Bio-based and Applied Economics 9(2): 171-200, 2020 ISSN 2280-6180 (print) © Firenze University Press ISSN 2280-6172 (online) www.fupress.com/bae Full Research Article DOI: 10.13128/bae-8287 Assessing preferences for rural landscapes: An attribute based choice modelling approach Cathal O’DOnOghue1, Stephen hyneS1, paul Kilgarriff2, Mary ryan3,*, anDreaS tSaKiriDiS3 1 National University of Ireland, Galway 2 Luxembourg Institute for Socio Economic Research 3 Teagasc, the Irish Agriculture and Food Development Authority Abstract. This study adopts a choice modelling framework to disentangle individu- al preferences for rural landscape attributes based on the viewing of photographs of the Irish countryside. Using ordered logit and standard panel and pooled regression models, societal preferences are quantified for rural landscape attributes, grouped into natural, agricultural and human-built non-agricultural categories. The prefer- ences of 430 individuals towards 50 rural landscape photographs are analysed. The results show positive preferences for landscapes with natural attributes such as cliffs, mountainous features, water and native trees, as well as preferences for neat/managed agricultural landscapes and traditional human-built features such as stone walls and planted hedgerows. The study shows negative preferences for features such as flood- ing, unmanaged landscapes, industrial turf cutting and mechanised features such as wind turbines. There is significant preference heterogeneity observed across the sam- ple particularity across the urban-rural residency divide. It is argued that analysing preferences for specific attributes of landscapes rather than preferences for individual landscape photographs allows for further applications particularly in the area of simu- lation. Keywords. Rural landscapes, choice modelling, ordered logit, attribute preference heterogeneity. JEL codes. Q18, Q24, Q57. 1. Introduction Agriculture is a multifunctional, natural resource based sector that takes place pre- dominately in rural areas. It provides private goods like the ‘5 fs’: food, feed, fuel, fibre and forest (Kern, 2002), generating income for farm families and contributing to the aes- thetic character of human-ecological systems. These landscapes also support the delivery of other public goods such as recreation and cultural heritage (Kantelhardt et al., 2015) *Corresponding author. E-mail: mary.ryan@teagasc.ie Editor: Francesco Vanni. 172 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis and ecosystem services (ES) relating to greenhouse gas emissions, water quality and bio- diversity (Vanni, 2014; van Zanten et al., 2014; OECD, 2015; Kantelhardt, 2006). These benefits, supplied by a sustainable agricultural sector, are reflected at EU policy level with increasing levels of funding dedicated to protecting rural landscapes and providing addi- tional public goods from farming. As landscape values are often perceived as public goods, in the sense that they are non-excludable and non-rival in consumption, markets cannot place a price on landscape features and quality of landscape services (Hanley et al., 2009), nor can they guarantee their adequate provision (Schaller et al., 2018; Villanueva et al., 2015; Rodríguez-Entrena et al., 2017). Thus, where there is a market failure, there is a case that governments should implement measures to ensure an adequate provision. To do that however knowledge is required in terms of the preferences of society for alternative landscape types and features. A wide range of studies, using different methodologies, have attempted to examine rural landscape preferences and values in order to guide policy and better target expendi- ture to the most ‘valued’ landscapes. There are a number of studies that use expert judge- ment to assess the aesthetic quality of landscapes (Frank et al., 2013; Hermes et al., 2018). However, the perception of value may vary with perspective. For example, land owners and agricultural scientists may place a higher value on landscape attributes that involve the delivery of provisioning of ecosystem services, while members of the general public may subjectively place a higher value on cultural ecosystem services such as the aesthetics and recreational opportunities (Lothian, 1999). Thus expert opinion may not reflect what is of value personally to individuals or the wider population (Tveit, 2009). Elsewhere, Kirillova et al. (2014) and Plieninger et al. (2013) perform a qualitative assessment of the cultural importance of landscapes, while willingness to pay (WTP) is assessed by Hynes et al. (2011; van Berkel and Verburg, (2014); Rodríguez-Entrena et al., (2017); Dupras et al., (2018); Bernués et al., (2019) and Huber and Finger, (2019). The publics’ stated preferences for landscapes and their features have also been surveyed (Howley, 2011; Howley et al., 2012; Schirpke et al., 2016, Santos-Martín et al., 2019). Stat- ed preference surveys often measure landscapes in a holistic way focusing on concepts or characteristics reflected in the landscape (Ives and Kendal, 2013; Tveit et al., 2006). Many landscape preference studies also employ non-monetary techniques where land- scapes are assessed through rankings of a number of photographs, or monetary techniques to estimate direct and indirect use values (e.g. forest fibres) and/or non-use values (e.g. biodiversity, wilderness, spiritual) for preserving landscapes (García-Llorente et al., 2012). Assessments based on cognitive attributes, such as landscape coherence, mystery, safety, and naturalness, provide a holistic assessment of a visual entity through its single compo- nents, rather than defining or focusing on specific physical landscape attributes, such as tree density or presence of hedges (Tagliafierro et al., 2013; van Zanten et al., 2014). Hynes and Campbell (2011) analysed the most appropriate economic valuation methodologies for agri-environment policies. They concluded that a holistic valuation approach should be used where the objective is the valuation of the landscape as a whole, whereas an attribute-based approach is appropriate if the objective is to understand preferences for individual compo- nents, which may allow for extrapolation using other GIS datasets in policy evaluation. Choice experiments have been utilised to assess the preference for individual char- acteristics (Hynes and Campbell, 2011; Rodríguez-Entrena et al., 2017; Dupras et al., 173Assessing preferences for rural landscapes 2018). Although they present monetary measures of the willingness to pay for landscape attributes, there is a limit to how many attributes can be considered, albeit some papers (such as Bernués et al., 2019) have an extensive array of choice attributes. Thus, it may be difficult to apply a choice experiment methodology to assess the preferences for a wide variety of landscape characteristics. García-Llorente et al. (2012) used photographs within the contingent valuation method to examine preferences for alternative landscape types. Follow-up expert opinion was employed to relate the observed willingness to pay for eco- system services connected to the different landscapes in the photographs. Two studies of particular relevance to this research are Howley (2011) and Schirpke et al. (2016). Howley (2011) assessed the effect of personal, geographic and environmental value orientations on landscape preferences. They did not however examine how the land- scape attributes themselves could influence preferences or whether the potential effects could vary across survey respondents according to their personal, socio-demographic and geographic characteristics. Schirpke et al. (2016) similarly examined attitudes in relation to landscape images by assembling specific landscape attributes using viewsheds from a digital elevation model. Although Schirpke et al. (2016) consider the relationship between socio-economic characteristics and holistic image-based landscape attributes (as does Howley, 2011), their study does not consider the differential preference for specific land- scape preferences across socio-demographic characteristics. This paper aims to contribute to the literature of landscape preference valuation by (a) investigating whether individuals’ characteristics interact with landscape attributes, and (b) how these interactions may ultimately affect public preferences for landscapes. The paper used data from Howley’s (2011) analysis and builds on Schirpke et al. (2016)’s approach by applying expert judgement as opposed to a combination of GIS-based and observational attributes to each of the photos. The literature is extended by utilising an attribute choice framework to disentangle individual preferences for a holistic image of a landscape photograph into preferences for specific attributes of that landscape. The approach adopted in this paper facilitates the creation of a formalised model of landscape preferences based on the component attributes. The study uses Ireland’s rural landscapes as a case study. The Irish rural landscape has, and still is undergoing considerable change. Agriculture remains the largest rural land use with the Irish agri-food sector accounting for over half of the country’s exports and almost 10% of the economy and employment (Teagasc, 2017). In many predominant- ly rural countries like Ireland, landscape images provide a visible representation of how the world sees the country and advertising campaigns such as Ireland’s ‘Origin Green’ are used to promote global agri-food exports. As rural based sectors and the public goods they provide are heavily influenced by public policy, societal preferences in relation to rural areas are important. Landscape aesthetics, as one of the most visual and under- standable public goods, is as a result, one of the most important drivers of support for the delivery of additional rural public goods. The next section of this paper presents a review of models of landscape preference as a basis for model development. Section 3 then describes the data used in the analysis. The methodology is reviewed in section 4 while section 5 presents results and discussion. Finally, policy relevant conclusions are provided in section 6. 