fieldsize Average size of forest and fields as today half the size double the size omitted FieldHalf FieldDouble Biodiversity Degree of biodiversity measured with bird indicator as today (55 Points) slight increase (85 Points) strong increase (105 Points) omitted Bio85 Bio105 Cornshare Share of corn on agricultural fields as today share of 30% share of 70% omitted Corn30 Corn70 Meadows share Share of meadows in % as today share of 25% share of 50% omitted Mead25 Mead50 Price Annual payment to a local landscape fund in Euro 0, 10, 25, 50, 80, 110, 160 287Does the place of residence affect land use preferences? reduction. For example, in forest rich areas, people may prefer a reduction in forest share (Sagebiel, Glenk, and Meyerhoff 2017). To account for such preferences, we used a posi- tive and a negative level. In the case of Cornshare and Meadowsshare, the direction of the change (reduction or increase) depends on the respondent’s current situation. However, absolute percentage values are useful as, in practice, land use changes are often announced in such values. We expected that people understand an absolute percentage value better than a relative change. Thus we used absolute percentage values for these attributes, taking into account that the change people value varies between respondents. Finally, the price attribute is framed as an annual payment to a newly introduced landscape fund per per- son for an unspecified period of time. We explained to the respondents that all residents who are affected by the land use change will have to contribute to the fund (i.e. a compul- sory payment) and that the money in the fund was to be exclusively used to finance and maintain the land use changes. The exact description of the payment vehicle was informed by focus group discussions. The framing of the payment vehicle as a fund was preferred to other possible payment vehicles and regarded as credible. Tax payments were not regarded as credible, because the land use change was local while taxes are usually collected at least at county level and often used for multiple purposes. The levels of the fund range from 10 to 160 Euro and is set to szero in the status quo alternative. Each choice set consists of three unlabelled landscape alternatives, where landscape 3 represents the status quo (Figure 1). The experimental design was created with the soft- ware package NGene, maximizing C-efficiency, which relates to the minimization of vari- Figure 1. Example of a choice set. 288 Julian Sagebiel, Klaus Glenk, Jürgen Meyerhoff ance of willingness to pay estimates. The design was optimized for a multinominal logit model with linearity in utility and priors close to zero. It consisted of 18 choice sets divid- ed into two blocks. Each respondent answered nine choice sets. The order of the choice sets was randomized across respondents. 2.2 Landscape categories and socio-demographics The German Federal Agency for Nature Conservation has developed a system to clas- sify landscapes within Germany. The intention behind this approach is to provide a basis for effective conservation and development of cultural landscapes along the objectives of the European Landscape Convention. Overall, the German land surface was divided in 858 landscapes including 59 urban conglomerations. The system comprises overall 24 landscape types that are assigned to the following six main categories (Gharadjedaghi et al. 2004):1 1. Coastal landscapes: This type is characterized by landscapes near the German coast of the North Sea and the Baltic Sea. 2. Forest landscapes: These landscapes have a large share of forests between 40% and 70%. 3. Cultural landscapes: These landscapes have a share of forest between 20% and 40% and a high share of one of the following items: water bodies, meadows and grassland, wine-growing, glaciers and rocks, orchards, wetlands, a combination of the items. 4. Farm- and grassland dominated landscapes: In contrast to the cultural landscapes, they have a share of forest that is less than 20%. They are further characterized by a large share of grassland and arable land. 5. Mining areas: Landscapes with more than 10 percent of the land surface under open cast mining. 6. Urban agglomerations: These landscapes comprise cities and areas with a high density of settlements and infrastructure. Table 2 summarizes the distribution of the respondents according to the landscapes. Each respondent is uniquely allocated to one of the categories. In this process, the actual place of residence was used to determine the landscape category rather than the percentage share of landscape categories surrounding the place of resi- dence. Figure 2 maps both the landscape categories and the respondents’ locations. We exclude five respondents from coastal landscapes and mining areas from the analysis as these categories are too small. The final sample size is 1409. 1 See https://www.bfn.de/en/activities/protecting-habitats-and-landscapes/landscapes-of-conservation-impor- tance/landscape-types.html for a brief description of the 24 landscape types. Table 2. Distribution of landscape categories. Landscape Category No. % Coastal landscapes 3 0.2 Forest landscapes 204 14.4 Cultural landscapes 326 23.1 Farm- and grassland landscapes 309 21.9 Mining areas 2 0.1 Urban Agglomerations 570 40.3 Total 1414 100.0 289Does the place of residence affect land use preferences? Figure 2. Spatial distribution of sample. 