The effects of motivations to go to the country on rural accommodation choice: A hybrid discrete choice model

Abstract

Based on the theory of constructive consumer choice process, we propose that the rural accommodation choice process depends on motivationals of tourists to go to the country. Discrete choice models have frequently been used to explain and predict choices from a set of finite alternatives, such as the choice of accommodation, but using only cognitive attributes as explanatory variables. The hybrid discrete choice (HDC) model also allows us to take into account unobservable or latent variables, like the motivations, and incorporate them through a multiple indicator multiple cause (MIMIC) model. Data collected in Murcia (Spain) from a stated choice survey are used to estimate a multinomial logit model and two specifications of the HDC model. Our results find that motivations affect the probability of accommodation rural choice. Furthermore, the effect of the motivations is different depending on the attributes of the accommodation.

Keywords

hybrid discrete choice models latent variables rural house choice rural tourism stated preference structural equation modelling

Introduction

The accommodation choice is one of the most important decisions of tourists (Chen et al., 2017; Sharpley, 2000) and a major component of their tourist expenditure (Laesser and Crouch, 2006; Masiero et al., 2015). Hence, understanding accommodation choice is a key element in tourism demand modelling (Alegre and Pou, 2016) and an important help to accommodation managers and owners in relation to an effective management and suitable investment decisions (Kim and Park, 2017).

Several studies have been conducted to evaluate the factors affecting the tourism accommodation choice, in particular for hotels. Attributes such as location, price, facilities, size of guest rooms, staff, cleanliness, silence or air conditioning have been identified as having a strong influence on the choice of hotel (Chen et al., 2017; Lockyer, 2005; Masiero et al., 2015; Merlo and Joao, 2011; Stringam et al., 2010). Most of these attributes are mere descriptors of the physical characteristics or qualities that give the accommodation a value and the desired features. However, other factors can be determined in the accommodation choice (Kim and Perdue, 2013; Martín et al., 2018). Recently, Kim and Park (2017) reported that the context in which tourists take their decisions, leisure versus business, affect hotel choice, showing that the hotel choice criteria can change depending on the goals, constraints and characteristics that the tourists had at the moment of making the decision.

In the rural accommodation choice, Albaladejo and Díaz (2009) showed that attributes like size or type of building, the quality of the furnishing and the services and activities offered were decisive in rural accommodation choice. But as in hotel choice, other factors can affect the rural accommodation choice. Not all rural tourists are inclined by natural environment and rural culture as Lane (1994) suggested. Several studies (Frochot, 2005; Molera and Albaladejo, 2007; Kastenholz et al., 1999, among others) tested the existence of several segments of rural tourists with different motivations or who seek different benefits from going to a rural environment. Some were in search of the traditional and authentic image in the countryside, others relaxation, others simply wanted to spend their time with the family and friends, while others were attracted by nature-based sports or activities. Thus, different groups of tourists, depending on the motivations or benefits sought, will probably have different preferences regarding the attributes that accommodation should offer, as proposed by Kim and Park (2017), who suggest that leisure and business travellers have different preferences when choosing a hotel.

We propose a model assuming that rural accommodation choice depends not only on the accommodation attributes, such as size, location, price or activities offered, but also on the motivations or needs that tourists seek to satisfy by going to a rural environment. Motivations are unobservable and difficult-to-measure variables and have a lot of dimensions, so they must be referred to as latent variables. A model that allows us to take into account observable (accommodation attributes) and latent variables to analyse the accommodation choice is the hybrid discrete choice (HDC) model. This modelling extends the discrete choice (DC) models incorporating the effect of the latent variables through a multiple indicator multiple cause (MIMIC) model (Márquez et al, 2018). To estimate the model, we use the data from a survey conducted in the Northwest area of the Region of Murcia over several weekends in the autumn of 2003. Part of this data has already been used in the article of Albaladejo and Díaz (2009). We are aware that the survey is over 15 years old, but the data necessary for our study are complex and difficult to obtain. Seeking other objectives, this survey was designed to get all these data.

In our study, the motivations or benefits sought by the tourists to go to the country are decisive in their choice of accommodation and in determining the attributes that an accommodation may have. Given that the motivations are difficult variables to observe and indicators are necessary to define them, an HDC model must be used. To the best of our knowledge, this is the first time that a study applies an HDC model in tourism or hospitality and incorporates motivations to investigate the accommodation choice process. In addition, the motivations imply certain restrictions or limitations when choosing the rural lodging. Thus, in our article, the hypothesis of Kim and Park (2017) that the choice context influences choice decisions for hotels is also extended to rural accommodations.

Furthermore, the findings of the study help promoters and owners of rural accommodation. Knowledge of tourists’ motivations affecting the accommodation choice may help to understand the preferences of a determined type of tourists, and thus in designing a suitable accommodation promotion for them. But, our findings also show that the motivations define heterogeneous preferences regarding the attributes of rural accommodation. This may be a help in deciding on investment in an accommodation offer aimed at improving the favourite features for specific types of tourists.

The next section analyses the modelling rural accommodation choice and reviews the DC models as a way to model the accommodation decision-making process. The third section introduces the choice model structure used in this study and the MIMIC model structure to estimate the latent variables as a function of indicators. The fourth section shows the empirical investigation, presents the models estimated and discusses the results. The fifth section summarizes the significant findings from the study and concludes the article.

Modelling rural accommodation choice

In the hospitality literature, the attributes influencing accommodation choice are basically descriptors of their physical characteristics (size of guest rooms, air-conditioning etc.) or qualities (service and food quality, cleanliness etc.), as shown by Lockyer (2005), Stringam et al. (2010) and Chen et al. (2017), among others. So, these traditional choice models assume that consumers value their choices based on only cognitive attributes (Kim and Perdue, 2013). However, other theoretical paradigms within consumer research and the decision-making process suggest the existence of more systems to value the alternatives in a choice. In the accommodation choice, Kim and Perdue (2013) use cognitive, affective and sensory attributes to analyse hotel choice. They report that two basic dimensions operate in the choice of the tourists: cognitive, relative to physical dimensions, and experiential, referring to affective and sensory attributes. Their empirical study was based on cognitive-experiential self-theory, one of the dual processing theories analysed in Epstein (2003), who provided a theoretical rationale for the usefulness of cognitive and experiential systems in information processing. Later, Kim and Park (2017), based on constructive decision processing theory (Bettman et al., 1998), proposed that hotel choice may vary depending on the choice context or choice goals. In particular, they examine the differences in the hotel choice between two types of tourists with different contexts (leisure vs. business).

In the rural context, accommodations are usually small businesses, such as rural hotels, B&B, rented houses and guesthouses (Ye et al., 2019), which are normally managed by local families as an economic survival measure and for revitalization of these areas. The work of Albaladejo and Díaz (2009) reported that cognitive attributes of accommodations, such as size, location, price or activities offered are decisive in rural lodging choice. But, following the proposal of Kim and Park (2017), the importance of cognitive attributes in rural accommodation choice can also vary depending on the different goals and constraints that the tourists had at the moment of making the decision to visit the countryside.

In its beginnings, rural tourism was related to people visiting the countryside seeking a natural environment and rural culture (Lane, 1994). However, several studies have shown that the rural tourism market is very broad and that motivations to go to the country are very varied. Kastenholz et al. (1999), Frochot (2005) and Molera and Albaladejo (2007), among others tested the existence of several segments of rural tourists with different motivations or seeking different benefits from going to a rural environment.