174 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis 2. Models of Landscape Preferences Increasingly, policy is focusing on the role of landscapes in the provision of ES, with landscape aesthetics being consistently included as an example of cultural ES. Many of these ES relate to the structure and composition of the landscape (Tscharntke et al., 2005; Zhang et al., 2007; van Berkel and Verburg, 2014; van Oudenhoven et al., 2012). A variety of eco- logical/landscape indicators have been used to estimate the relationship between landscape characteristics and the potential for supply of ES (Kienast et al., 2009; Burkhard et al., 2010; van Berkel and Verburg, 2014), whilst integrative analytical approaches and models have been developed to assess trade-offs between ES and economic decisions (Vidal-Legaz et al., 2013). Studies have also assessed the socio-cultural values of ecosystem services delivered by different landscape types (Hynes and Campbell, 2011; Martín-López et al. 2012). While the value of the agricultural provisioning function of landscapes can be quan- tified using farm activity data, the quantification of the aesthetic value of landscapes remains a challenge. There are however studies that focus on particular cultural services that can be attributed to visual landscape characteristics, rather than the totality of poten- tial ES. Such landscape preference studies use landscape photos to represent different types of landscapes (see for example Campbell et al., 2006; Rambonilaza and Dachary- Bernard, 2007; Moran et al., 2007; Hynes and Campbell, 2011). While the use of inter- views with photo-elicitation and ranking enables researchers to identify landscape prefer- ences and propose reasons underlying them, there are some criticisms of the reliability of evaluating aesthetic preference using photos. Bias in stated preferences may arise due to photo quality, light, weather, photo composition, and the number of photos presented (van Berkel and Verburg, 2014, Gill et al., 2015). However, empirical results from numer- ous studies support the use of landscape images and other visual approaches combined with questionnaires, as a reliable method for the public evaluation of landscapes (Svobo- dova et al., 2012; Häfner et al., 2018). 2.1 Landscape Attributes The concept of utilising landscape photographs as a proxy for landscape characteristics is commonplace in the literature (Kaltenborn and Bjerke, 2002; Arriaza et al., 2004). While a photographic image does not represent the actuality of the experience of being in a land- scape, there is a substantial literature that supports their use (Häfner et al., 2018). Accord- ing to Dramstad et al. (2006), preferences based on well-selected colour photographs of landscapes are similar to those made in the field. In this study, landscapes are decomposed into their individual attributes to examine the personal preferences for these attributes. In a meta-analysis, van Zanten et al. (2014) created a typology of landscape attributes consisting of two levels. At the first level there are four attribute groups: human influence on agricultural landscapes, land cover attributes, landscape elements and biophysical fea- tures. The second level decomposes level one attributes into their various components, e.g. farm system, level of fragmentation, mountains etc. Landscape scenes used in preference studies need to account for these different types of attributes. It is also important to distinguish the intensity of the various attrib- utes. Häfner et al. (2018) found there was a higher preference for point attributes such as 175Assessing preferences for rural landscapes individual trees, as opposed to lines of trees or hedgerows, with a higher frequency pre- ferred. The attributes extracted from landscape scenes for this analysis are also in line with those of De Ayala et al. (2012). They list the common attributes in landscape level dis- crete choice experiment studies as vegetation (e.g. trees, hedgerows), rural aspects (grass- land, farm buildings), wildlife, water, cultural heritage (monuments, traditional farming), boundaries (stone walls and fences) and recreation (walking trails, fishing). 2.2 Judgements Landscape has been described as the intersection between physical attributes of a place and individuals’ perceptions of that place (Hanley et al., 2009). Studies examining landscape values may use either expert judgement (objectivist approach), where the focus is on characterizing the landscape as an object, or personal preferences in the form of a survey (subjectivist approach), where the focus is on viewers’ experiences of the landscape (Lothian, 1999; Tveit et al., 2006). The objective approach considers landscape quality as an intrinsic attribute of the landscape, and requires an implicit understanding of human pref- erences for landscape. The subjectivist approach considers landscape quality as a human construct based on the interpretation of what is perceived as landscape through individuals’ memories, associations and imagination. In the subjectivist approach, landscapes provide a means of understanding preferences of landscape viewers (Lothian, 1999). Within the field of landscape aesthetics, evolutionary theories and cultural preference theories have been developed to explain landscape perception and identify the factors and mechanisms that shape human preferences towards landscapes (Häfner et al., 2018). When using personal preferences, the context in which the survey is collected is important. Studies that are context specific make upscaling of results difficult (van Zanten et al., 2014). Studies should therefore control for local context such as attitudes, location and demographics of the respondents. Education, for example, has been found to positively influence landscape preferences (Häfner et al., 2018). However, in an assessment of land- scape aesthetics, Frank et al. (2013) found few differences in the preference values across three different categories of respondents: the general population, experts and stakeholders. The location in which a respondent lives can also influence their preferences. Meta- analysis results show that urban residents have a higher preference for forest and natural landscapes (van Zanten et al., 2014). The landscape value of an area also includes the val- ue placed on it by tourists and those not living in an area. Kirillova et al. (2014) examined the aesthetic judgement of tourists using semi-structured interviews and disaggregated their judgements into a total of nine dimensions. Zoderer et al. (2016) found that tour- ists’ perceptions of landscape value vary with the land-use type and their socio-economic characteristics. In summary, some of the spatial, methodological and attribute choices in recent studies are presented in Table 1. 3. Methodological Framework A range of indicators is required to comprehensively describe landscapes. The European Landscape Convention (ELC, 2000) for example integrates biophysical, cul- 176 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis tural, social, and visual attributes of landscapes. In order to incorporate this integrated view and to combine public and expert opinion, Sowinska-Swierkosz and Chmielewski (2016) developed a methodological framework to identify Landscape Quality Objectives (LQOs) which include GIS analysis, quality assessments, social survey and expert value judgements. This study also combines expert and public viewpoints in developing a model that links the visual attributes of landscapes (as defined by agricultural scientists) with indi- viduals’ landscape preferences, socio-demographic data and GIS analysis. The main ben- efit of using such a modelling approach is the ability to rank a landscape, using personal preferences derived from a survey but without the need to conduct surveys in every loca- tion. Similar to the use of value transfer approaches this means that the parameters of the preference model can be used to estimate rank orderings of landscapes without the need for further primary surveys providing time and monetary savings to both researcher and policy maker (Hynes et al. 2018). In creating a formalised model of landscape preferences, it is first necessary to define the characteristics or attributes of landscapes. In doing so, choices are made (discussed previously), between broad holistic descriptions and more discrete, generalisable and quantifiable attributes of the landscape. The objective of this study to estimate a landscape preference model that is generalisable in an Irish context, thus a model of quantifiable landscape attributes is developed (equation 1) where: Max U = ∑i βi × li (1) Table 1. Choice of Landscape Attribute in Recent Studies. Paper Country General scene or attributes Scale (local or national) Expert or survey Häfner et al. (2018) Germany Attributes Local Stated preference survey (n=200) Hermes et al. (2018) Germany. 