290 Julian Sagebiel, Klaus Glenk, Jürgen Meyerhoff 2.3 Hypotheses and empirical strategy The geo-referenced respondents are distinguished by the landscape categories described in Table 2. The main aim is to find out whether respondents from different landscapes exhibit different preferences. Hence, the main hypothesis is: preferences and willingness to pay values for landscape attributes correlate with the landscape in which a respondent lives. We expect decreasing marginal utility, i.e. marginal willingness to pay is lower in landscapes where the status quo levels of defining attributes are already high. For example, marginal willingness to pay for more forest is lower in forest landscapes than in the other landscape categories. Additionally, we expect some kind of place attachment for attributes that dominate a landscape (Scannell and Gifford 2010). For example, a respondent living in a forest rich area is not willing to give up forest as it is a dominant characteristic of the landscape. In contrast, a respondent living in an area with a medium share of forest is more interested in gaining forest but also less averse against a loss in forests. Table 4 shows that in farm- and grassland landscapes, fieldsize is higher than in the other catego- ries, where it is rather similar. Hence, the hypothesis is that the willingness to pay for half the size differs between farm- and grassland landscapes and the other landscapes. Corn share is highest in the two cultural landscapes and lowest in urban agglomerations. As a high corn share is expected to be perceived negatively, and for most respondents the first level already implies an increase over the status quo, we expect negative willingness to pay values. These would be highest in cultural landscapes and lowest in urban agglomera- tions. Therefore, we focus on the second level of this attribute, i.e. an increase to 70%. The average share of meadows is relatively similar in all landscapes, so that large differences in willingness to pay may not be present. 3. Econometric approach In the analysis, we use a latent class logit model to investigate the effects of the land- scape categories on preferences and willingness to pay. The model is consistent with microeconomic theory, assuming rational individuals who maximize a utility function under constraints. An individual i chooses in t choice situations between a given set of alternatives n – each described by a conditional indirect utility function Uint – the alterna- tive that provides the maximum amount of utility. Each alternative is characterized by k attributes that have levels Aiknt. We assume the utility functions for alternatives to be linear and additive in the attributes, and add an error term eint which is Extreme Value Type I distributed to the random utility model. A utility function can be written as Uint=beta1Ai1nt+β2Ai2nt+…+βkAiknt+eint (1) where the βks are the corresponding utility coefficients. The probability of an individual choosing alternative n can be written as a conditional logit model: (2) 291Does the place of residence affect land use preferences? This model has a closed form and can be estimated using maximum likelihood. In order to incorporate preference heterogeneity, we apply a latent class logit model. We assume that a given number of preference classes S, differing in their utility parameters <βk|1,βk|2,…,βk|s>, exists. Each individual has probabilities to be member of the preference classes. The probabilities hs can be estimated with a multinomial logit model (3) where Xi are explanatory variables, in this case the landscape categories, and ζs are the coefficients. The unconditional choice probability to choose alternative m is given as (4) The latent class logit model as described in equation 4 introduces preference heter- ogeneity between classes. Within a class, preferences are fixed. To relax this assumption without introducing a large amount of new parameters, we extend the model to a scale- adjusted latent class model (Magidson and Vermunt 2008). In this model, each preference class s is separated by a constant which can be interpreted as a scale parameter. The scale parameter merely states that preferences for all attributes are higher in the one scale class than in the other scale class. Whether the differences between respondents are caused by different preferences (all very high, vs. all very low) or by differences in the error vari- ances (more random vs. less random choices) cannot be answered empirically (Hess and Train 2017). Still, the introduction of this parameter captures another dimension of heter- ogeneity, which can improve model fit significantly. As the scale classes are restricted in a way that all preference parameters differ similarly, willingness to pay values between scale classes are not affected. Technically, the scale parameter is estimated by another multino- mial logit model, and each respondent has a probability g to belong to scale class r – simi- lar to the preference classes. The unconditional choice probability in equation 4 becomes (5) If an earlier analysis has already identified some respondents belonging to a specific class, one can add a known-class parameter τr. This parameter is zero if a respondent can- not be assigned a priori to a certain class, leading to (6) In this study, we use the known class indicator to classify all respondents who have always chosen the status quo option into class 1. To determine the number of preference 292 Julian Sagebiel, Klaus Glenk, Jürgen Meyerhoff classes S, one can use statistical measures of fit such as the Bayesian Information Crite- rion (BIC), or the corrected Akaike Information Criterion (cAIC). Both BIC and cAIC penalize for more parameters and are therefore preferred over other information criteria. Additional to the statistical criteria, one can rely on own judgment concerning reasonable parameter estimates and knowledge gained from earlier analyses (Boxall and Adamowicz 2002; Scarpa and Thiene 2005). To calculate willingness to pay values for each class individually, the respective class preference parameter is divided by the class cost parameter. Confidence intervals of will- ingness to pay are calculated with the delta method. 