Social responsibility as a tool for destination competitiveness

Máster Universitario en Economía del Turismo, del Transporte y del Medio Ambiente Social Responsibility as a tool for Destination Competitiveness (Responsabilidad social como una herramienta para competitividad de los destinos) Presentado por Greta Skalisiute 23454608 Fdo.: Bajo la tutela de Jorge Araña Padilla Gran Canaria, 20/06/2014 TABLE OF CONTENTS TABLE OF CONTENTS ............................................................................................. 1 I. INTRODUCTION ................................................................................................. 1 II. THEORETHICAL FRAMEWORK ......................................................................... 3 2.1. What is Social Responsibility (SR) ................................................................ 3 2.2. SR measurement........................................................................................... 3 2.3. Dimensions of Social Responsibility .............................................................. 7 2.4. Destination Social Responsibility as a tool for Destination Competitiveness (National/ Regional Social Responsibility)............................................................... 9 III. METHODOLOGY ........................................................................................... 14 3.1. Using DCE to obtain monetary valuations of SR actions ............................... 14 3.2. Econometric Specification: Benefits and implications of departing from conditional logit model .......................................................................................... 17 3.3. The econometric model .................................................................................. 19 IV. DATA COLLECTION ...................................................................................... 24 V. RESULTS AND DISCUSSIONS ..................................................................... 26 REFERENCES ......................................................................................................... 29 TABLES .................................................................................................................... 34 Table 1. ................................................................................................................. 34 Table 3. ................................................................................................................. 35 Table 4. ................................................................................................................. 36 1 I. INTRODUCTION The overall number of international arrivals has grown an average of 4.1 % a year during the last ten years (UNWTO highlights, 2014). During the same period, the number of new competing destinations has been also growing at a fast pace. Therefore, one of the crucial challenges for modern tourist destinations is to identify and implement new competitive advantages (Crouch and Ritchie, 1999; Zhang, H., et al, 2011; Enright and Newton, 2004, 2005; Gomezelj and Mihalič, 2008). In this context, a growing number of Destination Managemnet Organizations (DMOs) and private tourist corporations have started to focus their management efforts in implementing different social responsibility‘s (SR) actions as a way to increase destinations competitiveness (Carroll, 1991; Mackey and Mackey, 2007). For instance, one of the largest tour operator TUI Travel PLC have decided to explicitly internalize the impacts of its operations on the environment and local communities by implementing, monitoring and managing a detailed accountability of SR actions (Coles et al, 2013). In this paper we propose the use of a Discrete Choice Experiment (DCE) in order to measure potential gains in destination competitiveness while undertaking Social Responsibility (SR) policies. Using DCE allows us to calculate consumers Willingness to pay (WTP) for implmenting different SR actions. There are at least three main advantages of using WTP as a measure of tourist preferences for SR. First, it provides a useful information for decision makers about how strong are the values that visitors attach to a SR policy. Therefore, it may be useful to inform pricing strategies for such products. Secondly, WTP measures can be employed as an important input in economic evaluations (e.g. cost benefit analysis), since it can be easily compared with the monetary costs of implementing social responsibility actions. Finally, WTP assessments are an appropriate instrument to make relative comparisons and rankings of the desirability of goods or services (Hole and Kolstad, 2012). 2 The results show that efficiently communicated destinations‘ SR actions may play a relevant role in tourists decision of where to travel. Thus, there is some room for DMOs to significantly improve their market shares by correctly adopting and efficiently communicating specific SR actions. In particular, it is estimated that on average tourist are willingness to pay a premium of 5% on top of current prices if the destinations they are travelling to performs particular SR policies. Visitors ranking of most valued SR dimensions were 1) labour conditions; 2) environmental conditions; 3) cultural activities for the community members. The rest fo the dissertation is organized as follows, section 2 reviews the existing literature on defining and measuring social responsibility at a corporation and at a destination level. Section 3 describes the DCE methodology employed to evaluate the proposed actions. Finally section 4 presents the main results and some discussion and further research. 