100m x 100m Scene National Expert Vidal-Legaz et al. (2013)Spain. No spatial component Scene Local Stated preference survey (n=226) van der Jagt et al. (2014)Scotland Scene Local Preference matrix survey (n=100) Zoderer et al. (2016) Italy Scene Local Stated preference survey (n=659) Frank et al. (2013) Germany Scene Local Survey consisting of laymen and experts (n=153) Bernués et al. (2019) Multiple countries (Spain, Norway, Italy) Attributes Country regional/ provincial Stated preference survey (n=1,044) Dupras et al. (2018) Canada (three regions; Saint-Jacque, Repentigny, and Montréal) Attributes Country regional Survey consisting of laymen (n=250) 177Assessing preferences for rural landscapes such that in maximising landscape preferences, or in economic terms utility from the landscape, U, a series of parameters βi are estimated that indicate a level of preference for individual attribute li. As a social science analysis, we are interested not only in the landscape attributes that are preferred but also preference heterogeneity across individuals or across groups of attributes , with personal characteristics and attitudes Z. Max Uj = ∑i βi × li × Zj (2) Individuals’ preference heterogeneity can be decomposed into different components. Beyond standard demographic characteristics in describing different groups, attitudinal factors are important (Swanwick, 2009). Appleton (1975) argues that individual preferenc- es for landscapes depend upon the relationship between an individual and their environ- ment, their experiences of the landscape, where individuals live and how they experience the landscape, while Howley (2011) finds heterogeneity in landscape preferences due to both demography and environmental orientations. The model should therefore account for the different drivers of preference variability (equation 3): Max Uj = ∑i βi × li × Zj (Demograhics,Attitudes,Location) (3) In order to understand the structure of individuals’ preferences for landscape attrib- utes, survey respondents were first asked to rank preferences for individual photographs on a 6-point Likert scale from (1) ‘not very highly’ to (6) ‘very highly’. While the ranking variable is potentially continuous over the range 1 to 6, discrete values were used for con- venience. Treating the ranking as an underlying continuous variable, an Ordinary Least Squares (OLS) model, of the form: Yi = β’Xi + εi (4) can be used for individual i, where Yi * is the dependent variable reflecting landscape pref- erences and Xi the explanatory variables and εi the error term. As an alternative modelling strategy the dependent variable can also be treated as dis- crete and the ranking is ordinal, an ordered logit model is employed (Greene, 2004): Yi * = β’Xi + εi (5) for individual i, where Yi * is the underlying latent variable reflecting landscape preferences and Xi the explanatory variables and εi the error term. Where there are six preference values 1,…,6, the following is the observed value of the dependent variable: Y = 1 if 0 < Yi * < μ1 (6)Y = 2 if μ1 < Yi * < μ2 Y = 6 if μ5 < Yi * < μ6 178 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis where Y is the preference value for the landscape image and μ, the vector of unknown threshold parameters that is estimated with the β vector. Since the dependent variable is an ordered, qualitative variable, we estimate the relationship between Y and X with an ordinal response model assuming a logistic distribution. However, as respondents were asked to rank their preference level, the difference between ranking variables has a meaning and is consistent between values. Given that the difference between values has a meaning, utilising the ordered logit loses information in the estimation. Thus even though the survey respondents use discrete values in their judgement, a continuous framework is also employed to model preferences. 3.1 Landscape Attributes In classifying landscape attributes, we move from preferences for individual pho- tographs to preferences for a number of specific attributes . These include agricultural attributes, natural attributes, human-built non-agricultural attributes, topography and other attributes. Given the nature of the data, where there are repeated values for each survey respondent for each of the 30 attributes selected, we employ a fixed effect panel data ordered logit model (Greene, 2001, 2004), which has been widely used for attitudinal studies (Fairlie et al., 2014), for the panel data continuous dependent variable: Yij = β’Zij + ui + εij (7) and for the panel data ordered logit (equation 8): Yij * = β’Zij + ui + εij (8) where Zij represents the landscape characteristics’ specific attributes, ui represents the indi- vidual fixed effect and where the panel data variance component σ2 u is also estimated. 3.2 Preference Heterogeneity We move from person-specific preferences (Xi) in the cross-sectional ordered logit model to landscape attributes (Zij) in the panel data model. Interaction terms (taste-shift- ers) between the personal and the landscape attributes are incorporated in equation 9 so that the influence of personal characteristics on preferences can be examined: XZij = Xi × Zij (9) to produce the following model: Yij * = β’Zij + β1’XZij + ui + εij (10) However, given that there are many landscape characteristic attributes, we combine the attributes into three aggregate characteristics representing natural, agricultural and human-built (non-agricultural) attributes: 179Assessing preferences for rural landscapes (11) 4. Data To assess the preferences of the public in relation to landscape attributes, a nation- ally representative survey1 of 430 individuals aged 15+ was conducted in Ireland in 2010 (Howley, 2011). The survey contained a number of components including: • personal information and demographic characteristics • preferences and attitudes to agriculture, the environment and natural resources • landscape characteristics This demographic and environmental information is later interacted with the respondents’ locations to generate ‘taste-shifters’. The initial parts of the survey also elic- ited responses in relation to the respondent’s environmental attitudes and orientations. Respondents were then asked to indicate their preferences (from 1 - not very highly ranked, to 6 – highly ranked) at an aesthetic level, for a range of photographs of rural landscapes. Respondents were asked to make full use of the ranking scale and to give the highest ranking to their most preferred landscapes. 4.1 Landscape preferences To ascertain landscape preferences, 50 photographs of rural landscapes with a vari- ety of different characteristics were presented to survey respondents. The photos used were selected from a database of 1,000 photos from the national agricultural development authority. They were selected in collaboration with colleagues to attempt to be representa- tive of rural settings, incorporating extensive farming landscapes along with intensive farming landscapes. As the process of selecting images to represent the range of land- scapes is relatively arbitrary, it is possible that a different set of photos would produce different outcomes. In order to improve reliability, photos were selected that had similar weather and light conditions. To ensure a representative sample, the survey was collected at different times of the day over the summer months. Tables 2 and 3 respectively report the six most preferred and the six least preferred landscapes. The most obvious conclusion is that there is a higher preference for water and coastal features in the landscape. Similarly, the presence of animals or heritage features is important. On the other hand, the least preferred landscapes contain human-built fea- tures such as motorways or wind turbines and also contain disorder such as flooding or unmanaged scrub and grassland or contain harvested peat bogs. In the Data Annex, we report the preferences for all photographs. Beyond the six most preferred, the next cohort of photos represents well-managed pastoral agriculture scenes and broadleaf forests/trees. Those photos ranked just above the least preferred landscapes, represent intensive cere- al and horticultural farming on the one hand, as well as marginal scrubland, along with conifer forest. 1 Quota sampling and survey validation are reported in Howley (2011). 180 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis Table 2. Most Preferred Landscapes (photo numbers correspond to ranks in Table A1-Data Annex). 1. Coastal image of sea and headland 2. Aerial photo of a river estuary 3. Coastal cliffs 4. Lake in rural setting 5. Horses in field 6. Large tree next to castle ruin in field Table 3. Least Preferred Landscapes (photo numbers correspond to ranks in Table A1). 50. Flooded farmland 49. New motorway cutting through landscape 48. Scrubland next to woodland 47. Barren hillside with wind turbine 46. Landscape of industrial bogland 45. Trees and scrubland with blue horizon 181Assessing preferences for rural landscapes 4.2 Landscape Attributes This study took a relatively simple approach to classifying attributes, attempting to score the significant presence of an attribute, rather than trying to grade the photo for the degree of importance of a particular attribute. Thus the presence of an attribute that was immediately visible on a quick inspection was scored as 1, as it was felt that these reflect the dominant attributes of an image. If an attribute was not immediately visible on a quick inspection, the attribute was scored as 0. Thus while each photograph has a specific rat- ing of 1-6, we have added additional dummy attributes or explanatory variables for each photo. In the dataset, it is expressed as a separate line for every attribute, with 1 for the presence of the attribute and a 0 otherwise. It thus appears as a panel, with personal char- acteristics invariant over the panel and landscape attributes varying over the panel. Table 4 describes the share of ratings from ‘not very highly’ (1) to ‘very highly’ (6) for these landscape attributes based on the original landscape rankings. Ranking these attrib- utes on the basis of where they appear in landscapes with ‘very highly’ ranked preferences, we note the higher preferences for the attributes lakes, cliffs, horses, water, monuments, hedgerows and Connemara-type landscape which can be collectively described as ‘land- scape descriptions’2. The next highly ranked attributes can be described as ‘pastoral agri- culture’ attributes such as livestock and pasture. At the other end of the preference scale, anthropogenic features such as wind turbines, fencing and problems like flooding and rough grazing landscapes (including gorse) have the lowest preference rankings. 