4. Results 4.1 Descriptive statistics of landscape categories We first analyze the relationship between socio-demographic variables and landscape categories. This step is important to understand whether and how potential differences in preferences could arise from differences in socio-demographics rather than the landscape respondents are living in. Most differences are found between urban agglomerations and the other landscapes (Table 3). Respondents from urban agglomerations are more educated and have fewer children. We use Kruskall-Wallis and t-tests to test for overall differences between the landscape categories. Statistically significant differences on a 5% level are present for all variables except personal income and sex. Although there are differences in socio-demo- graphics between landscape categories (especially between urban areas and all other are- as), we will not investigate those here. We acknowledge that the differences in preferences may be driven by socio-demographics rather than landscape categories, but this is not relevant for the policy question of how land use changes are perceived in different land- scapes. Our analysis thus only provides correlations. Using data from the German Federal Agency for Cartography and Geodesy (BKG) and the German Federal Institute of Research on Building, Urban Affairs and Spatial Development (BBSR), we investigated the actual status quo attribute levels of the respond- ents. Table 4 summarizes the actual status quo in the 15km radius by landscape categories. In most cases, there are relatively large differences between the landscape categories. For the sake of parsimony, we will not investigate the actual status quo and possible effects any further. Sagebiel, Glenk, and Meyerhoff (2017) conduct a detailed investigation of the actual status quo and its effects on willingness to pay. 4.2 Latent class analysis We estimate the latent class model described in section 3 using the software package LatentGold Choice 4.5 with the Syntax module. To select a specific number of classes we compared BIC and cAIC for two to eight class models, in the absence and presence of a scale class. We choose a model with five preference classes and two scale classes. This model turned out have the lowest BIC and cAIC values and offered plausible parameter values. 293Does the place of residence affect land use preferences? All attributes except price were dummy coded with the status quo level as today as the reference. The landscape categories entered the class membership function as dum- my coded variables with forest landscapes as the reference category. We did not include any socio-demographic variables as these are correlated with the landscape categories, Table 3. Frequencies and column percentages (in parentheses) of socio-demographic variables. Forest Cultural Farm- and grassland Urban Total Education Secondary or less 83 122 121 141 467 (40.9) (37.4) (39.4) (24.7) (33.2) Higher education 46 86 77 155 364 (22.7) (26.4) (25.1) (27.2) (25.9) University 74 118 109 274 575 (36.5) (36.2) (35.5) (48.1) (40.9) Sex Male 100 168 175 308 751 (49.0) (51.5) (56.6) (54.0) (53.3) Female 104 158 134 262 658 (51.0) (48.5) (43.4) (46.0) (46.7) Children in household Yes 68 135 102 134 439 (33.3) (41.4) (33.0) (23.5) (31.2) No 136 191 207 436 970 (66.7) (58.6) (67.0) (76.5) (68.8) Income Less than 1500 Euros 81 128 120 227 556 (39.7) (39.3) (38.8) (39.8) (39.5) 1500 to 2600 Euros 58 91 74 153 376 (28.4) (27.9) (23.9) (26.8) (26.7) More than 2600 Euros 65 107 115 190 477 (31.9) (32.8) (37.2) (33.3) (33.9) Age 19 to 29 32 67 60 132 291 (15.7) (20.6) (19.4) (23.2) (20.7) 30 to 39 50 62 58 110 280 (24.5) (19.0) (18.8) (19.3) (19.9) 40 to 49 39 89 92 144 364 (19.1) (27.3) (29.8) (25.3) (25.8) 50 to 59 40 64 55 102 261 (19.6) (19.6) (17.8) (17.9) (18.5) Older than 60 43 44 44 82 213 (21.1) (13.5) (14.2) (14.4) (15.1) 294 Julian Sagebiel, Klaus Glenk, Jürgen Meyerhoff potentially causing multicollinearity. 23% of all respondents chose the status quo alterna- tive in all choice situations and were assigned to class 1 with a probability of 1. As several respondents seemed to have ignored the price attribute, we fixed the price parameter to zero in class 3 to capture price non-attendance. In models without this restriction, at least one class is characterized by willingness to pay values three times as high as the highest price level of 160 Euro, which we consider implausible. In a first step, we describe the five classes in terms of estimated utility parameters and willingness to pay values. Then, we investigate the relationship between class membership and landscape categories. Table 5 shows the estimation results and Table 6 its willingness to pay values. The overall model is highly significant. The statistically significant coefficient for the scale class of -0.302 translates to scale class probabilities of 57.5% and 42.5% for scale classes 1 and 2, respectively, indicating that additional heterogeneity and correlation pat- terns are present. In Class 1, price, ForMinus10, FieldHalf, FieldDouble, Corn70 and Mead50 are highly significant and negative. Willingness to pay values range between -88 and -35 Euro, i.e. people are opting against all land use changes and