3 II. THEORETHICAL FRAMEWORK 2.1. What is Social Responsibility (SR) According to McGehee et al (2009), “SR may be defined as an overall ethic or vision that implies the need for businesses to contribute back to the communities and markets that have made them successful“. Although the history of social responsibility and philanthrophy started from 17th century, when companies first faced the public pressure of the awareness for the impacts of corporations on the environment, the modern era of SR probably started with Bowen‘s publication of “Social Responsibilities of Businessman“ in 1953 (Inoue and Lee, 2011). Inoue and Lee (2011) uses SR as a term which collects various voluntary activities adopted in a company, such as advanced human resource management programs, the reduction of environmentally dangerous substances, philanthropic activities, the support of local businesses and the production of products integrating social attributes. Various industries hence are now adopting SR responsible activities in order to satisfy their customers‘ demand, which are concerned with SR issues. Coles et al (2013) goes along with this statement and emphasized the idea for responsible modes of production and consumption. However, taking into account firms‘ financial performance, the question therefore arises how the investment in socially responsible activities affects company. Moreover, Inoue and Lee (2011) asked “whether or not the companies, actively involved in SR initiatives, outperform the firms that do not demostrate the same degree of social involvement“. Correspondingly, the same authors literally answered these questions by proposing several ideas investigated by earlier scholars, which idetintified SR as a source of competitive advantages that positively affects various aspects of firms performance, such as reputation, consumer satisfaction, attractiveness of a firm as an employer and organizational commitment among employees. 2.2. SR measurement 4 In spite of the growing attention to SR issues in the tourism sector, there is still some controversy regarding how to define and measure SR actions and their influences on corporations goals. Martínez et al (2013, pp. 366) argue that all the methodologies proposed in the literature have serious limitations, and therefore further research is necessary to improve the quality of such measures. Maigan and Farrel (2000) classified existing SR measurement methods into three approaches: (a) expert evaluations, (b) single-issue and multiple-issue indicators and (c) surveys of management. The first category of empirical investigation to evaluate social responsibility is based on the information provided by the experts of the bussineses, society area or the industry. (Maigan and Farrel, 2000, pp. 285) Martínez et al (2013), proposed an extension of Maigan and Farrel (2000) classification. It does account for the following aspects: (1) reputation indices or databases; (2) single and multiple-issue indicators; (3) content analysis of corporate publications; (4) scales measuring SR at the individual level; (5) and the scales measuring SR at the organizational level (Turker, 2009; Martínez et al, 2013). However, in this study we want to maintain consistency with the use of KLD data employed in previous relevant works (Inoue and Lee, 2011), therefore we will employ the first aspect -reputation indices or databases-. Objective measures In order to assess which specific companies behave responsibly towards the environment and the society, many empirical studies have used Fortune index ratings, e.g., Berman (1999). This index is mainly based on the intensive use of experts’ opinions through rating scales of several SR actions in large corporations. However, since the Fortune index is a somewhat subjective measure based on experts’ evaluations, some researchers have claimed the superiority of objective indicators. For instance pollution control index (published by Council of Economic priorities) or corporate criminality (Maigan and Farrel, 2000, pp. 285). These approaches only use a single dimension to measure SR. As a consequence, some 5 scholars have proposed the use of multi-dimensional measures in their analysis (Maigan and Farrel, 2000). For example Backhaus et al (2002) collected data from 297 undergraduate bussiness students and also included the sample drawn from Fortune index. Likewise, given Inoue and Lee (2011) proposed method, SR measurement includes Clarkson (1995) approach, which states that SR can be better assessed by a stakeholder framework that evaluates the companies relationship with their primary stakeholders. Respectively, the latter stakeholder group includes shareholders/ owners, employees, suppliers, customers, and public stakeholders such as community or the natural environment. For example Maigan and Ferrell (2004) as well as Smith (2003) incorporated stakeholder theory to suggest that SR prescribes the responsibility of corporation to meet or exceed the norms of various stakeholders, which in turn dictates desirable organizational behaviours (McDonalds, 2006). Robson and Robson (1996) as well as Sautter et al (1999) distinguished the primary stakeholder group, which included employees, tourists, local business suppliers, community residents, government and the environment. Consistent with ´stakeholder theory´ concept, and assuming it as an important part of objective measure, some of the recent studies have examined the influence of separate SR dimensions on different sectors. (Table 1) One of them was done by Berman (1999), who measured the link between each of the KLD categories and accounting-based financial performance on a stakeholder context. Since authors of the study implemented two implicit models, the obtained results were significant only from the first’s model’s perspective and showed that only two of five variables – employee relations and product quality, strongly affects company’s financial performance and hence improves profitability. However community, diversity, and the environmental dimensions showed insignificant level of the impact on firm financial performance. In contrast, Backhaus et al (2002) showed, that differently than in previously reviewed study, the