4.3 Environmental Attitudes To gain a deeper understanding of how environmental attitudes might influence land- scape preferences, the survey instrument included questions relating to preferences for landscapes as a provider of ES (in addition to its aesthetic or intrinsic value), or as a pro- vider of food and fibre, and questions relating to negative attitudes towards the environ- ment in general. The resulting environmental attitudes were aggregated using factor analy- sis as described by Howley (2011), resulting in three underlying factors that accounted for 61% of the underlying variation in responses to the attitudinal statements, namely ‘multi- functionalist’, ‘productivist’ and ‘environmental apathy’. These factors are used in the mod- els as explanatory variables. 4.4 Spatial heterogeneity Given the heterogeneity of landscapes, spatial heterogeneity of preferences for attrib- utes may exist. Previous approaches to account for this used distance decay, where WTP is a function of distance between residence and the site being valued (Hanley et al., 2003) or where area-based approaches improve basic distance decay using a radial analysis to model WTP as a function of both distance and quantity of the ES (Granado-Díaz et al., 2020). The distance decay function may also be impacted by the presence of substitute environmental attributes (Jørgensen et al., 2013). Use and non-use values are also impact- 2 Connemara is a remote, scenic, rugged landscape in the west of Ireland. 182 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis ed by distance (Jørgensen et al., 2013). For option value related reasons, non-users may prefer an improvement in local landscapes (Hanley et al., 2003). We also attempt to cap- ture some of the spatial heterogeneity of preferences by using an urban-rural classification based on the respondent’s location. Summary statistics for a variety of taste shifters are presented in Table 5. These are categorised in terms of city, town and rural dwellers and include characteristics of individ- Table 4. Landscape Attribute Summary Statistics showing shares of preference rankings from not very highly (1) to very highly (6). Attribute 1 2 3 4 5 6 Lakes 0.04 0.05 0.09 0.16 0.21 0.46 Cliffs 0.07 0.06 0.07 0.13 0.22 0.44 Horses 0.1 0.03 0.09 0.17 0.2 0.41 Water 0.06 0.08 0.13 0.15 0.23 0.36 Monuments 0.09 0.08 0.11 0.18 0.24 0.3 Hedgerows 0.07 0.07 0.15 0.19 0.24 0.28 Connemara-type landscape 0.04 0.19 0.16 0.17 0.19 0.25 Pasture 0.1 0.09 0.15 0.2 0.22 0.24 Sloping 0.11 0.12 0.16 0.18 0.21 0.23 Stonewalls 0.11 0.1 0.15 0.21 0.22 0.22 Cattle 0.13 0.1 0.15 0.2 0.21 0.21 Mountains 0.16 0.17 0.16 0.15 0.16 0.2 Neat Agricultural Landscape 0.1 0.13 0.16 0.2 0.21 0.2 Sheep 0.11 0.11 0.16 0.21 0.22 0.2 Green 0.11 0.13 0.17 0.19 0.2 0.2 Blue Sky 0.15 0.17 0.17 0.17 0.17 0.18 Bog (peatland) 0.15 0.16 0.16 0.17 0.17 0.18 Sunny 0.14 0.17 0.17 0.18 0.17 0.17 Native Trees 0.18 0.16 0.16 0.17 0.17 0.17 Old Buildings 0.1 0.14 0.17 0.21 0.21 0.17 Flowers 0.11 0.17 0.19 0.21 0.19 0.14 Flat 0.16 0.19 0.18 0.17 0.16 0.14 Cars and Machinery 0.22 0.2 0.17 0.15 0.14 0.13 Crops 0.13 0.17 0.19 0.21 0.18 0.12 Turf 0.18 0.23 0.19 0.16 0.1 0.12 Brown 0.19 0.21 0.18 0.16 0.14 0.12 Yellow 0.16 0.16 0.2 0.2 0.17 0.12 Unmanaged Landscape 0.19 0.21 0.2 0.16 0.13 0.12 Conifer Trees 0.17 0.18 0.21 0.18 0.15 0.11 Other Buildings 0.18 0.2 0.2 0.17 0.15 0.1 Gorse 0.26 0.21 0.18 0.14 0.12 0.09 Fencing 0.4 0.21 0.12 0.09 0.09 0.09 Turbine 0.21 0.21 0.19 0.18 0.13 0.08 Flooding 0.58 0.27 0.1 0.03 0.01 0.01 183Assessing preferences for rural landscapes ual respondents, along with their environmental attitudes, illustrating the degree to which preferences vary depending on where respondents live. Specifically, social and demo- graphic information includes respondent’s age range as a continuous variable with values of 1 (under 30) to 4 (60+), with dummy variables indicating respondents’ education level and whether they have a child. Two social groups were created; the first includes manual workers and unemployed individuals, whereas professional and managerial workers were classified in the second social class (high social class). In addition, respondents or family members who are involved in farming were compared with those without a farming back- ground. Similarly, dummy variables were created to control for the importance of land- scape in choosing where to live, the level of respondents’ satisfaction with respect to the area in which they live, the quality of surrounding landscape, and their concern about the environment and conservation. 5. Results The results of the models of landscape attribute preferences are considered separately for the ordinal logit and the continuous dependent variable panel and pooled OLS mod- els. The influence of personal characteristics on preferences, using taste-shifters (interac- tion terms) between the personal characteristics and landscape types are also presented and discussed. In Table 6, the coefficients for the landscape attributes are reported in terms of natu- ral, agricultural and human-built features, as well as other general attributes such as col- our and unmanaged landscapes. Although there are many variables, the OLS specification Table 5. Summary Statistics of Personal Characteristics and Environmental Preferences used as Taste Shifters. Personal Characteristic City Town Rural Total Has a Child (p) 0.365 0.352 0.424 0.379 Aged Under 30 0.256 0.246 0.250 0.251 Aged 30-50 0.410 0.423 0.394 0.409 Aged 50-60 0.103 0.092 0.152 0.114 Aged 60+ 0.231 0.239 0.205 0.226 University Educated 0.442 0.254 0.242 0.319 Believes landscape is important in choosing where to live 0.186 0.268 0.424 0.286 Satisfied with area in which they live 0.147 0.113 0.106 0.123 Believes surrounding landscape is of high quality 0.487 0.599 0.689 0.586 Higher Social Class 0.763 0.634 0.606 0.672 Farming Background 0.231 0.394 0.614 0.402 Care about Conservation 0.301 0.359 0.432 0.360 Concerned about the environment 0.186 0.324 0.242 0.249 Factor Loading: Multifunctionalist -0.114 0.128 -0.002 0.000 Factor Loading: Environmental Apathy -0.036 0.114 -0.081 0.000 Factor Loading: Productivist -0.173 0.073 0.126 0.000 184 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis Table 6. Coefficients of Panel and Pooled Ordered Logit Model and Panel and Pooled OLS Models for Landscape Attributes. Panel Ordered Logit Model Pooled Ordered Logit Model Panel OLS Model Pooled OLS Model Explanatory Variables Beta SD Beta SD Beta SD Beta SD Natural Landscape Characteristics                 Connemara type landscape 1.397* 0.419 1.454* 0.118 1.005* 0.073 0.98* 0.075 Lakes 0.988* 0.531 1.004* 0.149 0.67* 0.096 0.662* 0.094 Cliffs 1.033* 0.294 1.015* 0.083 -0.039 0.035 0.592* 0.052 Water 0.628* 0.202 0.632* 0.056 0.393* 0.041 0.383* 0.036 Flowers 0.37* 0.213 0.383* 0.058 0.214* 0.045 0.258* 0.038 Bogland 0.35* 0.016 0.349* 0.016 0.189* 0.01 0.213* 0.01 Sloping 0.193 0.182 0.165* 0.05 0.117* 0.033 0.11* 0.032 Native Trees 0.034 0.138 0.061 0.038 0.066* 0.03 0.057* 0.025 Mountains -0.188 0.26 -0.136* 0.072 -0.095* 0.049 -0.097* 0.046 Flat -0.389* 0.153 -0.254* 0.051 -0.113* 0.027 -0.138* 0.033 Conifer Trees -0.547* 0.314 -0.435* 0.087 -0.262* 0.064 -0.25* 0.057 Gorse -0.806* 0.269 -0.639* 0.08 -0.34* 0.047 -0.365* 0.052 Flooding -2.409* 0.482 -2.375* 0.134 -1.681* 0.086 -1.708* 0.085 Agricultural Landscape Characteristics                 Horses 0.533 0.467 0.6* 0.132 0.26* 0.084 0.26* 0.082 Neat Agricultural Landscape 0.424* 0.232 0.362* 0.067 0.167* 0.048 0.198* 0.043 Pasture 0.184 0.176 0.166* 0.049 0.115 0.194 0.109* 0.032 Crops -0.375 0.274 -0.368* 0.093 -0.249 0.298 -0.246* 0.06 Cut-Silage -1.245* 0.591 -1.145* 0.169 -0.755 0.65 -0.688* 0.11 Human Landscape Characteristics                 Monuments 0.947* 0.294 0.874* 0.096 0.618* 0.321 0.568* 0.061 Hedgerows 0.347 0.302 0.291* 0.095 0.297 0.329 0.257* 0.061 Stonewalls 0.117 0.309 0.109 0.095 0.123 0.338 0.123* 0.062 Old Buildings 0.026 0.3 0.036 0.094 0.073 0.328 0.082 0.06 Turf -0.122 0.397 -0.33* 0.125 -0.099 0.434 -0.26* 0.079 Turbine -0.271 0.326 -0.429* 0.105 -0.151 0.355 -0.284* 0.068 Other Buildings -0.167 0.212 -0.395* 0.075 -0.1 0.228 -0.273* 0.048 Cars and Machinery -0.382* 0.225 -0.509* 0.074 -0.285 0.244 -0.393* 0.047 Other Landscape Characteristics                 Yellow 1.019* 0.388 0.905* 0.107 0.646 0.428 0.564* 0.069 Green 0.129 0.155 0.138* 0.042 0.091 0.171 0.092* 0.027 Unmanaged Landscape 0.072 0.257 -0.051 0.074 0.035 0.283 -0.062 0.048 185Assessing preferences for rural landscapes is satisfactory from a multi-collinearity perspective, as the VIF (Variance Inflation Fac- tor) for all values is less than 10 (Kassie et al., 2008). In comparing the models, it is evi- dent that virtually all of the