The motivations or needs that tourists seek to satisfy when they decide to go to the countryside could imply certain restrictions or limitations when choosing the lodging where to spend their vacation time. If tourists are seeking to spend time with family and friends, they will tend to travel with children and will probably place greater weight on accommodations offering sports activities or other facilities, which will allow them to spend time in common. However, if tourists are very interested in nature and relaxation, they will tend to give a low value to activities and will presumably choose remote and quiet accommodations. Thus, motivations for going to the country could affect the choice of rural accommodation.

Taking into account the above, our proposal is that the rural accommodation choice depends not only on cognitive attributes but also on other variables that allow us define the motivations or needs of these rural tourists. But the determination of these motivations is no easy task. They represent consumers’ perceptions or attitudes about going to the country (Frochot, 2005) and, therefore, are not directly observable variables (Walker and Ben-Akiva, 2002). Generally, the unobservable or latent variables are measured using other variables known as indicators and are linked to sociodemographic characteristics of the tourists. One model that allows both types of variables to explain the choice is the DC model known as HDC model.

DC models

Over the last decades, DC modelling has had an increasingly important role in identifying the attributes defining the choices and in modelling individuals’ choice behaviour in tourism, hospitality and leisure (Crouch and Louviere, 2000). They have principally been applied to analyse not only destination choices (Eymann and Ronning, 1997; Morley, 1994; Seddighi and Theocharous, 2002) but also to travel modes (Kelly et al., 2007), heritage attraction (Apostolakis and Jaffry, 2005), park fees (Mmopelwa et al., 2007) or tourist information offices (Araña et al., 2016). Several applications can also be found in the hospitality sector, particularly in the hotel choice (Chen et al., 2017; Martín et al., 2018; Román and Martín, 2016; Victorino et al., 2005), in hotel room choice (Masiero et al., 2015, 2016) and the choice of rural house stay (Albaladejo and Díaz, 2009).

DC modelling was developed by McFadden (1974) who proposed the popular multinomial logit (MNL) model. This model, combining the random utility maximization (RUM) and the hedonic evaluation of alternatives, obtains the individual probabilities of choice among different alternatives – in this case, accommodations. The MNL model is highly popular because of its tractability but has been criticized because of the so-called independence of irrelevant alternatives property (Train, 2003) and by its limitation in capturing individuals’ differences in taste (Espino, Martín and Roman, 2008; Train, 2003). So, other models of the ‘Logit family’ have been developed which are aimed at relaxing restrictions, while maintaining tractability, such as the nested logit model, the generalized extreme value model and the mixed logit model (Train, 2003). For example, Chen et al. (2017) used a nested logit model to investigate whether the accommodation decision-making process is better explained by a hierarchical structure, tourists first chose between hotels and non-hotels and afterwards they chose a specific type of accommodation. The mixed logit (ML) has also been used to model the accommodation choice because it allows us to introduce heterogeneity preferences for the accommodation attributes among tourists (Albaladejo and Díaz, 2009; Masiero et al., 2015; Román and Martín, 2016).

Traditionally, these DC models consider only observable and tangible characteristics of the alternatives and the tourists as explanatory variables. However, other not easily observable factors can affect the choices, which are defined as latent variables and require effect indicators to be characterized (Park and Yoon, 2009). In recent decades, various attempts have been reported to include the effect of both observable and latent variables in choice models (Bahamonde-Birke and Ortúzar, 2014b; Bollen, 1989; Green, 1984; Keane, 1997). Nowadays, the most popular of these models that include latent variables are the HDC models. The advantages of these models have been shown from empirical and theoretical perspectives by several authors, such as Ashok et al. (2002), Ben-Akiva et al. (2002), Vredin-Johansson et al. (2006), Tam et al. (2010) and Alvarez-Daziano and Bolduc (2011), among others. In addition, there are a lot applications of the models to cycling demand (Maldonado-Hinarejos et al., 2014; Motoaki and Daziano, 2015).

The majority of the studies that use the HDC framework consider perceptions and attitudes as latent variables (Bahamonde-Birke, et al., 2017; Daly, et al., 2012; Raveau, et al., 2010; Walker and Ben-Akiva 2002; Yáñez et al., 2010), although other effects have also been studied, as Thorhauge et al., (2017) brought to light. None of these studies, however, account for the effect of motivations or benefits sought.

HDC models

DC models are based on RUM (McFadden, 1974), which introduces the concept of individual choice behaviour being intrinsically probabilistic. According to this theory, each individual has a utility function associated with each of the alternatives and this function is not known to the analysts with certainty. They consider that perceived utility of alternative j for individual i, $U_{i j}$ , consists of two components: a deterministic utility, $V_{i j}$ , observed by the analyst, which can be explained by the alternative’s attributes as well as by the decision-maker’s characteristics, and a random error term, $ε_{i j}$ , which indicates the unknown factors about individuals or measurement errors or imperfect information. That is,

U_{i j} = V_{i j} + ε_{i j}

Traditional models assume that only tangible and objective attributes define the deterministic utility, $V_{i j}$ . The HDC model also incorporates intangible elements associated with the attitudes, perceptions or other psychological characteristics of the individuals to define $V_{i j}$ . So, if we assume a linear specification, as usual, the deterministic utility, $V_{i j}$ , can be expressed as

V_{i j} = \sum_{k} β_{j k} x_{i j k} + \sum_{l} θ_{j l} η_{i j l}

where $x_{i j k}$ represents the observed attributes of the alternatives, $η_{i j l}$ the latent variables and $β_{j k}$ and $θ_{j l}$ are the parameters to be estimated.

Then,

U_{i j} = \sum_{k} β_{j k} x_{i j k} + \sum_{l} θ_{j l} η_{i j l} + ε_{i j}

The RUM theory also assumes that the individuals select the alternative with the maximum value of perceived utility from among the alternatives of a choice set. So, because of the presence of the random term, the probability that individual i selects alternative j in the choice set B is given by the expression

\begin{matrix} P_{i j} = Pr [U_{i j} \geq U_{i k} \forall k \in B k \neq j] \\ = Pr [ε_{i k} \leq ε_{i j} + (V_{i j} - V_{i k}) \forall k \in B k \neq j] \end{matrix}

Different HDC models can be derived depending on the assumptions made for the distribution of the random components of the utility. Commonly, this random component is assumed to distribute independent and identically distributed extreme value type-I.

Once the deterministic utility has been defined and the distribution of the random components of the utility has been considered, the HDC model can be estimated sequentially or simultaneously depending on how the available information is used (Raveau et al, 2010). The second stage of the sequential estimation requires that the expected value of the latent variables has been previously obtained. So, it is necessary that the intangible elements to be modelled as latent variables using a MIMIC model of structural equations.

Latent variables and structural equations

The unobservable or latent variables are identified by estimating them as a function of other observed variables, known as indicators. The indicators measure the level of agreement of the consumers with a set of questions concerning them. They can be continuous, binary or categorical variables and are obtained in a preference survey (Bahamonde Birke et al, 2017; Muthén, 1993). The most popular approach to estimate latent variables as a function of indicators is through a MIMIC model (Bahamonde Birke et al, 2017), which is a type of structural equation modelling (SEM). The SEM has been widely used in tourism literature (Nunkoo et al., 2013). An example of its use in the context of Spanish tourism destination is the work of Perles et al. (2011).