environmental issues, diversity and community relations dimensions are 6 highly significant to potential employee group. Based on the ‘Stakeholder theory’, ‘Signaling Theory’ and ‘Social Identity Theory’, authors measured the attractiveness of CSP’s (corporate social performance’s) influence for a job seekers’ which in turn has been recognized as having a high influence on a firm’s image, which is highly valuated between potential job candidates. Ultimately, Backhaus et al (2002) demonstrated, that firm’s involvement into SR activities improves overall corporate’s image not only from societal-concerned people perspective, but also for potential job seeker’s group. Hillman and Keim (2001) tested the relationship between shareholder value, stakeholder management and social issue participation. Authors demostrated that only one dimension – community relations, have a positive relationship to financial performance, while other dimensions were found as having insignificant or negative impact on corporate‘s financial performance. The explanation of such findings is simple – the use of a firm‘s financial resources always has an opportunity cost. (Hillman and Keim, 2001, pp.136) By implementing various philantrophic and charitable strategies into firm‘s activities, the cost of forgone opportunities increase. However, even if the SR actions requires a high cost, the implementation also requires a long-term vision. Thus, an expected feedback could be even higher not only from the financial perspective, but also from society‘s and environmental side, like higher stakeholder attention. For example, Kacperczyk (2009) showed that corporate attention to the environment, community and minorities dimensions influenced long-term shareholder value, since author tested the impact of corporate attention to non-shareholding stakeholders by shifting the power from shareholders to managers. Therefore, it is obvious that SR implementation into firm‘s activities demonstrates not only corporate‘s attention to environmental and societal issues but also has some profitable advantages. Given the diversity dimension, growing attention of diversity issues has also been acknowledged in tourism literature. For example, Klemm (2002) tested tourism participation of ethnic minority groups in Britain and partly approved the fact that different ethnic groups want different types of holidays because of their race and culture. In addition, Pritchard and Morgan examined gendering issues in tourism promotion (2000) and demonstrated, that according to some places, tourism promotion 7 privileges the male and heterosexual gaze above all others (Pritchard and Morgan , 2000, pp. 899). Thus, it could be stated that such finding suggest to create a new tourism products, considering to diversity issues. Finally, based on Turker’s (2009) suggested reputation indices, I propose that in order to objectively evaluate social activities in corporations, the measurement therefore combines reputation indices and databases (Fortune Index, Canadian Social Investment Database (CSID) and KLD Database) based on stakeholder theory as well as the survey dedicated to a certain group of respondents. Also, since we are interested in Inoue and Lee (2011) suggested five SR dimensions, KLD database corresponds to the concept of our research. Thus, in order to get acquainted to SR attributes, the next chapter will indroduce SR dimensions as well as SR division proposals suggested by other authors. Subjective measures: Ad hoc Surveys The third approach consists of surveying a certain group of community in order to measure SR. For example Klemm (2002) holded the survey of 80 Bradford citizens of Asian origin in order to find out their holiday preferences and differences between British population. In addition, Backhaus et al (2002) in their suvey tried to measure whether firm’s social performance influence potential employees perception of organizational attractiveness. According to Martínez et al (2013) the most surveys are mainly focused on the perception of SR activities, but not on corporate behaviours. (Martínez et al, 2013, pp. 369) Thus, it could be stated that in order to objectively operationalize SR measurement, it is necessary to combine more than one methods of SR measurement. 2.3. Dimensions of Social Responsibility While the most part of previous studies used one-dimensional measure that aggregates SR activities, some scholars suggest that SR consists of distinct 14  Thirdly, SR promotes higher degree of corporate concern for stakeholder group, which in turn have an ability to improve not only the relations with them but also and firm’s financial performance. (Berman, 1999)  Fourthly, company with implemented SR actions in its activities is determined as being more attractive to potential employees. (Backhaus et al, 2002)  Fifth, even if SR implementation on company‘s actions have a high cost, SR requires a long-term vision, which needs to be developed in a long term.  Finally, multinational companies are capable of making significant contributions through SR actions in developing countries. (Ite, 2004) III. METHODOLOGY 3.1. Using DCE to obtain monetary valuations of SR actions In order to get the desired results, in this section we propose discrete choice experiments (DCE) as a tool to be used to predict the impact of Tourist Destination Social Responsibility Actions on tourist demand. Discrete Choice Experiments (DCE) has been widely used to examine consumer decisions and preferences in various fields, such as marketing, transportation, environmental and health economics. According to Hoek and Gendall (2008), DCE examine attributes in the context of other product characteristics, thus it helps to create more realistic choice situation. In tourism research DCE has been widely utilized by investigators in order to test consumer´s behavior in different situations in a particular destination context. (Araña