coefficients are within the significance limits of the panel data ordered logit model, so that the models do not in general have substantial differences in their coefficients. We note however that the confidence intervals are wider for the panel data ordered logit than for the pooled version of the model or for the panel and pooled OLS specifications, reflecting perhaps that we utilise less of the information in the panel ordered logit model estimation than the in the pooled version or continuous dependent variable OLS models. Unsurprisingly the Breusch-Pagan Lagrangian multiplier finds the fixed effects insignificant. Therefore, we focus on the OLS pooled model for the discussion and for the introduction of the taste shifter interactions. Overall, the pseudo R2 is 24%, representing relatively large unexplained heterogeneity of landscape preferences. Of the natural attributes, the Connemara type landscape, which represents a remote rugged mountainous area, has the highest positive coefficient. This is followed by prefer- ences for cliffs, lakes and water as landscape attributes. Landscapes with flowers, native trees, bog (peat), sloping land and native trees have the next highest coefficients. Land- scapes with flooding have the lowest coefficient of the natural landscapes. The mountain landscape has an unexpected sign, but it shares considerable information with the Conne- mara type landscape. In relation to the agricultural landscape attributes, the presence of horses has the greatest positive significance, followed by neat agricultural land and pasture, whilst crops and cut-silage have negative coefficients. In relation to human-built landscape attributes, the presence of monuments has the highest positive and significant coefficient. Indeed, it has the second highest coefficient overall. Human-built landscape attributes associ- Panel Ordered Logit Model Pooled Ordered Logit Model Panel OLS Model Pooled OLS Model Brown -0.368* 0.204 -0.325* 0.056 -0.261 0.226 -0.238* 0.036 Constant         3.686 0.276 3.674 0.061 Cut Point 1 -2.623 0.258 -2.548 0.102 Cut Point 2 -1.435 0.256 -1.378 0.096 Cut Point 3 -0.202 0.256 -0.175 0.095 Cut Point 4 1.105 0.256 1.102 0.095 Cut Point 5 2.531 0.256 2.508 0.096 Sigma Squared (u)         0.358       Sigma Squared (e)         1.149   1.168   Rho         0.089                   0.223   Pseudo R2     0.079           Within         0.080       Between         0.866       Overall         0.240       N 20600.000   20600.000   20600.000   20600.000   Number of Groups 50.000       50.000       186 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis ated with farming such as hedgerows and stone walls have the next highest coefficient, followed by old buildings. Meanwhile negative preferences are observed on average for industrial or mechanised objects or activities such as wind turbines, cars and industrial turf-cutting. Also, yellow and green colours (conditional on other attributes) are positive while unmanaged rural landscapes have a negative coefficient. This preference for man- aged agricultural landscapes highlights the frequent mismatch between aesthetic prefer- ences and ecological diversity (Gobster et al., 2007). Interestingly, amongst the least pre- ferred landscapes are unmanaged (potentially biodiversity-rich) landscapes, perhaps reflecting evolutionary processes that favour landscapes that have a greater possibility of providing food and shelter. 5.1 Taste Shifters We interact personal characteristics and attitudes with preferences for natural, human built and agricultural attributes to form taste shifters. Interaction terms between the per- sonal characteristics and landscape types allow us to examine the influence of personal characteristics on preferences and are a means of controlling for observed heterogeneity in preferences within the model. In interacting personal characteristics and landscape charac- teristics, we group characteristics into natural, human and agricultural characteristics, thus reducing the degrees of freedom. Reflecting the fact that attributes have both positive and negative signs in Table 8, we break up the groups into positive and negative coefficients. We combine 15 personal characteristics with six different types of landscape attribute. Given that there are 90 combinations of these variables with potentially overlapping infor- mation and multi-collinearity, we use a Principal Component Analysis (PCA) to reduce the dimensionality, and present the detailed results in Table A2 Data Annex. Although there are 25 factors with an Eigenvalue of more than 1, accounting for 75% of informa- tion, on the grounds of parsimony, we select only those with an Eigenvalue of 2 or higher (Hair et al., 2010). To aid the interpretation of these, we employed a method known as Component Rotation (Bechtold and Abdulai, 2014). This method was used to distinguish between components and to facilitate the interpretation of components (see Table A3 Data Annex for detail on rotated components). The widely applied Varimax Rotation (Abdi and Williams, 2010) was also employed. Table 7 presents the interpretation of the principal components and the coefficients of the pooled OLS model interacted with the taste shift- ers, referencing both the socio-economic characteristics and the landscape attribute group associated with the principal component. For half of these principal components, a single socio-economic characteristic was found to be dominant combined with four landscape attribute groups, positive natural, positive agricultural, positive human and negative natu- ral, highlighting a coherent association with different landscape attribute types. Taste shifters capture preference heterogeneity relative to observed characteristics. For example, PC1 corresponds to a negative coefficient on agricultural landscapes for high social class (professional and managerial workers) city dwellers. A positive coefficient on this component suggests a less negative preference for crops and cut-silage than other groups. PC2 refers to the preferences of town dwellers for human and agricultural characteristics that have a positive coefficient. Here, a positive coefficient indicates a higher preference for these attributes than the general population. PC3 relates to preferences for natural attributes, 187Assessing preferences for rural landscapes where higher social classes, older respondents and those living in towns have lower than average preferences for these attributes. There is a similar impact on human attributes (PC4) with a negative score for higher-educated city dwellers or those with children. City dwell- ers in PC5 have lower than average preferences for human and agricultural attributes, while for PC6, town dwellers have higher preferences for both positive and negative agricultural attributes and more negative human attributes than average. In PC8, those with environ- mental concerns and landscape views have a lower preference for negative human aspects. The remaining principal components all relate to individual socio-economic charac- teristics interacted with the four sets of attributes highlighted above. Those that place a high ranking on the importance of landscape in choosing where to live have higher land- scape preferences than average, while those that are concerned about the environment or with multi-functional attitudes have lower preferences. In summary, grouping the landscape attributes into natural, agricultural (including human built) and non-agricultural human-built attributes, the results show positive asso- ciations with natural attributes such as cliffs, mountainous landscapes, landscapes with water and native trees, neat/managed agricultural landscapes and traditional human-built features such as stone walls and planted hedgerows. The results, as expected, show nega- tive associations with events such as flooding, unmanaged landscapes, industrial turf cut- ting and mechanised features. Table 7. Coefficients of Pooled OLS Model interacted with Taste Shifter Principal Components. Explanatory Variables Landscape Characteristics Interactions Interpretation Landscape attributes Coefficient Standard Error PC1 High social class city dwellers na 0.038 0.005*** PC2 Town dwellers ph pa 0.019 0.007** PC3 Older, town dwellers and higher social class nn -0.031 0.008*** PC4 Higher educated city dwellers with children nh -0.042 0.006** PC5 City dwellers ph pa -0.015 0.006*** PC6 Town dwellers pa, nh, na 0.015 0.006*** PC7 Satisfaction of area pn, ph, pa, nn, -0.030 0.004*** PC8 Environmentally concerned nh -0.035 0.006*** PC9 Importance of landscape in choosing where to live pn, ph, pa, nn, 0.040 0.005*** PC10 Farming background pn, ph, pa, nn, 0.005 0.005 PC11 Multi-functional agriculture pn, ph, pa, nn, -0.048 0.005*** PC12 Concerned about the environment pn, ph, pa, nn, -0.015 0.006** Note: pn – natural attributes (positive sign); ph – human attributes (positive sign) ; pa – agricultural attributes (positive sign); nn – natural attributes (negative sign); nh – human attributes (negative sign); na – agricultural attributes (negative sign). 188 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis There is significant preference heterogeneity however with different groups favouring or disfavouring different attributes. An urban-rural classification used to capture the spa- tial heterogeneity of preferences (based on the respondents’ locations) showed that those living in urban areas feel they have a lower quality of surrounding landscape compared to rural areas. Unsurprisingly those that have chosen to live in a rural landscape place the highest value on this type of landscape, while farmers have the highest preference for agricultural landscape attributes. Urban dwellers are more indifferent towards natural and farming landscapes. Underlying eco-centric attitudes are also important drivers. 