would need to be compensated. The positive and significant ASCsq means that Class 1 is characterized by preferences towards the status quo. Class 2 has a negative and significant ASCsq, indi- cating preferences for land use changes. ForMinus10, ForPlus10, FieldDouble, Bio105, Corn70, Mead50 and price are significant with the expected signs. The willingness to pay for ForMinus10 and ForPlus10 is -165 Euro and 64 Euro, respectively. People are will- ing to pay for increases in forest, but would need to be compensated nearly three times as much for decreases in forest. For a reduction of field sizes (FieldHalf), willingness to pay is nearly 20 Euro while a doubling of field sizes would need to be compensated with 45 Euro. Willingness to pay for increases in biodiversity is 32 Euro for an increase to 85 points and 51 Euro for 105 points. A share of corn of 30% is not significant but a share of 70% requires a compensation of 61 Euro. Willingness to pay for a share of meadows of 25% is positive (42 Euro) while a share of 50% is not significant and close to zero. In sum- mary, Class 2 is characterized by large positive and negative willingness to pay values for land use changes. Class 3 is the price non-attendance class. Respondents who disregard the cost attribute are likely choosing a land use change scenario over the status quo if they Table 4. Mean and standard deviation (in parenthesis) of actual status quo by landscape categories. Forest Cultural Farm- and grassland Urban Total Forest Share 41.7 29.8 17.5 18.4 24.2 (12.2) (11.2) (9.7) (10.0) (13.7) Field Size 17.7 17.5 25.9 17.0 19.2 (7.5) (6.8) (12.2) (6.7) (9.1) Corn Share 14.9 20.9 19.9 10.5 15.5 (10.3) (14.8) (15.8) (10.0) (13.5) Meadows Share 15.6 17.9 18.1 12.6 15.5 (6.1) (9.0) (12.1) (7.1) (9.1) 295Does the place of residence affect land use preferences? have a positive attitude towards policy change. This is reflected in the very large and nega- tive ASCsq. Similarly, the very large and negative coefficient for decreases in forest share can be explained by this phenomenon. Nearly all coefficients of the remaining attributes are significant and have the expected signs. Bio85 is significant and negative which could imply that members of this class have already a high degree of biodiversity and regard 85 points as a deterioration. Similarly, the positive coefficient of Corn30 implies that peo- ple have already high shares of corn and regard a 30% share as an improvement. Finally, Mead25 is not significant while Mead50 is significant and positive. Class 4 is character- ized by comparatively large negative willingness to pay values to avoid decreases in forest share, field size and a corn share of 70%. Interestingly, the willingness to pay for meadows share is negative for both 25% and 50%. In Class 5, positive willingness to pay values are significant and positive only for ForPlus10 (14 Euro) and Bio105 (12 Euro) and negative for FieldHalf (-16 Euro) and Mead25 (-19 Euro). This class comprises small or no utility gains from land use changes. The landscape categories have a significant impact on the probability to be member of a class. Forest landscapes and Class 1 are the reference categories, the parameters in Table 5. Latent class model with five classes. Class 1 Class 2 Class 3 Class 4 Class 5 ASCsq 0.760 -1.281*** -23.897** -3.993*** -3.535*** ForMinus10 -3.151*** -4.169*** -17.829* -1.287*** -0.062 ForPlus10 -0.316 1.624*** 0.926*** 0.317* 1.108** FieldHalf -2.678*** 0.474 -0.075 -0.961*** -1.305*** FieldDouble -2.120*** -1.132*** -0.275** 0.390** -0.361 Bio85 0.774 0.814** -5.791** 0.444 -0.455 Bio105 0.431 1.295*** 1.233*** 0.103 0.964* Corn30 -0.647 0.592 1.055*** 0.038 0.351 Corn70 -5.326*** -1.563*** 0.214 -1.195*** -0.763 Mead25 -1.173 1.056*** -0.035 -1.120*** -1.503*** Mead50 -2.525*** 0.090 0.773*** -0.802*** -0.586 price -0.060*** -0.025*** 0.000 -0.013*** -0.078*** Covariates of membership function Forest ref ref ref ref ref Cultural ref 0.172 0.877** 0.041 0.924* Grass/Farm ref 0.655** 1.245*** 0.2462 1.167** Urban ref 0.198 1.326*** 0.438 1.325*** Scale classes Scale Class 1 Scale Class 2 Constant ref -0.302** Log-Likelihood Observations Respondents -8976.153 12681 1409 * p < 0.10, ** p < 0.05, *** p < 0.01 , ref = reference category with parameter fixed to zero 296 Julian Sagebiel, Klaus Glenk, Jürgen Meyerhoff the membership function are interpreted relative to them. Class 1 is the largest class with a share of about 40%. Classes 2 to 4 have a share between 16% and 19%. Class 5 is the smallest class with a share of 9%. Note that Classes 1 and 5 are characterized by no or low willingness to pay values and make up nearly 50% of class membership. Table 7 shows class membership probabilities calculated for each landscape category separately. Differences in class membership between landscape categories are present in Classes 1, 3 and 5. Membership probabilities are rather homogeneous for Classes 2 and 4. Respondents from forest landscapes are more likely to be member of Class 1 compared to the other categories with a share of nearly 51% (against the class average of 39%) and less likely member of Classes 3 and 5 with shares of only 8% and 4% (compared to the class averages of 18% and 9%). Respondents from cultural landscapes are slightly more likely to be member of class 1 (42%) and less or equally likely in the other classes. Respondents from grass- and farmlands are less likely to be member of Class 1 (34% against 39%) and Class 4 (14% against 16%) and more likely to be member of Class 2 (24% against 19%). Finally, respondents from urban agglomerations are less likely to be member of Class 