and León, 2008, 2013; Fieldman and Vasquez-Parraga, 2013; Kelly et all, 2006; Unbehaun et all, 2008; Crouch et all, 2007; Crouch et all, 2008) In order to evaluate SR influence on tourists’ destination choice, we implement DCE with the purpose to obtain tourist’s willingness to pay (WTP) for SR activities in tourism sector. In this study we refer to Lancaster’s proposed approach to Consumers Theory which states that consumption is an activity where goods are considered as inputs, and the output is a collection of different characteristics (Lancaster, 1966, pp. 133). We assume this approach as an appropriate division, since we are interested in the tradeoffs between the multiply characteristics. Respectively, SR dimensions are considered 15 as multiple characteristics of the tourist product, therefore by statistically designed different configurations of tourist products we estimate the contribution of SR actions. Since we have price as a characteristic, the trade-off between price and SR actions will give us an information about WTP for SR actions in the tourism sector and hence the results are expected to allow us to preliminary evaluate destination´s competitiveness under SR activities, and consequently to demonstrate the SR importance as an international competitiveness tool among tourist destinations. The measurement of SR and the design of the survey were taken from the precursor study by Araña and León (2014) which in turn allowed us to identify the key aspects of SR which were the most valuable to consumers. The surveys described several scenarios, which included different levels of attributes, including a monetary aspect (price), and different level of involvement into SR actions. The survey presented a series of questions, where each question included two or more alternatives, from which respondents had to chose their preferred option. Kragt and Llewellyn (2014) stated that theoretical basis of choice experiments lies in ´random utility theory´ and aforementioned Lancaster´s characteristics theory of value´ (Kragt and Llewellyn, 2014). Given Lancaster´s proposition that any good consists of different attributes, the random utility model describes the utility 𝑈ijt that individual i gets from the choice j in situation t. Accordingly, the component Vijt is described as a ´systematic´ utility inherent of utility and is assumed as a linear systematic component. The variable 𝜀ijt is considered as a random unobserved error term. The equation therefore is defined as: 𝑈ijt = Vijt + 𝜀ijt (1) Under the utility function, the systematic component of utility (2) Vijt is expressed by the importance of functional attributes for visitors 𝛽´i followed by an explanatory variable x which includes functional attribute´s options. Next component of equation 𝛼 is considered as a vector of importance of SR attributes for visitor followed by Qijt which describes SR alternatives. Vijt= 𝛽´ixijt+ 𝛼´Qijt (2) 16 Considering our proposition to obtain the information about consumers WTP for Social Responsibility actions, after estimating the contribution of SR actions on tourist´s destination choices, we consider Qj as a vector of characteristics of SR alternatives, and qn as a particular SR dimension. SR attributes as well as functional attributes are represented in a table 2. Qj = (q1, q2,… q5) (3) Table 2. X Functional attributes 1. Labour conditions 2. Environmental issues 3. Community relations Q SR attributes 1. Price 2. Lenght of the stay Thus, we state that an individual ´i´ would choose alternative ´j´ in situation ´t´ if and only if: Uijt> Uikt ∀k≠j (4) The probability of this event can be represented as: Prob (choosing j) = Prob(Uijt> Uikt) (5) Prob [(𝛽´xijt + 𝛼´Qijt+ 𝜀ijt) > (𝛽´xikt + 𝛼´Qikt + 𝜀ikt)] (6) Prob [𝛽´(xijt - xikt) + 𝛼´(Qijt - Qikt) > (𝜀ijt - 𝜀ikt)] (7) At this point we need to specify the probability distribution function (8), which describes our focus on latent class model for scenarios where information is available on several covariates in order to obtain one dimensional response variable: F (𝜀ijt - 𝜀ikt) (8) 17 Thus, we let probability of belonging to a particular class depend on subject-specific variables through a multinomial logic model. The specification of econometric logic model is presented in the next section. 3.2. Econometric Specification: Benefits and implications of departing from conditional logit model Conventional DCE are based on conditional logit econometric specifications of function F(.) on equation (8), (McFadden, 1974). However, during the last twenty years or so there is a growing literature aimed at exploring more flexible functional forms for F(.) (see Keane and Wasi, 2013; or Hole and Kolstad, 2012; for recent reviews). These flexible econometric specifications try to allow DCEs to accommodate for preferences heterogeneity. Probably the explosion of these models come from Train and McFadden (2001), when they show that any utility maximization problem involved in responding to DCE’s can be approximated by an appropriately specified mixed logit model 1 . However, Balcombe et al (2009) raised several questions related to the economic interpretation of mixed logic models, which should be answered in order to understand the operation of mixed logic model. Which coefficients in the model should we assume to be fixed or randomly distributed? The first question is connected to the choice of which coefficients in the model should be considered as fixed and which should be randomly distributed. According to Balcombe et al (2009), this division of utility coefficient is the main source of instability for the WTP measurement. According to the same authors, fixing parameters is necessary if payment coefficients are random, because in this case the moments of WTP ratio may not exist. Respectively, if the parameters in such situation are fixed, the estimates of WTP could be expected to stabilize. Which distributions should be employed? 