6. Conclusions This study adopted an attribute choice framework to disentangle individual prefer- ences for a holistic image of landscape photographs into preferences for specific attributes of that landscape, and subsequently used these attributes in landscape preference models to relate societal preferences to quantifiable landscape attributes. The study further investi- gated whether individuals’ characteristics interact with landscape attributes and how these interactions ultimately affect public preferences for landscapes. This paper adopts a middle-ground approach between the methods found in the lit- erature for landscape preference modelling. On the one hand, it is ambitious in relation to the range of landscape attributes as in the case of Schirpke et al. (2016) or Bernués et al. (2019), but is less ambitious in focusing on preference attributes rather than will- ingness to pay, as in the stated preference valuation literature. It also extends the work of Schirpke et al. (2016) by considering the preference heterogeneity for specific landscape attributes. Although unobserved heterogeneity is not considered in this study, the variety of observed heterogeneity incorporated may be more useful for policy and from a simula- tion modelling perspective. Ultimately, the model results highlight differences in how peo- ple with different attitudes and characteristics rank landscape features. The impact of taste shifters on various groups illustrates the heterogeneity in rankings. As noted by Hynes et al. (2011) the attribute based approach to landscape prefer- ences allows the researcher to examine the general trade-offs which society is willing to make between different attributes of the countryside. On the other hand, modelling land- scape preferences based on landscape photos, such as in Howley’s (2011) study, is useful if the researcher is interested in understanding preferences for the wider landscape. The approach adopted here is particularly useful where one is interested in the utility gained or lost through a policy that may cause only incremental changes in the landscape or impact on only a small number of attributes. Interacting personal characteristics as taste shifters can help us to understand local preferences if the characteristics of the local popu- lation differ. The analysis does have the limitation of not being able to identify local pref- erences in terms of sense of place or relational value. Qualitative studies or localised sur- veys are needed to understand these more nuanced perspectives (Pérez-Ramírez et al., 2019; Vannier et al., 2019; Wartmann and Purves, 2018). Moving from a holistic view of landscapes to analysing preferences for specific attrib- utes of landscapes allows for further applications particularly in the area of simulation. Being able to assess preferences for an individual attribute makes it possible to extrapolate the preference ranking of a landscape in an area that has not been ranked directly. It is 189Assessing preferences for rural landscapes important to note however that the method adopted in this paper is based on the assump- tion that the sum of the singular landscape element’s preference scores equates to the preference ranking of the landscape as a whole. That simplifies the way in which humans value the environment and should be considered as a limitation of our study. As such, the method is more appropriate when there are only a limited number of attributes to be con- sidered in a given landscape. Human-built landscape characteristics such as stone walls and hedgerows are found to be positively associated with the preference rankings of photos in this study. Thus, future land-use changes and landscape development plans should promote the aesthetic role of stone walls and hedgerows and prioritise their conservation. Similarly, the recognition of the high aesthetic value that the public places on well-managed/neat agricultural land- scapes provides policy justification to incentivise farmers to maintain these public goods in future agri-environmental schemes. The results presented in this paper provide evidence of the preferences of a diverse range of individuals across a number of characteristics that should be of assistance to policy makers attempting to maximise the benefit for society from rural landscapes. The model developed here provides information for decision-makers to examine whether a proposed policy change involving one or more landscape attributes will have a positive or negative impact across the population, while also allowing for more targeted policy forma- tion by disaggregating the population into different preference cohorts. The approach adopted in this paper facilitates the creation of a formalised model of landscape preferences based on the component attributes. Decomposing complete land- scape images into quantifiable attributes is a common feature of preference studies and can help bridge the gap between the GIS literature and landscape analysis. The latter typi- cally takes quantifiable landscape attributes from GIS datasets to create typologies of dif- ferent types of landscapes. Meanwhile the former assesses societal preferences for holistic images. Our methodology can further allow for the application of societal preferences to quantifiable datasets of landscape attributes, rather than using expert judgement as is cur- rently the case. The approach developed in this study therefore, has implications for planners for Landscape Character Assessments (LCA) that often utilise a broad expert knowledge approach to developing LCA maps, which may under/over estimate the value of various landscape attributes. Future work will apply this methodology in a GIS landscape data- base to re-assess LCAs from a societal rather than an expert point of view. Future work should also test for the existence of spatial dependence and use spatial regression methods to examine spatial heterogeneity in more detail. Acknowledgement The authors acknowledge research funding received from Department of Agriculture, Food and the Marine (FARM-ECOS project REF 15/S/619), Teagasc Farm Management & Rural Development Department and the EU Interreg Atlantic Area Programme 2014– 2020 (EAPA_261/2016 ALICE). 190 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis References Abdi, H., Williams, L.J., 2010. Principal component analysis. Wiley Interdisciplinary Reviews Computational Statistics 2: 433–459. Appleton, J. (1975). The Experience of Landscape. Chichester: Wiley. Arriaza, M., Canas-Ortega, J.F., Canas-Madueno, J.A. and Ruiz-Aviles, P. (2004). Assessing the visual quality of rural landscapes. Landscape and Urban Planning 69(1): 115-125. Bechtold, K.B., Abdulai, A., 2014. Combining attitudinal statements with choice experi- ments to analyze preference heterogeneity for functional dairy products. Food Policy 47: 97–106. Bernués, A., Alfnes, F., Clemetsen, M., Eik, L.O., Faccioni, G., Ramanzin, M., Ripoll- Bosch, R., Rodríguez-Ortega, T. and Sturaro, E. (2019). Exploring social preferences for ecosystem services of multifunctional agriculture across policy scenarios. Ecosys- tem Services 39: 101002. Burkhard B, Kroll F. and Müller F. (2010). Landscapes’ capacities to provide ecosys- tem services– a concept for land-cover based assessments. Landscape Online 15: 1-22. Campbell, D., Hutchinson, W.G. and Scarpa, R. (2006). Quantifying the landscape benefits arising from the Rural Environment Protection Scheme: Results from a public sur- vey. Tearmann: Irish Journal of Agri-environmental Research 4(1): 14-29. Dachary-Bernard, J. and Rambonilaza, T. (2012). Choice experiment, multiple pro- grammes contingent valuation and landscape preferences: How can we support the land use decision making process? Land Use Policy 29(4): 846-854. De Ayala, A., Hoyos, D. and Mariel, P. (2012). Landscape valuation through discrete choice experiments: Current practice and future research reflections. Documento de Traba- jo BILTOKI DT2012.03, Departamento de Economia Aplicada III (Econometria y Estadística), Universidad del País Vasco, Bilbao. Retrieved from https://addi.ehu.es/ bitstream/handle/10810/8011/2012.03.pdf?sequence=6&isAllowed=y Dramstad, W.E., Tveit, M.S., Fjellstad, W.J. and Fry, G.L. (2006). Relationships between visual landscape preferences and map-based indicators of landscape structure. Land- scape and Urban Planning 78(4): 465-474. Dupras, J., Laurent-Lucchetti, J., Revéret, J.P. and DaSilva, L. (2018). Using contingent val- uation and choice experiment to value the impacts of agri-environmental practices on landscapes aesthetics. Landscape Research 43(5): 679-695. ELC (2000). European Landscape Convention. Florence, October 20, 2000. Fairlie, R.W., Hoffmann, F. and Oreopoulos, P. (2014). A community college instruc- tor like me: Race and ethnicity interactions in the classroom. American Economic Review 104(8): 2567-91. Fotheringham, A.S., Brunsdon, C. and Charlton, M. (2003). Geographically Weighted Regression: The Analysis of Spatially Varying Relationships. Wiley. Frank, S., Fürst, C., Koschke, L., Witt, A. and Makeschin, F. (2013). Assessment of land- scape aesthetics - validation of a landscape metrics-based assessment by visual esti- mation of the scenic beauty. Ecological Indicators 32: 222-231. García-Llorente, M., Martín-López, B., Iniesta-Arandia, I., López-Santiago, C. A., Aguil- era, P. A. and Montes, C. (2012). The role of multi-functionality in social preferences 191Assessing