1 (35% against 39%) and 2 (16% against 19%) and more likely in Classes 3 (21% against 18%), 4 (17% against 16%) and 5 (11% against 9%). The results are partly in line with our expectations. Forest landscapes and cultural landscapes have high shares of forest and are relatively bio-diverse, with many structural landscape elements. Such landscapes are generally associated with high recreational values. Respondents from these categories are more likely to be member of Class 1 which is char- acterized by status quo choices and strong opposition against reductions of forest share, increases in corn and changes in field size. This aligns with our expectation of place attach- ment. The zero willingness to pay for increases of forest indicates diminishing marginal utility. People from forest landscapes are also less likely to be members of Class 3, which is characterized by a strong tendency towards land use changes and cost non-attendance, and of Class 5, which is characterized by low willingness to pay, implying some heterogeneity within this landscape category. About 55% are allocated to Classes 1 and 5 (low willingness to pay), while the remaining share belongs to the other classes which are characterized by high willingness to pay and strong preferences for land-use changes. Farm- and grasslands are dominated by agriculture and monotonic landscapes with low shares of forest. Respondents from farm- and grasslands are more likely to be mem- bers of Class 2, which is characterized by rather large willingness to pay values. This result fits to our expectations of marginal diminishing utility. People living in this landscape have a low endowment of forest, biodiversity and meadows and are thus more willing to pay for an additional unit. Finally, respondents from urban agglomerations are more likely to be member of Class 3, i.e. are more likely to not attend to costs. While we have no expectation here, this result may be explained by hypothetical bias. The choice scenario is less realistic for people in urban areas and they are less used to the landscapes. They may have ignored the price attribute more often, while at the same time exhibit strong preferences for land use changes. It should be noted that our results indicate preference heterogeneity within landscape categories. We do observe deterministic patterns of distinct preferences between landscape categories. Each landscape category is present in each class with a probability close to the average group probability. Class 1 is the largest class for all landscape catego- 297Does the place of residence affect land use preferences? ries and Class 5 is the smallest class for all landscape categories. The effects that we identi- fied should be interpreted as tendencies. Additional to the latent class analysis, we have estimated separate conditional logit models by landscape categories and used Poe et al. tests (Poe, Giraud, and Loomis 2005) to test for differences in willingness to pay between landscape categories. While exact quantitative results differ, the key findings are similar irrespective of the approach used. The appendix provides more details on the conditional logit models, willingness to pay values and the Poe et al. test results. 5. Conclusion and policy implications This paper investigated preferences for land-use changes and compared willingness to pay values between different landscape categories in Germany. The data came from a dis- crete choice experiment inferring preferences for forest share, average size of forest and fields, degree of biodiversity, share of corn and share of meadows within the 15 kilometer radius of the respondents’ places of residence. The radius was chosen to represent a typical distance for everyday activities. As the places of residence were geo-referenced, we could combine the data with landscape categories compiled by the German Federal Agency for Nature Conservation. The categories comprised forest landscapes, cultural landscapes, Table 6. Willingness to pay values. Attribute Class 1 Class 2 Class 3 Class 4 Class 5 ForMinus10 -52.52*** -165.19*** - -100.67*** -0.7954 ForPlus10 -5.27 64.33*** - 24.75* 14.24*** FieldHalf -44.64*** 18.77* - -75.17*** -16.76*** FieldDouble -35.34** -44.83** - 30.51** -4.64 Bio85 12.90 32.24** - 34.75 -5.84 Bio105 7.17 51.32*** - 8.04 12.38** Corn30 -10.78 23.44 - 2.96 4.50 Corn70 -88.78*** -61.93* - -93.47** -9.80 Mead25 -19.55 41.82** - -87.58*** -19.31** Mead50 -42.09** 3.57 - -62.74** -7.53 * p < 0.10, ** p < 0.05, *** p < 0.01. Table 7. Class probabilities by landscape categories. Landscape Class 1 Class 2 Class 3 Class 4 Class 5 Forest 0.51 0.19 0.08 0.16 0.04 Cultural 0.42 0.19 0.17 0.14 0.09 Grass/Farm 0.34 0.24 0.19 0.14 0.09 Urban 0.35 0.16 0.21 0.17 0.11 Overall 0.39 0.19 0.18 0.16 0.09 298 Julian Sagebiel, Klaus Glenk, Jürgen Meyerhoff farm- and grassland landscapes and urban agglomerations. The aim of the study was to test whether preferences for land-use changes are correlated with these landscape catego- ries. To do so, we estimated a five-class latent class model and used the landscape catego- ries as explanatory variables in the class membership function. The classes can be distin- guished by different willingness to pay values. It turned out that people from forest land- scapes and cultural landscapes were less willing to pay for land-use changes and showed a preference towards the status quo situation. Further, people from urban agglomerations and farm- and grassland have high probabilities to be member of classes with large will- ingness to pay values. In summary, the results showed that the preferences do differ among