The MIMIC model has two parts, the structural equations, which explain the relationships between latent variables and a set of characteristics of the individuals and the alternatives, and the measurement equations, which link the indicators and latent variables. Thus, this model can be defined by

I_{i} = Λ η_{i} + ξ_{i}

η_{i} = Γ ω_{i} + ζ_{i}

where I_i corresponds to a $(P \times 1)$ vector of indicators associated with individual i, with P being the number of indicators; $Λ$ is a $(P \times L)$ matrix of unknown parameters that relates the L latent variables to the P indicators (called factor loadings); $η_{i}$ is a $(L \times 1)$ vector of latent variables, where L indicates the number of latent variables and $ξ_{i}$ is a $(P \times 1)$ vector of measurement errors normally distributed with a diagonal covariance matrix Θ. In the structural equation (6), $Γ$ is a $(L \times M)$ matrix of unknown parameters; $ω_{i}$ is a $(M \times 1)$ vector of explanatory variables causing the latent variables, where M is the number of explanatory variables causing the latent variables; and $ζ_{i}$ is a $(L \times 1)$ vector of structural errors or residuals that typically has a multivariate normal distribution with zero mean and a full covariance matrix Ψ.

When the indicators are continuous variables, the most widely used estimation method in SEM is the normal theory-based maximum likelihood (ML) method based on the sample covariance matrix, but it is not appropriate when the observed variables are not continuous or when multivariate normal distribution in the population is not plausible (Li, 2016).

For this reason, when the indicators are ordinal variables, the underlying response variable approach (Jöreskog, 1994; Jöreskog and Moustaki, 2001; Muthén, 1984) should be adopted. This method assumes that each observed ordinal variable I_i is considered to be generated by an underlying unobserved continuous variable, so the variable categorical variable I_i, with k_r response categories, takes a value k; with k = 1, 2,…, k_r, if and only if it is verified that there is a set of parameters τ_r_;k, called cutpoints or thresholds, such that

I_{r i} = \{\begin{matrix} 1 & if & - \infty = τ_{r, 0} < I_{r i}^{*} \leq τ_{r, 1} \\ 2 & if & τ_{r, 1} < I_{r i}^{*} \leq τ_{r, 2} \\ ⋮ & ⋮ \\ k_{r} & if & τ_{r, k_{r} - 1} < I_{r i}^{*} \leq τ_{r, k_{r}} = + \infty \end{matrix}

And in this case, equation (5) would have the expression

I_{i}^{*} = Λ η_{i} + ξ_{i}

where I_i has been replaced by $I_{i}^{*}$ .

Thus, when the indicators are categorical, the measurement equations are defined by equations (8) and (7), which represent a system of ordered probit models. In this case, two estimation methods have been proposed in the specialized literature. The first is an ML method with robust corrections (robust ML) based on the sample covariance matrix proposed by Satorra and Bentler (1994) and Yuan and Bentler (1998). This method was developed for approximately continuous but potentially non-normal variables observed, which could be compatible with ordinal observed variables with at least five response categories.

The second method, proposed by Muthén (1984; also see Jöreskog, 1994), uses the asymptotically distribution-free diagonally weighted least squares (DWLS) estimator along with a polychoric correlation matrix. The parameter estimates obtained by DWLS are not asymptotically efficient (Li, 2016). This can be overcome by implementing robust corrections to standard errors in the estimated asymptotic covariance matrix of the parameter estimates (Muthén et al., 1997). Additionally, a type of robust correction to χ² statistics can be also performed, which entails adjusting both the mean and the variance of the test statistic to make their shapes approximately the reference χ² distribution with its associated degrees of freedom (Asparouhov and Muthén, 2010).

Empirical investigation

Using data about rural accommodation choices made by tourists to the Region of Murcia, an MNL and two specifications of the HDC model are estimated to study the effect of including/omitting information about the motivations for going to the country on rural accommodation choice. The sequential method is used to estimate the HDC models. Although this method produces biased estimates and tends to underestimate the parameters’ standard deviations, it is still widely used because it requires significantly fewer computational resources (Bahamonde-Birke and Ortúzar, 2014a; Palma et al., 2018) as opposed to the simultaneous approach. In addition, unfortunately, there is no commercial software available to apply the simultaneous method.¹

Thus, the estimation of the HDC models requires two phases. First, the MIMIC model is applied to obtain parameter estimates for the equations relating the latent variables with the explanatory variables and motivation indicators. With these parameters in the structural equation, expected values of the latent variables which summarize each tourist’s motivations for going to the country are obtained. In a second stage, these latent variables are added to the set of explanatory variables that are used to estimate the two specifications of the HDC model.

Data and variables

The data for the study come from a survey conducted in the Northwest area of the Region of Murcia over several weekends in the autumn of 2003. As said in the introduction, we are aware that the survey is over 15 years old, but the data necessary to show the behavioural intentions to go to the country affecting the accommodation choice are complex and difficult to obtain. We needed individual data of the tourists which allowed us to collect information of the respondents’ profile and their travel conditions. We also wanted to identify the motivations for going to the countryside and to ascertain the preferences of the tourists in relation to the attributes which affected the choice of rural accommodation. To obtain more precise information, a stated preference experiment was required. This survey had all these data.

The main objective of the survey was to carry out a stated preference experiment which measured consumers’ preferences when choosing a hypothetical accommodation for their holidays in a rural area. Albaladejo and Díaz (2009) used these data and provide all the details regarding the experimental design and the survey therein. In addition, each participant was asked to complete a pencil-and-paper questionnaire about their demographic and socio-economic features (age, gender, marital status, level of education, income etc.), trip-related characteristics (expenditure per capita, group size, activities etc.) and a set of indicators (benefits sought), which were rated by the respondents on a five-point Likert-type scale (from 1 = strongly disagree to 5 = strongly agree), which allows us to obtain the most important motivations for visiting a rural area.

A total of 307 questionnaires were completed in the stated preference experiment, and they were all valid for the study. However, as presented in Table 2, there are individuals who did not complete the motivation indicators. Table 1 lists the respondent profiles and their travel conditions, and Table 3 presents the attributes and levels used in the choice experiment.

Table 1.

Profile of the respondents.

Characteristics	Level	Frequency	%
Socio-economic characteristics
Gender	Male	149	48.50
Gender	Female	158	51.50
Age (years)	≤25	83	27.00
	26–40	186	60.60
	>40	38	12.40
Place of residence	Murcia	222	72.30
Place of residence	Rest of Spain	85	27.70
Level of education	Primary	59	19.20
	Secondary or similar	92	30.00
	university diploma, degree or similar	158	50.80
Occupation	Salaried worker/self-employed	232	75.60
Occupation	Unemployed/student retired/household duties	75	24.40
Civil status	Married	130	42.30
	Single	168	54.70
	Divorced or separated/widowed	9	2.90
Travel conditions
Travelling with children	Yes	124	40.40
Travelling with children	No	183	59.60
Travelling with friends	Yes	158	51.50
Travelling with friends	No	149	48.50
Travelling with family	Yes	55	17.90
Travelling with family	No	252	82.10
Using the Internet to organize your trip	Yes	129	42.00
Using the Internet to organize your trip	No	178	58.00
Total expense (without accommodation) over €60	Yes	131	42.10
Total expense (without accommodation) over €60	No	178	57.30
Travel group six people or more	Yes	208	67.10
Travel group six people or more	No	101	32.90
Travel to rural areas more than twice a year	Yes	38	12.40
Travel to rural areas more than twice a year	No	289	92.50
Travel frequency	More than 5 times a year	23	7.50
Travel frequency	Five limes or less a year	284	87.60

Table 2.

Descriptives of the indicators.