preferences for rural landscapes toward semi-arid rural landscapes: An ecosystem service approach. Environmental Science and Policy 19: 136-146. Gill, N., Dun, O., Brennan-Horley, C. and Eriksen, C. (2015). Landscape preferences, amenity, and bushfire risk in New South Wales, Australia. Environmental Manage- ment 56(3): 738-753. Gobster, P. H., Nassauer, J. I., Daniel, T. C. and Fry, G. (2007). The shared landscape: What does aesthetics have to do with ecology? Landscape Ecology 22(7): 959-972. Granado-Díaz, R., Gómez-Limón, J.A., Rodríguez-Entrena, M. and Villanueva, A.J. (2020). Spatial analysis of demand for sparsely located ecosystem services using alternative index approaches. European Review of Agricultural Economics 47(2): 752-784. Greene, W. (2001). Estimating econometric models with fixed effects. Working Papers 01-10, New York University, Leonard L. Stern School of Business, Department of Economics. Greene, W. (2004). The behaviour of the maximum likelihood estimator of limited dependent variable models in the presence of fixed effects. Econometrics Journal 7(1): 98-119. Häfner, K., Zasada, I., van Zanten, B.T., Ungaro, F., Koetse, M. and Piorr, A. (2018). Assessing landscape preferences: A visual choice experiment in the agricultural region of Märkische Schweiz, Germany. Landscape Research 43(6): 846-861. Hair, J.F., Black, W.C., Babin, B.J., Anderson, R.E., Tatham, R.L., 2010. Multivariate Data Analysis, Prentice Hall. Prentice Hall, New Jersey.Appleton, J. (1975). The Experience of Landscape. Chichester: Wiley. Hanley, N., Ready, R., Colombo, S., Watson, F., Stewart, M. and Bergmann, E.A. (2009). The impacts of knowledge of the past on preferences for future landscape change. Journal of Environmental Management 90(3): 1404-1412. Hanley, N., Schläpfer, F. and Spurgeon, J. (2003). Aggregating the benefits of environmen- tal improvements: Distance-decay functions for use and non-use values. Journal of Environmental Management 68(3): 297-304. Hermes, J., Albert, C. and von Haaren, C. (2018). Assessing the aesthetic quality of land- scapes in Germany. Ecosystem Services 31: 296-307. Howley, P. (2011). Landscape aesthetics: Assessing the general publics’ preferences towards rural landscapes. Ecological Economics 72: 161-169. Howley, P., O’Donoghue, C. and Hynes, S. (2012). Exploring public preferences for tradi- tional farming landscapes. Landscape and Urban Planning 104(1): 66-74. Huber, R. and Finger, R. (2019). A metaanalysis of the willingness to pay for cultur- al services from grasslands in Europe. Journal of Agricultural Economics. doi: 10.1111/1477-9552.12361 Hynes, S. and Campbell, D. (2011). Estimating the welfare impacts of agricultural land- scape change in Ireland: A choice experiment approach. Journal of Environmental Planning and Management 54(8): 1019-1039. Hynes, S., Campbell, D. and Howley, P. (2011). A Holistic vs. an attribute-based approach to agri-environmental policy valuation: Do welfare estimates differ? Journal of Agri- cultural Economics 62(2): 305-329. Hynes, S., Ghermandi, A., Norton, D. and Williams, H. (2018). Marine Recreational Eco- system Service Value Estimation: A Meta-Analysis with Cultural Considerations. Ecosystem Services 31: 410-419. 192 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis Ives, C.D. and Kendal, D. (2013). Values and attitudes of the urban public towards peri- urban agricultural land. Land Use Policy 34: 80-90. Jørgensen, S.L., Olsen, S.B., Ladenburg, J., Martinsen, L., Svenningsen, S.R. and Hasler, B. (2013). Spatially induced disparities in users’ and non-users’ WTP for water quality improvements - Testing the effect of multiple substitutes and distance decay. Ecologi- cal Economics 92: 58-66. Kaltenborn, B.P. and Bjerke, T. (2002). Associations between environmental value orienta- tions and landscape preferences. Landscape and Urban Planning 59(1): 1-11. Kantelhardt, J., Schaller, L. and Viaggi, D. (2015). Do agricultural landscapes provide socio-economic benefits in rural regions? 29th International Association of Agricul- tural Economists Conference, August 9-14, Milan, Italy. Kassie, M., Pender, J., Yesuf, M., Kohlin, G., Bluffstone, R. and Mulugeta, E. (2008). Esti- mating returns to soil conservation adoption in the northern Ethiopian highlands. Agricultural Economics 38(2): 213-232. Kern, M. (2002). Food, Feed, Fibre, Fuel and Industrial Products of the Future: Challenges and Opportunities. Understanding the Strategic Potential of Plant Genetic Engineer- ing. Journal of Agronomy and Crop Science 188(5): 291 – 305. Kienast, F., Bolliger, J., Potschin, M., De Groot, R.S., Verburg, P.H., Heller, I., Wascher, D. and Haines-Young, R. (2009). Assessing landscape functions with broad-scale envi- ronmental data: insights gained from a prototype development for Europe. Environ- mental Management 44:1099-1120. Kirillova, K., Fu, X., Lehto, X. and Cai, L. (2014). What makes a destination beautiful? Dimensions of tourist aesthetic judgment. Tourism Management 42: 282-293. Lothian, A. (1999). Landscape and the philosophy of aesthetics: Is landscape quality inherent in the landscape or in the eye of the beholder? Landscape and Urban Plan- ning 44(4): 177-198. Martín-López, B., Iniesta-Arandia, I., García-Llorente, M., Palomo, I., Casado-Arzuaga, I., García del Amo, D., Gómez-Baggethun, E., Oteros-Rozas, E., Palacios-Agundez, I., Willaarts, B., González, JA., Santos-Martín, F., Onaindia, M., López-Santiago, C., Montes, C. (2012). Uncovering ecosystem service bundles through social preferenc- es. PLoS ONE 7(6): e38970. Moran, D., McVittie, A., Allcroft D.J. and Elston, D.A. (2007). Quantifying public prefer- ences for agri-environmental policy in Scotland: A comparison of methods. Ecologi- cal Economics 63(1): 42-53. OECD (2015). Public Goods and Externalities. Agri-environmental Policy Measures in Selected OECD Countries. OECD Publishing, Paris. Pérez-Ramírez, I., García-Llorente, M., Benito, A., & Castro, A. J. (2019). Exploring sense of place across cultivated lands through public participatory mapping. Landscape Ecology 34(7): 1675-1692. Plieninger, T., Dijks, S., Oteros-Rozas, E. and Bieling, C. (2013). Assessing, mapping, and quantifying cultural ecosystem services at community level. Land Use Policy 33: 118- 129. Rambonilaza, M. and Dachary-Bernard, J. (2007). Land-use planning and public prefer- ences: What can we learn from choice experiment method? Landscape and Urban Planning 83 (4): 318-326. 193Assessing preferences for rural landscapes Rodríguez-Entrena, M., Colombo, S. and Arriaza, M. (2017). The landscape of olive groves as a driver of the rural economy. Land Use Policy 65: 164-175. Santos-Martín, F., Zorrilla-Miras, P., García-Llorente, M., Quintas-Soriano, C., Montes, C., Benayas, J., Gómez Sal, A., Paracchini, M.L. (2019). Identifying win–win situations in agricultural landscapes: an integrated ecosystem services assessment for Spain. Landscape Ecology 34: 1789–1805. Schaller, L., Targetti, S., Villanueva, A., Zasada, I., Kantelhardt, J., Arriaza, M., Bal, T., Bossi Fedrigotti, V., Handan Giray, F., Hafner, K., Majewski, E., Malak-Rawlikowska, A., Nikolov, D., Paoli, J.C., Piorr, A., Rodríguez-Entrena, M., Ungaro, F., Verburg, P., van Zanten, B. and Viaggi, D. (2018). Agricultural landscapes, ecosystem services and regional competitiveness - Assessing drivers and mechanisms in nine European case study areas. Land Use Policy 76: 735-745. Schirpke, U., Timmermann, F., Tappeiner, U. and Tasser, E. (2016). Cultural ecosystem servic- es of mountain regions: Modelling the aesthetic value. Ecological Indicators 69: 78-90. Sowinska-Swierkosz, B.N. and Swierkosz, T.J. (2016). A new approach to the identification of Landscape Quality Objectives (LQOs) as a set of indicators. Journal of Environ- mental Management 184: 596-608. Svobodova, K., Sklenicka, P., Molnarova, K. and Salek, M. (2012). Visual preferences for physical attributes of mining and post-mining landscapes with respect to the soci- odemographic characteristics of respondents. Ecological Engineering 43: 34-44. Swanwick, C. (2009). Society’s attitudes to and preferences for land and landscape. Land Use Policy 26: S62-S75. Tagliafierro, C., Longo, A., Van Eetvelde, V., Antrop, M. and Hutchinson, W.G. (2013). Landscape economic valuation by integrating landscape ecology into landscape eco- nomics. Environmental Science and Policy 32: 26-36. Teagasc (2017). Agriculture in Ireland: The Irish agri-food industry. https://www.teagasc.ie/ rural-economy/rural-economy/agri-food-business/agriculture-in-ireland/ Tscharntke, T., Klein, A.M. and Kruess, A. (2005). Landscape perspectives on agricultural intensification and biodiversity - ecosystem service management. Ecological Letters 8: 857-874. Tveit, M.S. (2009). Indicators of visual scale as predictors of landscape preference; a com- parison between groups. Journal of Environmental Management 90(9): 2882-88. Tveit, M., Ode, Å. and Fry, G. (2006). Key concepts in a framework for analysing visual landscape character. Landscape Research 31(3): 229-255. van Berkel, D.B. and Verburg, P.H. (2014). Spatial quantification and valuation of cultural ecosystem services in an agricultural landscape. Ecological Indicators 37: 163–174. van der Jagt, A.P.N., Craig, T., Anable, J., Brewer, M.J. and Pearson, D.G. (2014). Unearth- ing the picturesque: The validity of the preference matrix as a measure of landscape aesthetics. Landscape and Urban Planning 124: 1-13. van Oudenhoven, A.P.E, Petz, K., Alkemade, R., Hein, L., de Groot, R.S. (2012). Frame- work for systematic indicator selection to assess effects of land management on eco- system services. Ecological Indicators 21: 110-122. van Zanten, B.T., Verburg, P.H., Koetse, M.J. and van Beukering, P.J. (2014). Preferences for European agrarian landscapes: A meta-analysis of case studies. Landscape and Urban Planning 132: 89-101. 