landscape cat- egories, but not as systematically as we had expected. Although we find systematic differ- ences in preferences between landscape categories, all landscape categories are relatively evenly distributed across classes. As the latent class analysis has shown, preference hetero- geneity exists also within the landscape categories. That is, each respondent, independent of which landscape category the respondent is from, has a probability of at least 8% to be member of any class. The analysis has implications for policy makers. Our study provides evidence that there are differences in preferences determined by the place of residence. Integrating such differences in landscape planning and cost-benefit analyses may help to improve decisions and induce land-use changes to areas where people appreciate them most or are least reluctant towards a change. A relevant example is the share of corn among agricultural fields. While an increase in the production of energy corn can potentially help to reduce carbon dioxide emissions, it is largely regarded as a disfigurement of the landscape. Our study revealed that opposition to corn is generally large, but stronger in forest and cul- tural landscapes than in other landscapes. Similarly, increases in forest share should take place in areas with limited forests and near urban agglomerations. Areas characterized by high recreational values such as forest and cultural landscapes should be preserved. Here, people tend more towards the status quo and changes are less appreciated by residents. There is limited interest in increases in forest shares or biodiversity, and at the same time a large resistance against reductions. In contrast, respondents from urban agglomerations and farm- and grasslands are more likely to benefit from increases in forest shares and biodiversity. Here, significant welfare effects of such measures are more likely. Our find- ings may also be used to inform the design of agri-environmental schemes. For example, compensation may be higher for measures to increase agro-biodiversity in a rather mono- tonic landscape or near urban areas, because benefits of measures are greater. Similar studies have investigated land use changes on a broader scale. In their meta- analysis, van Zanten et al. (2014) have found preferences for various landscape elements such as smaller field sizes, but no spatial determinants of preferences. Garrod et al. (2012) have found that preferences for improving ecosystem services depend on the landscape where they are present. This result is in line with our findings, yet our study differs as the proposed land use change always took place at the person’s place of residence. In Gar- rod et al. (2012), this was not the case. To our knowledge, our study is the first study that identifies spatial differentiated preferences for local land use changes. This study is limited by the fact that we did not investigate the underlying sources for the differences. The landscape categories differ in the status quo of the investigated attrib- 299Does the place of residence affect land use preferences? utes and in socio-demographic variables. Thus, we are not able to identify the causal effect of living in a certain landscape on preferences. Yet, the study insights provide correlation patterns which are sufficient to foster an understanding of the variation of preferences and willingness to pay between qualitatively different regions. Acknowledgments We thank Henry Wuestemann for support in the processing of the GIS data and Dr. Roland Goetzke and Raphael Knevels from the Federal Institute of Research on Building, Urban Affairs and Spatial Development (BBSR) for preparing the landscape structure data. Financial support from the German Ministry of Education and Research within the pro- ject CC-LandStraD (Grant Number: 01LL0909C) is gratefully acknowledged. References Boxall, Peter C, and Wiktor L Adamowicz. 2002. “Understanding Heterogeneous Prefer- ences in Random Utility Models: A Latent Class Approach.” Environmental and Resource Economics 23 (4): 421–46. Broch, Stine Wamberg, Niels Strange, Jette B. Jacobsen, and Kerrie A. Wilson. 2013. “Farmers’ Willingness to Provide Ecosystem Services and Effects of Their Spatial Distribution.” Ecological Economics 92: 78–86. Brouwer, Roy, Julia Martin-Ortega, and Julio Berbel. 2010. “Spatial Preference Heteroge- neity: A Choice Experiment.” Land Economics 86 (3): 552–68. Butchart, Stuart HM, Matt Walpole, Ben Collen, Arco van Strien, Jörn PW Scharlemann, Rosamunde EA Almond, Jonathan EM Baillie, et al. 2010. “Global Biodiversity: Indicators of Recent Declines.” Science 328 (5982): 1164–8. Campbell, Danny, W George Hutchinson, and Riccardo Scarpa. 2009. “Using Choice Experiments to Explore the Spatial Distribution of Willingness to Pay for Rural Landscape Improvements.” Environment and Planning A 41: 97–111. Campbell, Danny, Riccardo Scarpa, and W George Hutchinson. 2008. “Assessing the Spa- tial Dependence of Welfare Estimates Obtained from Discrete Choice Experiments.” Letters in Spatial and Resource Sciences 1 (2-3): 117–26. Doerpinghaus, Annette, and Gerhard Ludwig. 2005. Methoden Zur Erfassung von Arten Der Anhänge Iv Und V Der Fauna-Flora-Habitat-Richtlinie. Bundesamt für Natur- schutz. Garrod, Guy, Eric Ruto, Ken Willis, and Neil Powe. 2012. “Heterogeneity of Preferences for the Benefits of Environmental Stewardship: A Latent-Class Approach.” Ecological Economics 76: 104–11. Gharadjedaghi, B, R Heimann, K Lenz, Ch Martin, V Pieper, A Schulz, A Vahabzadeh, P Finck, and U Riecken. 2004. “Verbreitung Und Gefährdung Schutzwürdiger Land- schaften in Deutschland.” Natur Und Landschaft 79 (2): 71–81. Glenk, Klaus. 