Indicator	n	Mean	SD	Median	Skew	Kurtosis
Peaceful atmosphere	306	4.297	0.902	5	−1.285	1.351
Attractive landscapes	306	4.062	1.089	4	−1.018	0.257
Environmental quality and nature	305	3.816	1.066	4	−0.653	−0.261
Not crowded	303	4.099	1.123	4	−1.284	0.989
Rustic atmosphere	305	3.964	1.027	4	−0.800	0.020
Having a good time with family	301	3.369	1.695	4	−0.397	−1.563
Relationships with local residents	303	2.997	1.323	3	0.023	−1.037
Opportunities for meeting people	305	2.931	1.327	3	0.075	−1.039
Opportunities for children	302	2.669	1.720	2	0.283	−1.667
Opportunities for resting	305	4.125	1.111	4	−1.279	0.949
Independence and flexibility	302	3.914	1.126	4	−0.847	−0.058
Short travel distance	304	3.997	1.115	4	−1.074	0.491
Good price	307	4.072	1.076	4	−1.067	0.424

Table 3.

Attributes and levels of the experimental design.

Attribute	Levels	Denomination
Type of building	4	Original
		New
		Mill house
		Wooden
Location	4	Town
		Huerta*
		Country-mountain
		Village
Number of bedrooms	4	1
		2
		3
		4
Price per room	4	€80
		€100
		€120
		€140
Hiring horses	2	Yes
Hiring horses	2	No
Play area	2	Yes
Play area	2	No
Meal service	2	Yes
Meal service	2	No
Swimming pool	2	Yes
Swimming pool	2	No
Sports facilities	2	Yes
Sports facilities	2	No
Mini-farm	2	Yes
Mini-farm	2	No
Bathroom	2	Shared
Bathroom	2	Private
Type of rent	2	Whole house
Type of rent	2	By room
‘Q’ Quality award	2	Yes
‘Q’ Quality award	2	No
Booking	2	Internet
Booking	2	Telephone

* Irrigated fruit and vegetable growing area.

Estimation of MIMIC model

As a first step, an exploratory factorial analysis (EFA) was conducted on the responses to the questions on motivations to identify the latent variables. Table 4 presents the results obtained with the EFA. According to these results, it seems reasonable to choose a four-factor solution, which explains about 73% of the variance of the information contained in the 13 indicator variables. Factor 1 will be called Environment because it refers to the surroundings enjoyed in a rural context. Factor 2, Individuality, alludes to the conditions for peaceful, individual travel. Family will be the name of factor 3 because the indicators are related to travel with family. Finally, Sociability will be factor 4, which explains the trips motivated by contact with other persons.

Table 4.

Exploratory factor analysis.

Item	Factor 1	Factor 2	Factor 3	Factor 4
Peaceful atmosphere	0.7055	0.1387	0.0500	0.0526
Attractive landscapes	0.8052	−0.0647	0.0682	0.02039
Environmental quality and nature	0.7229	−0.0803	0.2790	0.0488
Not crowded	0.8117	0.0820	−0.1601	−0.0807
Rustic atmosphere	0.7186	0.0661	−0.1208	0.0582
Having a good time with family	0.0167	−0.0020	0.9221	0.05210
Relationships with local residents	0.0622	−0.0131	0.0248	0.9205
Opportunities for meeting people	−0.0500	0.0235	−0.0198	0.9593
Opportunities for children	0.0018	0.0425	0.9348	−0.0188
Opportunities for resting	0.1494	0.7118	0.1952	−0.0396
Independence and flexibility	0.0570	0.7871	−0.0977	0.1349
Short travel distance	−0.0732	0.7768	0.1346	0.0296
Good price	−0.0095	0.8350	−0.0880	−0.0626
Eigenvalue	4.9807	1.8752	1.4849	1.0950
Proportion explained variance	0.2321	0.1985	0.1524	0.1429
Cumulative proportion explained variance	0.2321	0.4306	0.5829	0.7258

Note: Extraction method: principal components analysis on polychoric correlation matrix and rotation method: oblimin.

Before estimating the MIMIC model, it is convenient to test the reliability and validity of the measurement model embedded in the MIMIC model by means of a confirmatory factor analysis, which is presented in Table 5. Goodness-of-fit statistics² for this model indicate an acceptable overall fit for the data (comparative fit index = 0.975, Tucker–Lewis index = 0.965, root mean square error of approximation = 0.052, standardized root mean square residual = 0.057 and weighted root mean square residual = 0.775). Convergent validity is established by statistically significant standardized factor loadings, which take values in a range from 0.64 to 0.99. All construct reliabilities (Cronbach’s α and composite reliability) are above the recommended threshold of 0.60 (Bagozzi and Yi, 1988). With respect to the average variance extracted for any two factors (Fornell and Larcker, 1981), it should be noted that it is greater than 0.5, indicating that there is no problem of reliability, and it is also smaller than the squared correlation between them, so the discriminant validity is verified, as seen in Table 6.

Table 5.

Reliability, internal consistency and convergent validity of the measurement equations.

Indicator	Loadings	t	CA (α)	CR (Ω)	AVE
Environment
E1. Peaceful atmosphere	0.846***	23.780
E2. Attractive landscapes	0.745***	18.572
E3. Environmental quality and nature	0.676***	15.542	0.840	0.842	0.518
E4. Not crowded	0.675***	14.792
E5. Rustic atmosphere	0.641***	14.523
Individuality
I1. Opportunities for resting	0.872***	22.143
I2. Independence and flexibility	0.759***	21.408	0.831	0.835	0.562
I3. Short travel distance	0.637***	12.581	0.831	0.835	0.562
I3. Short travel distance	0.637***	12.581
I4. Good price	0.709***	14.296
Family
F1. Having a good time with family	0.924***	13.708	0.906	0.906	0.828
F2. Opportunities for children	0.895***	12.572	0.906	0.906	0.828
F2. Opportunities for children	0.895***	12.572
Sociability
S1. Relationships with local residents	0.993***	18.464	0.884	0.894	0.811
S2. Opportunities for meeting people	0.797***	17.199	0.884	0.894	0.811
S2. Opportunities for meeting people	0.797***	17.199
p-value (χ²)	0.000
CFI	0.975
TLI	0.965
RMSEA	0.065
p-value RMSEA 0.05	0.052
SRMR	0.057
WRMR	0.775

Note: CA: Cronbach’s α; CR: composite reliability; AVE: average variance extracted; CFI: comparative fit index; TLI: Tucker–Lewis index; RMSEA: root mean square error of approximation; SRMR: standardized root mean square residual; WRMR: weighted root mean square residual.

***p < 0.001.

Table 6.

Discriminant validity.

	Environment	Individuality	Family	Sociability
Environment	0.518
Individuality	0.393	0.562
Family	0.192	0.049	0.828
Sociability	0.130	0.123	0.113	0.811

Note: AVE on the main diagonal and ρ² on lower triangular. AVE: average variance extracted.

After identifying the latent variables, we apply a MIMIC model to relate the latent variable with the explanatory variables and indicators. The structure of our MIMIC model is shown in Figure 1. The links between the latent variables and the indicators (i.e. the measurement equations) are assumed to be of ordered probit form, while links between tourists’ observable characteristics and their latent variable (i.e. the structural equations) are assumed to be linear.

Figure 1.

Path diagram of the MIMIC model. MIMIC: multiple indicator multiple cause.

Estimation of this MIMIC model is carried out using the weighted least squares means and variance (WLSMV) adjusted estimator implemented in the R package lavaan (version 0.5-23.1097) for latent variable analysis (Rosseel, 2012). Note that the WLSMV estimator implies a DWLS estimation, robust standard errors and the mean and variance adjusted test statistics to compensate the loss of efficiency when the full weight matrix is not engaged (Li, 2016). Table 7 reports the results of the MIMIC model³ that measures tourists’ motivations.