194 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis Vanni, F. (2014). Agriculture and Public Goods: The Role of Collective Action. Netherlands: Springer. Vannier, C., Bierry, A., Longaretti, P. Y., Nettier, B., Cordonnier, T., Chauvin, C., Lavorel, S. (2019). Co-constructing future land-use scenarios for the Grenoble region, France. Landscape and Urban Planning 190: 103614. Vidal-Legaz, B., Martínez-Fernández, J., Picón, A. S. and Pugnaire, F. I. (2013). Trade- offs between maintenance of ecosystem services and socio-economic develop- ment in rural mountainous communities in southern Spain: A dynamic simulation approach. Journal of Environmental Management 131: 280-297. Villanueva, A.J., Targetti, S., Schaller, L., Arriaza, M., Kantelhardt, J., Rodriguez-Entrena, M., Bossi-Fedrigotti, V. and Viaggi, D. (2015). Assessing the role of economic actors in the production of private and public goods in three EU agricultural landscapes. Journal of Environmental Planning and Management 58(12): 2113-2136. Wartmann, F. M., & Purves, R. S. (2018). Investigating sense of place as a cultural eco- system service in different landscapes through the lens of language. Landscape and Urban Planning 175: 169-183. Zhang, W., Ricketts, T.H., Kremen, C., Carney, K. and Swinton, S.M. (2007). Ecosystem services and dis-services to agriculture. Ecological Economics 64(2): 253-260. Zoderer, B.M., Tasser, E., Erb, K-H., Stanghellini, P. S.L. and Tappeiner, U. (2016). Iden- tifying and mapping the tourists’ perception of cultural ecosystem services: A case study from an Alpine region. Land Use Policy 56: 251–261. 195Assessing preferences for rural landscapes Data Annex: Assessing population landscape characteristic preferences using disag- gregated attributes for rural landscapes Table A1. Ranking of Photos by Survey Participants. Rank Photo Description 1 Coastal image of sea and headland 2 Aerial photo of a river estuary 3 Coastal cliffs 4 Lake in rural setting 5 Horses in field 6 Large tree next to castle ruin in field 7 Rolling hills, with conifers and well-kept fields 8 Copper beech tree in parkland 9 Sandy Beach 10 Stream flowing through Deciduous forest 11 Patchwork quilt of fields and river 12 The Rock of Cashel Historic Monument 13 Rich farmland and hillside in background 14 Remote hillside, with trees 15 Large rock in field on hillside 16 Field of sheep in lowland good grass and stone walls 17 Hillside of bluebells and deciduous trees 18 Remote (Connemara) mountainous landscape 19 Traditional farm building 20 Forest track in deciduous trees 21 Stonewalls with neat field of sheep 22 Stonewall with cows in field and trees on hillside 23 Dairy cows in field 24 Large field after silage cut 25 Stonewalls with neat field of oil seed rape 26 Sheep in front of traditional farmhouse 27 Statue of harpist in rural village 28 Hilly Woodland and Trees 29 Large field of cereal crops 30 Wildflower in field of ferns 31 Hillside of conifer trees 32 Trees and field of rushes 33 Mature forest 34 Rows of horticulture crops in field 35 Neat rows of cereal crops 36 Rocky mountain with extensive agriculture 37 Tillage field after harvest with blue sky 38 Mechanical cutting of turf from bog 39 Hillside of conifer 196 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis Rank Photo Description 40 Reeds and scrubland 41 Marginal land with trees in background 42 Large horticulture field 43 Barren bogland 44 Heather in bogland 45 Trees and scrubland with blue horizon 46 Landscape of industrial bogland 47 Barren hillside with wind turbine 48 Scrubland next to woodland 49 New motorway cutting through landscape 50 Flooded farmland Table A2. Principal Component Analysis. Principal Component Eigenvalue Cumulative Proportion of Variance P. Component 1 8.79458 0.0977 P. Component 2 6.75435 0.1728 P. Component 3 5.62239 0.2352 P. Component 4 4.78296 0.2884 P. Component 5 4.50848 0.3385 P. Component 6 3.63383 0.3789 P. Component 7 3.32083 0.4157 P. Component 8 2.9566 0.4486 P. Component 9 2.75087 0.4792 P. Component 10 2.55781 0.5076 P. Component 11 2.43572 0.5346 P. Component 12 2.2716 0.5599 P. Component 13 1.72736 0.5791 P. Component 14 1.71575 0.5981 P. Component 15 1.64843 0.6165 P. Component 16 1.49928 0.6331 P. Component 17 1.44856 0.6492 P. Component 18 1.39257 0.6647 P. Component 19 1.34819 0.6797 P. Component 20 1.27425 0.6938 P. Component 21 1.15625 0.7067 P. Component 22 1.09155 0.7188 P. Component 23 1.08972 0.7309 P. Component 24 1.07077 0.7428 P. Component 25 1.03101 0.7543 197Assessing preferences for rural landscapes Ta bl e A 3. R ot at ed c om po ne nt s (o rt ho go na l v ar im ax ) w ith lo ad in g < 0. 3. C or re la tio ns Ty pe C om p1 C om p2 C om p3 C om p4 C om p5 C om p6 C om p7 C om p8 C om p9 C om p1 0 C om p1 1 C om p1 2 En vi ro nm en ta l A pa th y pn M ul tif un ct io na l pn 0. 43 72 Pr od uc tiv ist pn C hi ld pn A ge pn U ni ve rs ity E du ca te d pn So ci al C la ss pn C ity pn To w n pn Fa rm in g Ba ck gr ou nd pn 0. 52 74 Im po rt an ce o f L an ds ca pe in c ho os in g w he re to li ve pn 0. 52 21 Sa tis fa ct io n of A re a pn 0. 51 58 Q ua lit y of S ur ro un di ng L an ds ca pe pn C on ce rn ed a bo ut th e en vi ro nm en t pn 0. 54 44 C ar e ab ou t C on se rv at io n pn En vi ro nm en ta l A pa th y ph M ul tif un ct io na l ph 0. 54 15 Pr od uc tiv ist ph C hi ld ph A ge ph U ni ve rs ity E du ca te d ph So ci al C la ss ph C ity ph 0. 53 63 To w n ph 0. 53 23 Fa rm in g Ba ck gr ou nd ph 0. 34 01 Im po rt an ce o f L an ds ca pe in c ho os in g w he re to li ve ph 0. 35 49 198 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis C or re la tio ns Ty pe C om p1 C om p2 C om p3 C om p4 C om p5 C om p6 C om p7 C om p8 C om p9 C om p1 0 C om p1 1 C om p1 2 Sa tis fa ct io n of A re a ph 0. 38 08 Q ua lit y of S ur ro un di ng L an ds ca pe ph C on ce rn ed a bo ut th e en vi ro nm en t ph 0. 34 69 C ar e ab ou t C on se rv at io n ph En vi ro nm en ta l A pa th y pa M ul tif un ct io na l pa 0. 43 15 Pr od uc tiv ist pa C hi ld pa A ge pa U ni ve rs ity E du ca te d pa So ci al C la ss pa C ity pa 0. 35 94 To w n pa 0. 33 58 0. 31 34 Fa rm in g Ba ck gr ou nd pa 0. 48 66 Im po rt an ce o f L an ds ca pe in c ho os in g w he re to li ve pa 0. 49 83 Sa tis fa ct io n of A re a pa 0. 49 35 Q ua lit y of S ur ro un di ng L an ds ca pe pa C on ce rn ed a bo ut th e en vi ro nm en t pa 0. 49 43 C ar e ab ou t C on se rv at io n pa En vi ro nm en ta l A pa th y nn M ul tif un ct io na l nn 0. 44 12 Pr od uc tiv ist nn C hi ld nn A ge nn 0. 44 26 U ni ve rs ity E du ca te d nn So ci al C la ss nn 0. 33 84 C ity nn To w n nn 0. 37 78 199Assessing preferences for rural landscapes C or re la tio ns Ty pe C om p1 C om p2 C om p3 C om p4 C om p5 C om p6 C om p7 C om p8 C om p9 C om p1 0 C om p1 1 C om p1 2 Fa rm in g Ba ck gr ou nd nn 0. 37 69 Im po rt an ce o f L an ds ca pe in c ho os in g w he re to li ve nn 0. 39 79 Sa tis fa ct io n of A re a nn 0. 41 21 Q ua lit y of S ur ro un di ng L an ds ca pe nn 0. 34 42 C on ce rn ed a bo ut th e en vi ro nm en t nn 0. 40 1 C ar e ab ou t C on se rv at io n nn En vi ro nm en ta l A pa th y nh -0 .3 38 1 M ul tif un ct io na l nh 0. 37 Pr od uc tiv ist nh C hi ld nh 0. 30 44 A ge nh 0. 38 86 U ni ve rs ity E du ca te d nh 0. 30 16 So ci al C la ss nh 0. 40 32 C ity nh 0. 46 38 To w n nh 0. 34 59 Fa rm in g Ba ck gr ou nd nh Im po rt an ce o f L an ds ca pe in c ho os in g w he re to li ve nh 0. 41 22 Sa tis fa ct io n of A re a nh Q ua lit y of S ur ro un di ng L an ds ca pe nh C on ce rn ed a bo ut th e en vi ro nm en t nh 0. 43 73 C ar e ab ou t C on se rv at io n nh 0. 39 08 En vi ro nm en ta l A pa th y na M ul tif un ct io na l na Pr od uc tiv ist na C hi ld na A ge na U ni ve rs ity E du ca te d na 200 Cathal O’Donoghue, Stephen Hynes, Paul Kilgarriff, Mary Ryan, Andreas Tsakiridis C or re la tio ns Ty pe C om p1 C om p2 C om p3 C om p4 C om p5 C om p6 C om p7 C om p8 C om p9 C om p1 0 C om p1 1 C om p1 2 So ci al C la ss na 0. 30 04 C ity na 0. 48 51 To w n na 0. 56 Fa rm in g Ba ck gr ou nd na Im po rt an ce o f L an ds ca pe in c ho os in g w he re to li ve na Sa tis fa ct io n of A re a na Q ua lit y of S ur ro un di ng L an ds ca pe na C on ce rn ed a bo ut th e en vi ro nm en t na 0. 30 44 C ar e ab ou t C on se rv at io n na N ot e: p n – na tu ra l a tt rib ut es ( po si tiv e si gn ); ph – h um an a tt rib ut es ( po si tiv e si gn ) ; p a – ag ric ul tu ra l a tt rib ut es ( po si tiv e si gn ); nn – n at ur al a tt rib ut es (n eg at iv e si gn ); nh – h um an a tt rib ut es (n eg at iv e si gn ); na – a gr ic ul tu ra l a tt rib ut es (n eg at iv e si gn ); Economics of culture and food in evolving agri-food systems and rural areas Severino Romano1, Francesco Vanni2, Davide Viaggi3 On the relevance of the Region-Of-Origin in consumers studies Fabio Gaetano Santeramo1, Emilia Lamonaca1,*, Domenico Carlucci2, Biagia De Devitiis1, Antonio Seccia1, Rosaria Viscecchia1, Gianluca Nardone1 Benefits for the local society attached to rural landscape: An analysis of residents’ perception of ecosystem services Stefano Targetti1, Meri Raggi2, Davide Viaggi1 Assessing preferences for rural landscapes: An attribute based choice modelling approach Cathal O’Donoghue1, Stephen Hynes1, Paul Kilgarriff2, Mary Ryan3, Andreas Tsakiridis3 Exploring governance mechanisms, collaborative processes and main challenges in short food supply chains: the case of Turkey Yaprak Kurtsal, Emel Karakaya Ayalp, Davide Viaggi