1 Actually it can be shown that all the proposed models (latent class models, GMNL, …) are just particular specifications of the mixed function within the mixed logic models. 18 The second question raised by Balcombe et al (2009) encompasses the problem of selecting statistical distributions when including random parameters. With basis on Hole and Kolstad (2012) the distribution of preferences follows a particular distribution, e.g. a normal distribution. In addition, since there exist a number of possible distributions, the classical methods are found to better handle bounded distributions, while Bayesian methods perform better with ‘transformation of the normal’ distribution (Balcombe et al, 2009). With accordance to the same authors, since the most popular distribution was recognized the ‘normal’ one, it implies the idea that there will always be some respondents who will have extremely positive or negative values in respect of various attributes. As a result, this may lead to inappropriate assumption to impose on the model (Balcombe et al, 2009). Hole and Kolstad (2012), support this idea by saying that “it is unreasonable to assume that all individuals have the same marginal utility of income, so this implies an undesirable trade-off between reality and modelling convenience.” In addition, from the alternative – log-normal distribution point of view, the preferences for income are considered to be heterogeneous, but WTP distribution can be highly distorted, and therefore may also produce unrealistic estimates of the WTP means. Should we adopt a WTP space or a preference space? As a solution to the latter problem, Hole and Kolstad (2012) suggest to estimate the mixed logit model in WTP space rather than in preference space. Likewise, the third question raised by Balcombe et al (2009) links to the dilemma if we should estimate WTP in preference space or in WTP space. With reference to Balcombe et al (2009), in WTP space distributions of the marginal rates of substitution are estimated directly, thus this leads to production of more stable WTP estimates. It is also essential to mention that the instability of the WTP estimates in preference space is especially strong “where the parameter of the payment attribute is variable and is not bounded above zero” (Balcombe et al, 2009). How should we compare alternative specifications? 19 Finally, the fourth question asks on what basis the model comparison should be substantiated and with respect to Balcombe et al (2009), there is no theoretical basis on which the comparison of the models should be based on. As it was observed by Kamakura and Wedel, the distribution of preference coefficients need to be considered by investigator and therefore the distribution, which most accurately represents the data becomes an empirical issued (Balcombe et al, 2009). 3.3. The econometric model This section describes most of the different econometric specification that have been proposed in the discrete choice literature, and is mainly borrowed from Balcombe et al (2009). General model specification The utility that the jth individual receives from the sth choice in the nth choice set is assumed to be of the form Uj,s,n = x’j,s,ng( β j) + es,j,n (1) where x’j,s,n denotes the k × 1 vector of attributes presented to the jth individual (j = 1, …. , J) in the sth option (s = 1, …., S) of the nth choice set (n = 1, …., N). yj,s,n denotes an indicator variable that equals 1 if the jth individual indicates that they would choose the sth option within the nth choice set, and 0 if they would not. βj is a (k × 1) vector describing the preferences of the jth individual and g( ∙ ) is transformation of the parameters from and to the space of k vectors. The erros es,j,n is ‘extreme value' (Gumbel) distributed, is independent of x’s,j,n, and is uncorrelated across individuals or across choices. Without loss of generality, the parameters β j are ordered so that they may contain fixed parameters cj in the first block, and random parameters b’j in the second β ’j = (c’j, b’j) (2) Both cj and bj can be conditioned on variables describing the characteristics of the jth individual. Preferences may therefore be determined by a vector zj, a (h × 1) column vector describing the characteristics of the jth individual (h being 1 and z’j being 1, for 20 all j, if there are no characteristics). Therefore, defining Zj = Ik , the components of β ’j are defined as Cj = Z’j α c bj = Z’j α b + uj (3) and uj is a independently and identically normally distributed vector with variance covariance matrix Ω. The error {uj} are assumed to be uncorrelated across individuals. The function g(∙) may take many forms (Train and Sonnier, 2005). In considering estimation in WTP space, we also use reparameterisations of the form g( β j) = g1( β 1j)(1, g2( β 2j),….,gk( β kj))’ (4) in which case the quantities g2( β 2j),….,gk( β kj) are marginal rates of substitution with the numeraire element of attribute vector (the first element in the case above). If this type of transformation is used then we say that estimation is taking place in WTP space. Otherwise, estimation is being performed in ‘preference space’. The set of all stated choices by respondents is Y = {yj,s,n}j,s,n. The set of characteristics describing all respondents is Z = {zj}j. The set of options given to the jth individual is xj = {xj,s,n}s,n and the set of all options sets given to all respondents is X = {Xj}j.The data D are, therefore, the collection D = {Y, Z, X}. Faced with a set of choices, the jth individual will prefer x𝑠𝑘n if Uj,𝑠𝑘,n > Uj,𝑠𝑞,n for all k not equal to q. Each respondent has a probability (π) of misreporting, along with a probability (𝜆s) that misreporting (should it occur) will be in favour of option s (where ∑𝜆s = 1). The parameters related to misreporting are denoted as Λ = (π, 𝜆1, …., 𝜆S-1). The collection of all parameters describing the model are denoted as Θ = (𝛼, Ω, Λ) and the ser {bj}j will be denoted as B the ‘latent data’. Finally, for convenience, the definite multiple integral ∫𝛽𝑛…∫𝛽1db1 … dbn is expressed as ∫B dB. Priors Bayesian estimation requires priors for α and Ω and Λ. For α these are specifies as being normally distributed with mean μ, and variance𝐴0: (α’c, α’b)’ = α ~ fN(α|μ, 𝐴0) (5) where A0 is a diagonal matrix. If there are fixed and random elements, the associated means for αc and αb, respectively, are μc and μb. Likewise, A0 contains the diagonal 21 blocks A0,b and A0,c. The prior for the covariance matrix of the random parameters is ditributed inverse Wishart with parameters 𝑇0 and 𝑉0 Ω~𝑓IW(Ω|𝑇0, 𝑉0) The ‘hyper parameters‘ μ , 𝐴0,𝑇0,𝑣0 are set a priori. The misreporting parameters are assumed to have a uniform prior, subject to inequality constraints π~𝑓𝑈(0,1) (6) (𝜆1, ...., λs-1)~ 𝑓𝑈(0,1)s-1 × I (∑𝜆 𝑆−1 𝑆=1 ≤ 1) (7) where (∑𝜆 𝑆−1 𝑆=1 ≤ 1) is equal 1 if the constraint is obeyed and zero otherwise. The integrating constant of this distribution is 1/(S – 1)! Together this set of priors is denoted as P(Θ), and the priors on Λ only, as P (Λ). Full-data likelihood, the likelihood and marginal likelihood The full-data (or complete) likelihood function is the likelihood expresses in terms of the parameters and latent data (B). For the ML with misreporting, the full-data likelihood is Lf (B,Θ,D) =∏( 𝑗∏ ∏ 𝑝𝑗,𝑠,𝑛 𝑌𝑗,𝑠,𝑛 𝑛𝑠 )f(B|Ω, αb, Z) (8) Integrating out the latent data gives the likelihood L(Θ,D)= ∫B ∏( 𝑗∏ ∏ 𝑝𝑗,𝑠,𝑛 𝑌𝑗,𝑠,𝑛 𝑛𝑠 ) dF(B|Ω, αb, Z) (9) In the absence of latent data we could write Lf({},Θ, D) = L(Θ, D). L(Θ, D) is likelihood of the model. It is this quantity that is maximised in classical estimation. The marginal likelihood, given priors on the parameters P(Θ), is ℳ(D)= ∫Θ L(Θ,D)P(Θ)dΘ (10) The larger marginal likelihood indicates greater support for a particular model. Model estimation As for the model estimation we also follow the procedure and recommendations proposed by Balcombe et al (2009). It follows that the poseriors of αb and Ω are known given values of αc and Λ.∂ When some of the coefficients are fixed or the model contains misreporting probabilities, M-H steps are required to map the posterior distributions of αc and Λ. Alternatively, if all parameters are fixed, a M-H algorithm can be employed to estimate the model. Investigations into the performance of estimation 22 algorithms revealed that the rates of convergence depended on the types of transformations g(βj) that were used and whether misreporting was introduced, because of the extra steps required to compute the model. For the analysis, the posterior distributions are mapped using 10,000 draws from the posterior sampler. Convergence of the sampler is monitored in several ways. First, visual plots of the sampled values are produced as sampler. Covergence of the sampler is monitored in several ways. First, visual plots of sampled values are produced as the sampler runs for the sequences of α, Ω, and λ. Second, a degree of dependence of the sampled values is examined by estimating the autocorrelation coefficients of the sequential values of the sampler. The ‘skip‘ (only very ‘skipth‘ iteration is recorded) was then set so as to allow a lesser degree of dependence should autocorrelation be too high. Third, a modified t-test for hypothesis of ‘no-difference‘ between the first and second half of the sampled values (with subset eliminated from the midle) was conducted on the sequence of α parameters. This used an estimate of long-run covariance matrix (the spectral density matrix of the sequence at frequency zero) provided by the spectral kernel methods. Choice of priors We refined our priors using Monte-Carlo trials. These indicated that setting T0 = 𝜈 0. Ik inflated estimates of the covariance matrices Ω, generated by the sampler and inflated values of α also. This tendency depended on the values of Ω, used to generate the data, the number of attributes, the sample size and the number of choice sets given to each respondent. When all parameters were random, setting T0 = ν0I/10 and 𝜈 0 = k(k + 1)/2 being the number of free elements in the covariance matrices Ω and were dominated by the data in cases where the elements of Ω were larger. Our experiments also indicated that these priors gave similar, but slightly better results, to simply using ν0 = k + 3 or k + 4. Proper priors are used for α because marginal likelihood values cannot be computed without them. Non-informative priors can be obtained by setting the diagonal elements of A0 to very large values. However, the priors employed here are set more 23 informatively, with A0 = 10Ik and μ = 0. In a standard linear regression framework these priors would, in most cases, be considered highly informative. But, given the attribute values emplyed in this study these priors are only weakly informative, although they can sometimes substantially improve the performance of the sampler. This is particularly so in specifications that attempt to truncate the distributions of bj since the whole of the distribution can become massed at a point of truncation, and α can become non-identified. In such circumstances, an informative prior can prevent this parameter wandering into non-identified regions idefinitely. Calculating the marginal likelihood The ratio of the marginal likelihoods gives the posterior odds for the two models (measuring the relative support for these models) given that the prior odds are even. Marginal likelihood calculations can be practically problematic in cases where the parameter space has many dimensions. Using M to denote different models, denote ℳD, M) as an estimate of the marginal likelihood for the Mth model The method of Gelfand and Dey (GD) (1998) estimtes the marginal likelihood using In ℳ(D,M) = -ln [𝐺−1 ∑𝜓(Θ𝑖) 𝑃(Θ𝑖)𝐿(Θ𝑖,𝑫,𝑀) 𝐺 𝑖=1 ] (11) or alternatively ℳ(D,M) = -ln [𝐺−1 ∑𝜓(Θ𝑖,𝑩𝑖) 𝑃(𝐁i|Θ𝑖)𝑃(Θ𝑖)𝐿(𝑩𝑖,Θ𝑖,𝑫,𝑀) 𝐺 𝑖=1 ] (12) where Θ i and Bs,i are draws from their posterior