Table 7.

MIMIC component estimates.

Measurement equation (R²)	Estimate	SE	T	p-value
Environment	Reliability: 0.859 (α), 0.860 (Ω)
E1. Peaceful atmosphere (0.693)	1
E2. Attractive landscapes (0.488)	0.823***	0.063	13.025	0.000
E3. Environmental quality and nature (0.576)	0.902***	0.072	12.553	0.000
E4. Not crowded (0.521)	0.853***	0.066	13.016	0.000
E5. Rustic atmosphere (0.482)	0.818***	0.062	13.293	0.000
Individuality	Reliability: 0.871 (α), 0.873 (Ω)
I1. Opportunities for resting (0.747)	1
I2. Independence and flexibility (0.672)	0.935***	0.071	13.100	0.000
I3. Short travel distance (0.653)	0.919***	0.076	12.139	0.000
I4. Good price (0.465)	0.750***	0.072	10.409	0.000
Family	Reliability: 0.899 (α), 0.899 (Ω)
F1. Having a good time with family (0.810)	1
F2. Opportunities for children (0.822)	1.016***	0.156	6.513	0.000
Sociability	Reliability: 0.905 (α), 0.905 (Ω)
S1. Relationships with local residents (0.805)	1
S2. Opportunities for meeting people (0.850)	1.034***	0.085	12.143	0.000
Environment (0.164)
Place of residence: Murcia	−0.239	0.131	−1.823	0.068
Level of education: University diploma, degree or similar	−0.327	0.161	−2.033	0.042
Travelling with family	0.337	0.204	1.651	0.099
Travelling with children	0.308	0.185	1.663	0.096
Travel group six people or more	−0.389	0.133	−2.925	0.003
Travel frequency: More than five times a year	0.378	0.224	1.686	0.092
Individuality (0.294)
Place of residence: Murcia	−0.219	0.131	−1.678	0.093
Level of education: Secondary or similar	−0.467	0.183	−2.559	0.010
Level of education: University diploma, degree or similar	−0.685	0.181	−3.792	0.000
Travel group six people or more	−0.749	0.143	−5.253	0.000
Family (0.631)
Status civil: Single	−0.669	0.194	−3.453	0.001
Level of education: University diploma, degree or similar	−0.368	0.190	−1.940	0.052
Travelling with friends	−0.445	0.119	−2.242	0.025
Travelling with family	0.605	0.247	2.455	0.014
Travelling with children	0.749	0.198	3.788	0.000
Travel frequency: More than five times a year	0.576	0.270	2.132	0.033
Sociability (0.224)
Gender: Male	−0.205	0.124	−1.654	0.098
Age: 26–40	0.482	0.196	2460.000	0.014
Level of education: University diploma, degree or similar	−0.661	0.162	−4.076	0.000
Occupations: Salaried worker/self-employed	0.280	0.156	1.792	0.073
Using the internet to organized your trip	−0.235	0.132	−1.774	0.076
Total expense (without accommodation) over €60	0.238	0.128	1.851	0.064
Travel frequency: More than five times a year	0.543	0.251	2.169	0.030
p-value (χ²)	0.048
CFI	0.984
TLI	0.98
RMSEA	0.023

Note: MIMIC: multiple indicator multiple cause; CFI: comparative fit index; TLI: Tucker–Lewis index; RMSEA: root mean square error of approximation.

The sign of the indicators’ coefficients means the latent variables can be interpreted and the structural equations help to explain how latent variables are built by means of a causal relationship (Motoaki and Daziano, 2015; Palma et al., 2018). The indicators’ coefficients with a positive sign show that these indicators are positively correlated with the latent variable that explains them, so high levels in the latent variable are caused by high levels in these indicators. For example, high levels of the Environment latent variable are correlated with high levels of environmental quality and nature, tranquillity, an attractive landscape and a non-crowded place. In addition, the Environment motivation is related with tourists who travel more than five times a year, in a small group (five people maximum) and with family and children, living outside Murcia and not having a higher education level.

The Individuality latent variable alludes to the search for rural accommodation that is close to their home, has a reasonable price, allows them to rest and provides them a feeling of independence and flexibility. This latent variable is explained by place of residence, level of education and travel group size. For example, a higher Individuality latent variable is expected for people living outside the Region of Murcia, preferring to travel in small groups and with the lowest level of education.

The Family latent variable is connected with individuals who are interested in having a good time with family and opportunities for children. A higher level in the Family latent variable is expected for individuals travelling more than five times a year, with family and children, while for the individual who travels with friends, is single and has a higher level of education, it is expected to be lower.

Finally, the Sociability latent variable is associated with people who see choosing a rural accommodation as an opportunity to meet people and be in contact with local residents. Thus, high levels of the Sociability variable are expected for people aged 26–40 years, salaried workers or self-employed, with a travel budget exceeding €60 and who travel more than five times a year.

Estimation of DC models

Once the MIMIC model has been appraised, the expected values of latent variables for each individual are estimated using a least squares regression approach. These latent variable scores, which are numerical values that indicate an individual’s relative standing on the latent variable, have been added afterwards to the set of explanatory variables to estimate the DC models.

To detect preferences of tourists regarding the attributes of rural accommodations, three DC models are considered, an MNL and two specifications of HDC model. The MNL model only accounts for homogeneous preferences between tourists, so, their indirect utility functions depend only on the attributes of rural accommodations presented in Table 8. In the first HDC specification, individual-specific latent variables are entered directly in the indirect utility function and in the linear form. The second HDC specification also considers interaction between latent variables and the attributes of the alternatives.

Table 8.

Modelling variables.

Attributes	Types
House price	Continuous variable
House size	Continuous variable
Original building	1 if the lodging is a typical house in the area
New building	1 if the house is of new building in the area
Wooden house	1 if the lodging is a wooden house
Town house	1 if the house is located on the outskirts of a town
Farm house	1 if the house is located in area of huerta*
Country house	1 if the house is located in area of country or mountain
Mini-farm	1 if the house has mini-farm
Hiring horses	1 if it is possible riding horses
Rented whole house	1 if the whole house is rented
Shared bathroom	1 if the bedrooms do not have their own bathroom
Quality	1 if the house bears ‘Q’ awarded
Internet booking	1 if the house can be booked on the Internet
Latent motivations
Environment	Continuous variable
Individuality	Continuous variable
Family	Continuous variable
Sociability	Continuous variable

* Irrigated fruit and vegetable growing area.

A classical sequential selection procedure is used to define the predicting variables of each specification. This involves estimating alternative models removing predicting variables and the model chosen is the best one according to the likelihood ratio test and information criteria of Akaike. The three specifications proposed are estimated using NLOGIT 5.0 econometric software (Greene, 2012), and the results are displayed in Table 9 in conjunction with the main goodness-of-fit measures.

Table 9.

Logit models estimations results.