2011. “Using Local Knowledge to Model Asymmetric Preference Formation in Willingness to Pay for Environmental Services.” Journal of Environmental Man- agement 92 (3): 531–41. 300 Julian Sagebiel, Klaus Glenk, Jürgen Meyerhoff Glenk, Klaus, Robert J Johnston, Jürgen Meyerhoff, and Julian Sagebiel. 2019. “Spatial Dimensions of Stated Preference Valuation in Environmental and Resource Eco- nomics: Methods, Trends and Challenges.” Environmental and Resource Economics, 1–28. Gregory, RD, D Noble, R Field, J Marchant, M Raven, and DW Gibbons. 2003. “Using Birds as Indicators of Biodiversity.” Ornis Hungarica 12 (13): 11–24. Hess, Stephane, John M Rose, and David A Hensher. 2008. “Asymmetric Preference For- mation in Willingness to Pay Estimates in Discrete Choice Models.” Transportation Research Part E: Logistics and Transportation Review 44 (5): 847–63. Hess, Stephane, and Kenneth Train. 2017. “Correlation and Scale in Mixed Logit Models.” Journal of Choice Modelling 23: 1–8. Johnston, Robert J, and Mahesh Ramachandran. 2014. “Modeling Spatial Patchiness and Hot Spots in Stated Preference Willingness to Pay.” Environmental and Resource Eco- nomics 59 (3): 363–87. Magidson, Jeroen K., and Jeroen K. Vermunt. 2008. “Removing the Scale Factor Con- found in Multinomial Logit Choice Models to Obtain Better Estimates of Prefer- ence.” In Sawtooth Software Conference Proceedings, 139–55. Sawtooth Software. Poe, Gregory L, Kelly L Giraud, and John B Loomis. 2005. “Computational Methods for Measuring the Difference of Empirical Distributions.” American Journal of Agricul- tural Economics 87 (2): 353–65. Sagebiel, Julian, Klaus Glenk, and Jürgen Meyerhoff. 2017. “Spatially Explicit Demand for Afforestation.” Forest Policy and Economics 78: 190–99. Scannell, Leila, and Robert Gifford. 2010. “Defining Place Attachment: A Tripartite Organizing Framework.” Journal of Environmental Psychology 30 (1): 1–10. Scarpa, Riccardo, and Mara Thiene. 2005. “Destination Choice Models for Rock Climb- ing in the Northeastern Alps: A Latent-Class Approach Based on Intensity of Prefer- ences.” Land Economics 81 (3): 426–44. Schmitz, K, PM Schmitz, and TC Wronka. 2003. “Valuation of Landscape Functions Using Choice Experiments.” German Journal of Agricultural Economics 52 (8): 379–89. Shoyama, Kikuko, Shunsuke Managi, and Yoshiki Yamagata. 2013. “Public Preferences for Biodiversity Conservation and Climate-Change Mitigation: A Choice Experiment Using Ecosystem Services Indicators.” Land Use Policy 34: 282–93. Tversky, Amos, and Daniel Kahneman. 1981. “The Framing of Decisions and the Psychol- ogy of Choice.” Science 211 (4481): 453–58. van Zanten, Boris T, Peter H Verburg, Mark J Koetse, and Pieter JH van Beukering. 2014. “Preferences for European Agrarian Landscapes: A Meta-Analysis of Case Studies.” Landscape and Urban Planning 132: 89–101. 301Does the place of residence affect land use preferences? Appendix In order to further investigate differences between landscape categories, we estimate separate conditional logit models for the landscape categories. Table 8 provides the esti- mation results. Table 8. Conditional logit models by landscape. (1) Forest (2) Cultural (3) Farm- and Grasslands (4) Urban ASCsq 0.101 0.0171 -0.0959 0.0436 (0.201) (0.155) (0.154) (0.112) ForMinus10 -0.647*** -0.594*** -0.532*** -0.458*** (0.135) (0.103) (0.106) (0.0778) ForPlus10 0.139 0.284*** 0.384*** 0.401*** (0.104) (0.0789) (0.0769) (0.0562) FieldHalf -0.152 -0.314*** -0.283*** -0.221*** (0.119) (0.0905) (0.0906) (0.0670) FieldDouble -0.302*** -0.345*** -0.267*** -0.0984* (0.107) (0.0791) (0.0773) (0.0561) Bio85 0.0381 0.129 0.0650 0.204*** (0.127) (0.0994) (0.102) (0.0753) Bio105 0.0126 0.246*** 0.327*** 0.417*** (0.118) (0.0884) (0.0843) (0.0614) Corn30 -0.0983 0.176* 0.000266 -0.0231 (0.125) (0.0972) (0.0965) (0.0706) Corn70 -0.691*** -0.390*** -0.550*** -0.548*** (0.127) (0.0930) (0.0919) (0.0670) Mead25 0.0596 0.0768 -0.131 0.0247 (0.136) (0.103) (0.105) (0.0761) Mead50 -0.234* -0.151 -0.182* -0.125* (0.130) (0.0939) (0.0938) (0.0683) price -0.00615*** -0.00755*** -0.00520*** -0.00583*** (0.00115) (0.000892) (0.000858) (0.000630) N 5508 8802 8343 15390 pseudo R2 0.171 0.117 0.087 0.081 AIC 3366.8 5717.0 5604.7 10379.3 BIC 3446.1 5802.0 5689.0 10471.0 χ2 691.3 753.6 529.8 916.5 Log-Likelihood (NULL) -2017.1 -3223.3 -3055.2 -5635.9 Log-Likelihood -1671.4 -2846.5 -2790.3 -5177.6 Standard errors in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01. 302 Julian Sagebiel, Klaus Glenk, Jürgen Meyerhoff All models are highly significant and differences between the landscape categories are visible. In forest landscapes, ForPlus10 is not significant, according to the hypothesis that respondents living in areas with a lot of forests have a limited preference for an increase in the share of forests. An increase in biodiversity to 85 points is only significant in urban agglomerations, where people are characterized by a low degree of biodiversity. Hence, an increase to 85 points has already a positive effect on utility. In the other categories, biodi- versity is significant only at the 105 point level. In order to better understand the differ- ences, Table 9 displays the estimated willingness to pay values for the different categories and Figure 3 gives a graphical overview of the willingness to pay values and correspond- ing 95% confidence intervals. Finally, Table 10 provides the p-values of the Poe test. If the p-value is larger than 0.95 or smaller than 0.05, the willingness to pay values are signifi- cantly different. The Poe test has to be