distributions. The ‘tuning functions‘ ψ(Θi) or ψ(Θi,Bi) are densitiwa with tails that are sufficiently thin so the fractions within the expressions (11) or (12) are bounded from above. Alternatively, the tuning functions can be set equal to the priors, in which case expressions (11) and (12) collapse to harmonic means which are known to be unstable (Raftery, 2006). The second estimate (12) is the easier to calculate, for a given choice of ψ(Θi,Bi). It does not require recording draws of B, which would be memory intensive, providing ψ(Θi,Bi) and P(Bi|Θi) are recorded when running the sampler. However, GD method tends to give poor estimates in high dimensional problems (Raftery, 2006). As B contains up to J × k elements, our approach is to use (11) in preference (12) in order to mitigate the negative impacts of this dimensionality. In calculating (11), L( Θ , D, M) can be simulated by making successive draws of Bt (t = 1, ...., T) from f(B|α, Ω, Z) and calculating the likelihood as outlined by (Train, 30 Coles, T., et al, (2013). ‘Tourism and corporate social responsibility: A critical review and research agenda’, Tourism Management Perspectives, Vol.6, UK, pp. 122141 Crouch, G., I., Brent Ritchie, J., R., (1999). ‘Tourism, Competitiveness, and Societal Prosperity’, Journal of Business Research, Vol. 44, pp. 137-152 Crouch, G., I., et al. (2007). ‘Discretionary Expenditure and Tourism Consumption: Insights from a Choice Experiment’, Journal of Travel Research, Vol. 45, pp. 247-258 Crouch, G., I., et al., (2009). ‘Modelling consumer choice behavior in space tourism’, Tourism Management, Vol. 30, pp. 441-454 Crouch, G., I., Ritchie, B., J., R., (1999). ‘Tourism Competitiveness, and Societal Prosperity’, Journal of Bussiness Research, vol 44, pp.137-152 Dodds, R., Kuehnel, J., (2010). ‘CSR among Canadian mass tour operators: good awareness but little action’, International Journal of Contemporary Hospitality Management, Vol. 22, No. 2, pp. 221-244, Canada Enright, M., J., Newton, J., (2004). ‘Tourism destination competitiveness: a quantitative approach’, Tourism Management, Vol. 25, pp. 777-788 Enright, M., J., Newton, J., (2005). ‘Determinants of Tourism Destination Competitiveness in Asia Pacific: Comprehensiveness and Universality’,Journal of Travel research, Vol.43, pp.339-350 Feldman, P., M., Vasquez-Parraga, A., Z., (2013). ‘Consumer social responses to CSR initiatives versus corporate abilities’, Journal of Consumer Marketing, Vol. 30, No. 2, pp. 100-111 Font, X., et al., (2012). ‘Corporate social responsibility: The disclosureperformance gap’, Tourism management, Vol. 33, pp. 1544-1553 Gelfand, A., E., Dey, D., K. (1998). ‘Bayesian Model Choice: Asymptotics and exact calculations’, Stanford University, Department of statistics, pp. 1-18 Gjølberg, M., (2009). ‘Measuring the immeasurable? Constructing an index of CSR practices and CSR performance in 20 countries’, Scandinavian Journal of Management, Vol. 25, pp. 10-22 Gomezelj, D., O., Mihalič, T., (2008).’Destination competitiveness – Applying different models, the case of Slovenia’, Tourism management, Vol.29, pp. 294-307 31 Golja, T., Nižić, M., K., (2010). ‘Corporate Social Responsibility in Tourism – the most popular tourism destinations in Croatia: comparative analysis’, Management, Vol. 15, pp. 107-121 Henderson, J., C., (2007). ‘Corporate social responsibility and tourism: Hotel companies in Phuket, Thailand, after the Indian Ocean tsunami’, Hospitality management, Vol. 26, pp. 228-239 Hillman, A., J., Keim, G., D., ‘Shareholder value, stakeholder management, and social issues: What’s the bottom line?’, Strategic Management Journal, Vol. 22, No. 2, 2001, pp. 125-139 Holden, A., ‘In need of new environment ethics for tourism?’, Annals of Tourism Research, Vol. 30, No. 1, 2003, pp. 94-108 Hsee, C., and Leclerc, F. (1998), ‘Will products look more attractive when presented separately or together?’, Journal of Consumer Research, Vol 25, pp 175– 185. Inoue, Y., Lee, S., (2011). ‘Effects of different dimensions of corporate social responsibility on corporate financial performance in tourism-related industries’, Tourism Management, Vol. 32, pp. 790-804 Ite, U., E., (2004). ‘Multinationals and Corporate Social Responsibility in Developing Countries: A case study of Nigeria’, Wiley InterScience (www.interscience.wiley.com), pp. 1-11 Kacperczyk, A., ‘With greater power comes greater responsibility? Takeover protection and corporate attention to stakeholders’, Strategic Management Journal, Vol. 30, 2009, pp. 261-285 Klemm, M., S., ‘Tourism and Ethnic Minorities in Bradford: The Invisible Segment’, Vol. 41, 2002, pp. 85-91 Kelly, J., et al., (2007). ‘Stated preferences of tourists for eco-efficient destination planning options’, Vol. 28, pp. 377-390 Koop, G., (2003). ‘Bayesian Econometrics’, Wiley Publishers, England Kragt, M., E., Llewellyn, R., S., (2014). ‘Using a Choice Experiment to Improve Decision Support Tool Design’, Applied Economic Perspectives and Policy, pp. 1-21 Lancaster, K., J., (1966). ‘A NEW APPROACH TO CONSUMER THEORY’, The Journal of Political Economy, Vol. 74, No. 2, pp. 132-157 32 Laufer, W., S., (2003). ‘Social Accountability and Corporate Greenwashing’, Journal of Business Ethics, Vol.43, No. 3, pp. 253-261 Li, H., Maddala, G., S., (1997). ‘Bootstrapping cointegrating regressions’, Journal of Econometrics, pp. 297-318 Mackey, A., Mackey T., B., (2007). ‘Corporate Social Responsibility and Firm Performance: Investor Preferences and Corporate Strategies’, Vol. 32, No.3, pp. 817835 Maigan, I., Ferrell, O., C., (2000). ‘Measuring Corporate Citizenship in two Countries: The Case of the United States and France’, Journal of Business Ethics, Vol. 23, pp. 283-297 Martínez, P., et al. (2013). ‘Measuring Corporate Social Responsibility in tourism: Development and validation of an efficient measurement scale in the hospitality industry’, Journal of Travel & Tourism Marketing, 30:4, pp. 365-385 