Attribute	MNL			HDC model			HDC model with interactions
Attribute	Estimate	SE	t-Stat	Estimate	SE	t-Stat	Estimate	SE	t-Stat
House price	−0.31607***	0.07373	−4.29	−0.30546***	0.07299	−4.19	−0.41247***	0.09004	−4.58
House size	0.69366***	0.08484	8.18	0.49815***	0.08666	0.75	0.71020***	0.10856	6.54
Original building	0.28731***	0.07578	3.79	0.28365***	0.07545	3.76	0.37372***	0.1266	2.95
New building	−0.57385***	0.08304	−6.91	−0.53652***	0.08044	−6.67	−0.65258***	0.09647	−6.76
Wooden house	0.17621**	0.07819	2.25	0.16063**	0.07677	2.09	0.15842**	0.07847	2.02
Town house	−0.42545***	0.08195	−5.19	−0.40943***	0.08006	−5.11	−0.48601***	0.09348	−5.2
Farm house	0.11323	0.07860	1.44	0.09377	0.07663	1.22	0.09929	0.07889	1.26
Country house	0.27964***	0.07696	3.63	0.27993***	0.07601	3.68	0.31698***	0.08606	3.68
Mini-farm	0.06831	0.04459	1.53	0.06801	0.04418	1.54
Hiring horses	0.11353**	0.04508	2.52	0.10813**	0.04438	2.44	0.12591***	0.04564	2.76
Rented whole house	−0.08113*	0.04503	−1.8	−0.08833**	0.04432	−1.99	−0.09236	0.06996	−1.32
Sports facilities							0.1242	0.08453	1.47
Quality	0.15929***	0.04505	3.54	0.15995***	0.04435	3.61	0.16195***	0.04553	3.56
Internet booking							0.14050**	0.06954	2.02
Environment				−0.13316	0.15832	−0.84	−0.17286	0.1636	−1.06
Individuality				−0.39968***	0.12001	−3.33	−0.46777***	0.13499	−3.47
Family				−0.14542*	0.08663	−1.68	−0.36009***	0.12972	−2.78
Sociability				0.87707***	0.1175	7.46	0.99179***	0.16224	6.11
New building × Environment							−0.18516**	0.08905	−2.08
Country house × Environment							0.11648	0.07214	1.61
Swimming pool × Environment							−0.06338	0.04325	−1.47
Rented whole house × Environment							0.13689*	0.0701	1.95
House size × Individuality							0.09131**	0.03736	2.44
Original building × Individuality							0.17635**	0.07445	2.37
Sports facilities × Individuality							0.06694	0.05099	1.31
Mini-farm × Individuality							−0.08146**	0.03268	−2.49
Rented whole house × Individuality							−0.08723	0.06517	−1.34
Internet booking × Individuality							0.10530**	0.04785	2.2
House size × Family							0.09671***	0.03678	2.63
New building × Family							0.15218**	0.06155	2.47
Meal service × Family							−0.05915	0.04025	−1.47
Mini-farm × Family							0.11305***	0.03795	2.98
House price × Sociability							0.15609*	0.0877	1.78
House size × Sociability							−0.24262**	0.10723	−2.26
Original building × Sociability							0.18033**	0.08392	2.15
Town house × Sociability							0.13714*	0.08053	1.7
Play area × Sociability							−0.07643	0.04754	−1.61
Meal service × Sociability							0.08091*	0.04785	1.69
Swimming-pool × Sociability							0.0699	0.04534	1.54
Sports facilities × Sociability							−0.12195**	0.05765	−2.12

Fit indices
Observations	1160			1160			1160
Individuals	290			290			290
Number of parameters	12			16			39
Log-likelihood function	−1099.66035			−1053.09728			−1013.11369
Restricted log-likelihood function	−1274.39025			−1274.39025			−1274.39025
McFadden pseudo R²	0.137108629			0.173646157			0.20502084
Adjusted McFadden pseudo R²	0.128080995			0.16207865			0.17733853
AIC/N	1.916655776			1.843271172			1.813989121

Note: MNL: multinomial logit; HDC: hybrid discrete choice; AIC: Akaike information criterion.

***p < 0.01.

**p < 0.05.

*p < 0.1.

The MNL specification (first column in Table 9), which assumes constant coefficients across individuals, appears to the data with an acceptable value of 0.14 for its adjusted R². The coefficients of the variables house price, new building, town house and rented whole house are negative, while the rest are positive. Thus, the tourists of the Murcia Region prefer rural accommodations of low price and large size, not built for the express purpose of being rented out, located in the countryside, that bear the ‘Q’ quality control distinction and have the possibility of hiring horses.

The first HCD specification including latent variables in the indirect utility function in linear form (second column in Table 9) shows an adjusted R² of 0.17, 26.7% higher than in the MNL model. This indicates that inclusion of the latent attributes identified by the structural equations resulted in an improvement in the goodness-of-fit of the choice model. The significant attributes of the alternatives are the same as in the MNL model and although the values of their coefficients are different, their signs coincide. Therefore, the interpretation of attributes of alternatives is similar to the interpretation of the MNL model. Regarding motivations, Individuality, Family and Sociability are statistically significant in the rural accommodation choice. The coefficient of Sociability motivation has a positive sign, so the higher motivation to establish social relationships, the higher the probability of choosing a rural accommodation, all other things being equal. However, the coefficients of Individuality and Family motivations are negative, so the probability of lodging in a rural accommodation is lower for tourists with high motivations to travel with the family or to make a short, calm, individual trip.

Lastly, the third column of Table 9 shows the results of the HCD model with interactions among the individual-specific latent motivations and the attributes of rural accommodations. All attributes were interacted with latent variables; however, only significant interactions were kept in the final model. The last specification has the highest value for the adjusted R² (0.20). Our results show heterogeneous preferences among respondents, as some coefficients vary specifically for certain segments of tourists, according to their latent motivation.

Thus, individuals whose motivation for choosing rural accommodation is Environment, value more negatively than the rest of the tourists a new building accommodation and value positively that it can also be rented by rooms. On the other hand, the original houses are more positively perceived by consumers whose main motivation is Individuality together with the possibility of booking accommodation online.⁴ For the segment of tourists motivated by Family, the possibility that a rural accommodation has a mini-farm is positively valued, ceteris paribus. In addition, they prefer a larger accommodation than the rest of the tourists and value a new building negatively, although less so. Finally, for tourists motivated mainly by Sociability, the price, the size and situation of the house in the town are important but to a lower degree in relation to other tourists, while the original building is much more important. For this segment, the probability of choosing accommodation with a meal service is higher, and lower if sports facilities are available.

Concluding remarks

Based on the theory of constructive consumer choice process of Bettman et al. (1998), this article proposes that the motivational goals of tourists to go to the country influence the choice of rural accommodation and influence the importance of the attributes of accommodation. To contrast this hypothesis, three DC models were estimated for tourists who visited the Northwest area of the Region of Murcia: an MNL and two specifications of the HDC model. The MNL model considers as explanatory variables only objective and tangible attributes defining the accommodations, such as size, location, price or activities offered, while the HDC specifications also introduce latent variables which identify the tourists’ motivations for going to the country.

The inclusion of latent variables in the DC models is made using the MIMIC model of structural equations, which links them to their indicators and sociodemographic characteristics of the tourists. In our study, four latent variables identifying the motivations of rural tourists are found: Environment, Individuality, Family and Sociability. These variables included in the DC models allow us to detect heterogeneous preferences among tourists. Our estimation of the HDC models shows that the tourists with a greater motivation to socialize with both local residents and other tourists have a higher probability of choosing a rural accommodation in the Region of Murcia than tourists motivated by spending time with the family or motivated by making a short, calm, individual trip. But the motivations can also de ne heterogeneous preferences regarding the attributes of rural accommodations, so, for example, the fact that the accommodation has a mini-farm positively influences the choice of accommodation for tourists motivated to travel with the family and children, but negatively influences tourists who search to make a short, calm, individual trip.

The two estimated HDC models offer significant improvements in model goodness of fit than the MNL model and give evidence sustaining that motivations may affect the probability of rural accommodation choice and that may incline tourists towards certain attributes of the lodgings. However, the estimation of the HDC models requires much more information and a more complex estimation process than the MNL model. So, according to Palma et al. (2018) and Vij and Walker (2016), the HDC models may be used only for studies in a particular context. Such is the case of our study, which aimed to make a deeper analysis of the choice of rural accommodation, proposing that the motivational goals of the tourists also affect their choices.