interpreted with care. Significant differences will only appear when confidence intervals are small enough. Hence, if the test does not reject the hypotheses that the willingness to pay values are similar, it does not necessarily mean that they are not. It rather means that we cannot show that they are. Differences in willingness to pay are significant for ForPlus10, Bio105, Corn30 and Corn70. ForPlus10 is not significant for forest landscapes and is significantly higher in open cultural landscapes and urban agglomerations. An increase in biodiversity is valued most in open cultural landscapes and in urban agglomerations and is significantly higher than in forest landscapes. An increase in corn share to 70% has the highest negative will- ingness to pay in forest landscapes and in open cultural landscapes. There are very few differences between open cultural landscapes and urban agglomerations and no significant differences for Meadows Share and Bio85, which however maybe due to the large con- fidence intervals. FieldHalf and FieldDouble are nearly always significant, but again, no significant willingness to pay differences exist. Thus, preferences for this attribute are rela- tively similar. The results from the Poe test are corresponding to the findings from the latent class analysis. In both exercises, people from open cultural landscapes and urban agglomera- tions seem to have relatively equal preferences. Similarly, people from forest landscapes and from structurally rich cultural landscapes exhibit similar preferences. The main hypotheses of decreasing marginal utility seem partly confirmed. For example, people in forest landscapes have no willingness to pay for an increase, but a strong willingness to pay against a decrease. However, not in all cases, the results correspond to our expecta- tions. 303Does the place of residence affect land use preferences? Table 9. Willingness to pay for different landscape models. (1) Forest (2) Cultural (3) Farm- and grassland (4) Urban ForMinus10 -105.2*** -78.72*** -102.3*** -78.52*** (29.01) (16.00) (26.24) (15.52) ForPlus10 22.54 37.66*** 73.74*** 68.74*** (16.21) (10.10) (16.33) (10.44) FieldHalf -24.75 -41.61*** -54.31*** -37.89*** (20.69) (13.79) (20.61) (12.65) FieldDouble -49.10** -45.62*** -51.34*** -16.88 (21.77) (13.10) (19.00) (10.26) Bio85 6.188 17.10 12.49 34.97*** (20.38) (12.77) (19.15) (12.37) Bio105 2.044 32.57*** 62.91*** 71.57*** (18.98) (10.26) (14.17) (9.419) Corn30 -15.98 23.26* 0.0511 -3.970 (21.72) (11.95) (18.54) (12.29) Corn70 -112.4*** -51.64*** -105.6*** -93.96*** (35.27) (16.03) (29.91) (18.47) Mead25 9.688 10.17 -25.24 4.244 (21.66) (13.45) (21.35) (12.97) Standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01. Table 10. Poe test results. ForMinus10 ForPlus10 FieldHalf FieldDouble Bio85 Bio105 Corn30 Corn70 Mead25Mead50 Forest vs. Cultural 0.845 0.905 0.251 0.481 0.879 0.987 0.992 0.965 0.696 0.558 Forest vs. Farm 0.412 0.998 0.174 0.529 0.644 0.993 0.805 0.589 0.327 0.517 Forest vs. Urban 0.795 0.997 0.318 0.711 0.877 1.000 0.837 0.683 0.675 0.492 Cultural vs. Farm 0.074 0.986 0.33 0.55 0.185 0.775 0.056 0.046 0.14 0.449 Cultural vs. Urban 0.4 0.975 0.597 0.779 0.486 0.908 0.025 0.041 0.474 0.403 Farm vs. Urban 0.889 0.321 0.733 0.683 0.802 0.623 0.505 0.587 0.867 0.458 304 Julian Sagebiel, Klaus Glenk, Jürgen Meyerhoff Figure 3. Willingness to pay confidence intervals by sample. −1 50 −1 00 −5 0 W illi ng ne ss to P ay Wooden landscapes Structurally rich landscapes Open cultural landscapes Urban agglomerations ForMinus10 0 20 40 60 80 10 0 W illi ng ne ss to P ay Wooden landscapes Structurally rich landscapes Open cultural landscapes Urban agglomerations ForPlus10 −1 00 −5 0 0 50 W illi ng ne ss to P ay Wooden landscapes Structurally rich landscapes Open cultural landscapes Urban agglomerations FieldHalf −1 00 −8 0 −6 0 −4 0 −2 0 0 W illi ng ne ss to P ay Wooden landscapes Structurally rich landscapes Open cultural landscapes Urban agglomerations FieldDouble −4 0 −2 0 0 20 40 60 W illi ng ne ss to P ay Wooden landscapes Structurally rich landscapes Open cultural landscapes Urban agglomerations Bio85 −5 0 0 50 10 0 W illi ng ne ss to P ay Wooden landscapes Structurally rich landscapes Open cultural landscapes Urban agglomerations Bio105 −6 0 −4 0 −2 0 0 20 40 W illi ng ne ss to P ay Wooden landscapes Structurally rich landscapes Open cultural landscapes Urban agglomerations Corn30 −2 00 −1 50 −1 00 −5 0 0 W illi ng ne ss to P ay Wooden landscapes Structurally rich landscapes Open cultural landscapes Urban agglomerations Corn70 −1 00 −5 0 0 50 W illi ng ne ss to P ay Wooden landscapes Structurally rich landscapes Open cultural landscapes Urban agglomerations Mead25 −8 0 −6 0 −4 0 −2 0 0 20 W illi ng ne ss to P ay Wooden landscapes Structurally rich landscapes Open cultural landscapes Urban agglomerations Mead50 Figure 3: willingness to pay CI by Sample 20 Investigating determinants of choice and predicting market shares of renewable-based heating systems under alternative policy scenarios Cristiano Franceschinis, Mara Thiene Multi-country stated preferences choice analysis for fresh tomatoes Maria De Salvo1,*, Riccardo Scarpa2,3,4, Roberta Capitello2, Diego Begalli2 “Not my cup of coffee”. Farmers’ preferences for coffee variety traits. Lessons for crop breeding in the age of climate change Abrha Megos Meressa, Ståle Navrud* Does the place of residence affect land use preferences? Evidence from a choice experiment in Germany Julian Sagebiel1,*, Klaus Glenk2, Jürgen Meyerhoff3 The use of latent variable models in policy: A road fraught with peril? Danny Campbell*, Erlend Dancke Sandorf