Mathies, C., et al., (2013). ‘The Effects of Customer-Centric Marketing and Revenue Management on Travelers’ Choices’, Journal of Travel Research, Vol. 52, No. 4, pp. 479-493 McDonald, L., M., ‘Use of different corporate social responsibility (CSR) Initiatives as a crisis mitigation strategy’, Academy of World Business, Marketing & Management Development Conference Proceedings, Vol. 2, No. 118, 2006, pp. 1365-1375 McDonald, L., M., Rundle-Thiele, S., ‘Corporate social responsibility and bank customer satisfaction’, Vol. 26, No. 3, 2008, pp. 170-182 Nicolau, J., L., (2008). ‘Corporate Social Responsibility. Worth-Creating Activities’, Annals of Tourism Research, Vol. 35, No. 4, pp. 990-1006 Su, X., et al., (2013). ‘Profit, Responsibility, ans The Moral Economy of Tourism’, Annals of Tourism Research, Vol. 43, pp. 231-250 Nunkoo, R., (2012). ‘Public Trusr in Tourism Institutions’, Annals of Tourism Research, Vol. 39, No. 3, pp. 1538-1564 Perez, A., Rodriguez del Bosque, I., (2013). ‘Measuring CSR Image: Three Studies to Develop and to Validate a Reliable Measurement Tool’, Journal of Business Ethics, Vol. 118, pp. 265-286 Pritchard, A., Morgan, N., J., (2000). ‘PRIVILEGING THE MALE GAZE: Gendering Tourism Landscapes’, Annals of Tourism Research, Vol. 27, No. 4, pp. 884905 33 Raftery, A., E., (1996). ‘Hypothesis testing and model selection. In Markov Chain Monte Carlo in Practice (W.R. Gilks, D.J. Spiegelhalter and S. Richardson, eds.)’, London: Chapman and Hall, pp. 163--188 Robson, J., Robson, I., (1996). ‘From shareholders to stakeholders: critical issues for tourism marketers’, Tourism Management, Vol. 17, No. 7, pp. 533-540 Sautter, E., T., Leisen, B., (1999). ‘MANAGING STAKEHOLDERS: A Tourism Planning Model’, Annals of Tourism Research, Vol. 26, No. 2, pp. 312-328 Swimberghe K., R., Wooldridge, B., R., (2014). ‘Drivers of Customer Relationships in Quick-Service Restaurants: The Role of Corporate Social Responsibility’, Cornell Hospitality Quarterly, pp. 1-11 Train, K., E., (2003). ‘Discrete Choice Methods and Simulation’, Cambridge University Press, pp. 1-383 Train, K., Sonnier, G., (2004). ‘Mixed Logit with Bounded Distributions of Correlated Partworths*’, University of California, Los Angeles, pp. 1-23 Turker, D., (2009). ‘Measuring Corporate Social Responsibility: A Scale Development Study’, Journal of Business Ethics, Vol. 85, No. 4, pp. 411-427 Unbehaun, W., et al., (2008). ‘Trends in winter sport tourism: challenges for the future’, Tourism Review, Vol. 63, No. 1, pp. 36-47 UNWTO Tourism Highlights, 2014 Edition. (accessed on June, 10th of 2014 at http://mkt.unwto.org/publication/unwto-tourism-highlights-2014-edition) Weber, M., (2008). ‘The business case for corporate social responsibility: A company-level measurement approach for CSR’, European Management Journal, Vol. 26, pp. 247-261 Zhang, H., et al, (2011). ‘The evaluation of tourism destination competitiveness by TOPSIS & information entropy – A case in the Yangtze River Delta of China’, Tourism Management, Vol.32, pp. 443-451 34 TABLES Table 1. Title Dimensions Emplyee relations Product quality Community relations Environmenta l issues Diversity issues Methology Sector Sautter and Leisen (1999) Significant Significant Significant Significant Significant Theoretical Framework Tourism Robson and Robson (1996) Significant Significant Significant Significant Significant Theoretical Framework Tourism Berman, Brett and Karl (1999) Significant Significant Insignificant Insignificant Insignificant Comparison of 'Strategic stakeholder management model' and 'Stakeholder commitment model' Tourism + others (Combined) Backhaus et al (2002) Insignificant Insignificant Significant Significant Significant Signaling Theory, Social Identity Theory Tourism + others (Combined) Hillman and Keim (2001) Insignificant Insignificant Significant Insignificant Insignificant Market ValueAdded (MVA), SM (Stakeholder Management), SIP (Social Issue Participation) Tourism + others (Combined) Kacperczyk (2009) Insignificant Insignificant Significant Significant Significant Shareholder model, Stakeholder model, Difference-indifference (DD) method Tourism + others (Combined) Byrd, Bosley and Dronberger (2009) Insignificant Insignificant Significant Significant Insignificant Liker-style questions, ANOVA test Tourism Holden (2003) Insignificant Insignificant Insignificant Significant Insignificant Methodologic al framework Tourism Klemm (2002) Insignificant Insignificant Insignificant Insignificant Significant Personal survey (80 respondents) Tourism Pritchard and Morgan (2000) Insignificant Insignificant Insignificant Insignificant Significant Theoretical Framework Tourism 35 Table 3. Model number Attribute coefficients fixed or random WTP space %CR Model rank ML SE Price Env Cult Anim Labour 1 F F F F F ∙ 20 -1052,31 0,003 2 F F F F F ∙ 63.1 19 -1201,57 0,231 3 N N N N N ∙ . 12 -909,90 0,601 4 N N N N N ∙ 97.7 11 -907,72 0,516 5 L N N N N ∙ . 7 -904,36 0,629 6 L N N N N ∙ 93.2 8 -904,79 0,714 7 L N N N N √ 1 -895,9 0,191 8 L N N N N √ 95.2 5 -901,75 0,731 9 F N N N N ∙ 16 -926,91 0,763 10 F N N N N ∙ 97.1 15 -925,32 0,942 11 L N N N ∙ √ . 18 -1089,42 0,071 12 L N N N ∙ √ 66.9 17 -958,4 0,156 13 T N N N N ∙ . 13 -900,36 0,93 14 T N N N N ∙ 97.4 14 -901,42 0,75 15 T N N N N √ . 9 -885,56 0,618 16 T N N N N √ 96.8 10 -907,32 0,63 17 L C C C C ∙ 3 -900,53 0,25 18 L C C C C ∙ 97.3 6 -903,21 0,43 19 L C C C C √ 2 -898,33 0,75 20 L C C C C √ 97.5 4 -891,66 0,69 Notes: CR, percentage correctly reporting; ML, logged marginal likelihood; SE, standard error on the estimated marginal likelihood. F, fixed; N, normal; L, lognormal; T, triangular; C, censored normal (negatives massed on zero). 1 Table 4. Mean WTP per visitor (euros) for each SR actions proposed SR dimension Single format WTP (DCCV)* Joint format WTP (DCE) Labour 12.8 15.9 Environmental 16.7 23.7 Social projects 11.4 4.2 * Single format WTP comes from Araña and león (2014).