This complex analysis developed in this study contributes to the literature by using a DC model, the HDC models, which allow us to analyse the impact of decisive and not directly observable factors on the accommodation choice, such as the tourists’ motivations. Other previous accommodation choice studies use DC models, but as far as we know, this is the first to incorporate motivations to investigate the accommodation choice process.

The results of this study also support the conceptual framework of Bettman et al. (1998), who show that the rural accommodation choice varies depending on the decision context of the tourist (Environment, Individuality, Family or Sociability) and decision context also drives their preferences in relation to the attributes of the accommodation. In consequence, the choice context is decisive in the tourism accommodation choice, be it hotel (Kim and Park, 2017) or rural accommodation.

Finally, our findings also show that tourists may incline towards certain attributes of the accommodation. Consequently, this study could be a valuable tool for planners and managers of rural establishments when ascertaining tourists’ preferences for facilities and amenities that are offered by a rural accommodation. It provides information about those attributes of the accommodations which are appreciated by the tourists, those which are not valued by any tourists whatever their motivation for going to the country, or those which only are attractive for tourists who have a specific motivation.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was partially supported by the Agencia Estatal de Investigacion of the Spanish Government under research projects ECO2016-76178-P and PID2019-107192GB-I00 to the first author (Isabel Albaladejo). These projects are co-financed by FEDER funds.

ORCID iD

Isabel P Albaladejo

M Teresa Díaz-Delfa

Notes

Author biographies

Isabel P Albaladejo is an associate professor in Quantitative Methods at the University of Murcia. She teaches economic mathematics at the Faculty of Economics and Business. Her research interests include discrete choice modelling, tourist behaviour, rural tourism, tourism demand, dynamic econometric models, economic growth and tourism models. Her contributions have been published in Tourism Management, Tourism Economics, Current Issues in Tourism, International Journal of Tourism Policy and Spanish tourism research journals.

M Teresa Díaz Delfa is an associate professor in Quantitative Methods at the University of Murcia. She teaches statistics and probability at the Faculty of Economics and Business. Her research interests include discrete choice modelling, tourist behaviour, rural tourism and tourism demand. Her contributions have been published in Tourism Management.

References

Albaladejo

Díaz

(2009) Tourist preferences for rural house stays: evidence from discrete choice modelling in Spain. Tourism Management 30: 805–811.

Alegre

Pou

(2016) The length of stay in the demand for tourism. Tourism Management 27: 1343–1355.

Alvarez-Daziano

Bolduc

(2011) Incorporating pro-environmental preferences towards green automobile technologies through a Bayesian hybrid choice model. Transportmetrica 9: 74–106.

Apostolakis

Jaffry

(2005) A choice modelling application for Greek heritage attractions. Journal of Travel Research 43: 309–318.

Araña

Leon

Carballo

, et al. (2016) Designing tourist information offices: the tale of the human factor. Journal of Travel Research 55(6): 764–773.

Ashok

Dillon

Yuan

(2002) Extending discrete choice models to incorporate attitudinal and other latent variables. Journal of Marketing Research 39: 31–46.

Asparouhov

Muthén

(2010) Simple second order chi-square correction. Available at: http://www.statmodel.com/download/WLSMVnewchi21.pdf (accessed 20 March 2020).

Bahamonde-Birke

Ortúzar

(2014a) Is sequential estimation a suitable second best for estimation of hybrid choice models? Transportation Research Record 2429(1): 51–58.

Bahamonde-Birke

Ortúzar

(2014b) On the variability of the hybrid discrete choice models. Transportmetrica A: Transport Science 10(1): 74–88.

10.

Bahamonde-Birke

Kunert

Link

, et al. (2017) About attitudes and perceptions: finding the proper way to consider latent variables in discrete choice models. Transportation 44: 475–493. DOI: 10.1007/s11116-015-9663-5.

11.

Bagozzi

(1988) On the evaluation of structural equation models. Journal of the Academy of Marketing Science 16 (1): 74–94.

12.

Ben-Akiva

Walker

T Bernardino

, et al. (2002) Integration of choice and latent variable models. In: Mahmassani En

(ed.), Perpetual Motion: Travel Behaviour Research Opportunities and Challenges. Pergamon, Amsterdam, pp. 431–470. DOI: 10.1016/B978-008044044-6/50022-X.

13.

Bettman

Luce

Payne

(1998) Constructive consumer choice processes. Journal of Consumer Research 25: 187–217.

14.

Bollen

(1989) Structural Equations with Latent Variables. Chichester: John Wiley.

15.

Chen

Tsai

Chiu

(2017) The decision-making process of and the decisive factors in accommodation choice. Current Issues in Tourism 20(2): 111–119.

16.

Crouch

Louviere

(2000) A review of choice modeling research in tourism, hospitality and leisure. Tourism Analysis 5: 97–104.

17.

Daly

Hess

Patruni

, et al. (2012) Using ordered attitudinal indicators in a latent variable choice model: a study of the impact of security on rail travel behaviour. Transportation 39: 267–297.

18.

Eymann

Ronning

(1997) Microeconometric models of tourists’ destination choice. Regional Science and Urban Economics 27: 735–761.

19.

Epstein

(2003) Cognitive-experiential self-theory of personality. In: Millon

Lerner

(eds) Handbook of Psychology: Personality and Social Psychology, Vol. 5. Hoboken, NJ: Wiley, pp. 159–184.

20.

Espino

Martín

Roman

(2008) Analyzing the effect of preference heterogeneity on willingness to pay for improving service quality in an airline choice context. Transportation Research Part E: Logistics and Transportation Review 44(4): 593–606.

21.

Fornell

Larcker

(1981) Structural equation models with unobservable variables and measurement error. Journal of Marketing Research 18(1): 39–50.

22.

Frochot

(2005) A benefit segmentation of tourists in rural areas: a Scottish perspective. Tourism Management 26: 335–346.

23.

Green

(1984) Hybrid models for conjoint analysis: an expository review. Journal of Marketing Research 21: 155–169.

24.

Greene

(2012) Nlogit Version 5.0 Reference Guide. Plainview, NY: Econometric Software, Inc.

25.

Bentler

(1999) Cutoff criteria for fit indexes in covariance structure analysis: conventional criteria versus new alternatives. Structural Equation Modeling 6(1): 1–55.

26.

Jöreskog

(1994) Structural Equation Modeling with Ordinal Variables Using LISREL (IMS Lecture Notes-Monograph Series 24). DOI: 10.1214/lnms/1215463803.

27.

Jöreskog

Moustaki

(2001) Factor analysis of ordinal variables: a comparison of three approaches. Multivariate Behavioral Research 36: 347–387.

28.

Jung

Yoo

(2016) A study on passengers’ airport choice behavior using hybrid choice model: a case study of Seoul metropolitan area, South Korea. Journal of Air Transport Management 57: 70–79.

29.

Kastenholz

Davis

Paul

(1999) Segmenting tourism in rural areas: the case of North and Central Portugal. Journal of Travel Research 37: 353–363.

30.

Keane

(1997) Modelling heterogeneity and state dependence in consumer choice behaviour. Journal of Business and Economic Statistics 153: 310–327.

31.

Kelly

Haider

Williams

, et al. (2007) Stated preferences of tourists for eco-efficient destination planning options. Tourism Management 28: 377–390.

32.

Kim

Perdue

(2013) The effects of cognitive, affective, and sensory attributes on hotel choice. International Journal of Hospitality Management 35: 246–257.

33.

Kim

Park

(2017) The moderating role of context in the effects of choice attributes on hotel choice: a discrete choice experiment. Tourism Management 63: 439–451.

34.

Laesser

Crouch

(2006) Segmenting markets by travel expenditure patterns: the case of international visitors to Australia. Journal of Travel Research 44: 397–406.

35.

Lane

(1994) What is rural tourism? Journal of Sustainable Tourism 2(1–2): 7–21.

36.

(2016) The performance of ML, DWLS, and ULS estimation with robust corrections in structural equation models with ordinal variables. Psychological Methods 21: 369–387.

37.

Lockyer

(2005) Understanding the dynamics of the hotel accommodation purchase decision. International Journal of Contemporary Hospitality Management 17(6): 481–492.

38.

Maldonado-Hinarejos

Sivakumar

Polak

(2014) Exploring the role of individual attitudes and perceptions in predicting the demand for cycling: a hybrid choice modelling approach. Transportation 41: 1287–1304.

39.

Márquez

Cantillo

Arellana

(2018) Assessing the influence of indicators’ complexity on hybrid discrete choice model estimates. Transportation 47: 373–396.

40.

Martín

Román

Mendoza

(2018) Determinants for sun-and-beach self-catering accommodation selection: a stated preference approach. Tourism Economics. The Business and Finance of Tourism and Recreation 24(3): 319–336.

41.

Masiero

Nicolau

Law

(2015) A demand-driven analysis of tourist accommodation price: a quantile regression of room bookings. International Journal of Hospitality Management 50: 1–8.

42.

Masiero

Pan

Heo

(2016) Asymmetric preference in hotel room choice and implications on revenue management. International Journal of Hospitality Management 56: 18–27.

43.

McFadden

(1974) Conditional logit analysis of qualitative choice behaviour. In: Zarembka

(ed) Frontiers in Econometrics. New York: Academic Press, pp. 105–142.

44.

Merlo

Joao

IDS

(2011) Consumers attribute analysis of economic hotels: an exploratory study. African Journal of Business Management 5: 8410–8416.

45.

Mmopelwa

Kgathi

Molefhe

(2007) Tourists’ perceptions and their willingness to pay for park fees: a case study of self-drive tourists and clients for mobile tour operators in Moremi Game Reserve, Botswana. Tourism Management 28: 1044–1056.

46.

Molera

Albaladejo

(2007) Profiling segments of tourists in rural areas of south-eastern Spain. Tourism Management 28: 757–767.

47.

Morley

(1994) Experimental destination choice analysis. Annals of Tourism Research 21: 780–791.

48.

Motoaki

Daziano

(2015) A hybrid-choice latent-class model for the analysis of the effects of weather on cycling demand. Transportation Research Part A: Policy and Practice 75: 217–230.

49.

Muthén

(1984) A general structural equation model with dichotomous, ordered categorical and continuous latent indicators. Psychometrika 49: 115–132.

50.

Muthén

(1993) Goodness of fit with categorical and non-normal variables. In: Bollen

Long

(eds) Testing Structural Equation Models. Newbury Park, CA: Sage, pp. 205–234.

51.

Muthén

du Toit

SHC

Spisic

(1997) Robust inference using weighted least squares and quadratic estimating equations in latent variable modeling with categorical and continuous outcomes. Available at: https://www.statmodel.com/download/Article_075.pdf (accessed 20 March 2020).

52.

Nunkoo

Ramkissoon

Gursoy

(2013) Use of structural equation modeling in tourism research: past, present, and future. Journal of Travel Research 52(6): 759–771.

53.

Park

Yoon

(2009) Segmentation by motivation in rural tourism: a Korean case study. Tourism Management 30: 99–108.

54.

Palma

Ortúzar

Rizzi

, et al. (2018) Modelling consumers’ heterogeneous preferences: a case study with Chilean wine consumers. Australian Journal of Grape and Wine Research 24: 51–61.

55.

Perles

Ramon

Sevilla

(2011) Determinants of the competitive advantage of residential tourism destinations in Spain. Tourism Economics 17(2): 373–403.

56.

Raveau

Daziano

Yáñez

, et al. (2010) Sequential and simultaneous estimation of hybrid discrete choice models: some new findings. Transportation Research Record Journal of the Transportation Research Board 2156: 131–139.

57.

Román

Martín

(2016) Hotel attributes: asymmetries in guest payments and gains – a stated preference approach. Tourism Management 52(C): 488–497.

58.

Rosseel

(2012) Lavaan: an R package for structural equation modeling. Journal of Statistical Software 48(2): 1–36. Available at: http://www.jstatsoft.org/v48/i02/ (accessed 20 March 2020).

59.

Satorra

Bentler

(1994) Corrections to test statistics and standard errors in covariance structure analysis. In: von Eye

Clogg

(eds) Latent Variable analysis: Applications for developmental research. Sage Publications, Inc, pp. 399–419.

60.

Seddighi

Theocharous

(2002) A model of tourism destination choice: a theoretical and empirical analysis. Tourism Management 23: 475–487.

61.

Sharpley

(2000) The influence of the accommodation sector on tourism development: lessons from Cyprus. International Journal of Hospitality Management 19(3): 275–293.

62.

Soto

Márquez

Macea

(2018) Accounting for attitudes on parking choice: An integrated choice and latent variable approach. Transportation Research Part A: Policy and Practice 111: 65–77.

63.

Stringam

Betsy

Gerdes

John

Vanleeuwen

Dawn

(2010) Assessing the importance and relationships of ratings on user-generated traveler reviews. Journal of Quality Assurance in Hospitality & Tourism 11: 73–92.

64.

Tam

Lam

WHK

(2010) Incorporating passenger perceived utility service quality in airport ground access mode choice model. Transportmetrica 6: 3–17.

65.

Thorhauge

Cherchi

Walker

, et al. (2017) The role of intention as mediator between latent effects and behavior: application of a hybrid choice model to study departure time choices. Transportation 46: 1421–1445.

66.

Train

(2003) Discrete Choice Methods With Simulation. Cambridge: Cambridge University Press.

67.

Vij

Walker

(2016) How, when and why integrated choice and latent variable models are latently useful. Transportation Research Part B: Methodological 90: 192–217.

68.

Victorino

Verma

Plaschka

, et al. (2005) Service innovation and customer choices in the hospitality industry. Managing Service Quality 15: 555–576.

69.

Vredin-Johansson

Heldt

Johansson

(2006) The effects of attitudes and personality traits on mode choice. Transportation Research Part A: Policy and Practice 40: 507–525.

70.

Walker

Ben-Akiva

(2002) Generalized random utility model. Mathematical Social Sciences 43: 303–343.

71.

Yáñez

Raveau

Ortúzar

(2010) Inclusion ol latent variables in Mixed Logit models: Modelling and forecasting. Transportation Research Part A: Policy and Practice 44: 744–753.

72.

Xiao

Zhou

(2019) Small accommodation business growth in rural areas: effects on guest experience and financial performance. International Journal of Hospitality Management 76: 29–38.

73.

Muthén

(2002) Evaluation of Model fit Indices for Latent Variable Models With Categorical and Continuous Outcomes (Technical Report). Los Angeles: University of California, Los Angeles, Graduate School of Education and Information Studies.

74.

Yuan

K-H

Bentler

M P

(1998) Normal theory based test statistics in structural equation modeling. British Journal of Mathematical and Statistical Psychology 51: 289–309.