Culture-sensitive tourists are more price insensitive

Abstract

The purpose of this article is to analyze the effect of the cultural interest manifested by tourists when planning a vacation on their sensitivity to price. The proposed hypothesis states that tourist price sensitivity is moderated, at the moment of choosing a destination, by cultural interest. For this purpose, we measure and identify tourists’ price sensitivities—individual by individual—from real choices, i.e., tourist price sensitivity is estimated for each individual by observing the destination she actually selects. The empirical application is carried out on a sample of 2,127 individuals, and the operative formalization used to estimate individual price sensitivities follows a Random-Coefficient Logit Model; and to detect the way these sensitivities relate to the search for culture, an ANOVA procedure is employed. The results show an incremental effect of cultural interest on tourist price insensitivity; i.e., people looking for culture find their price sensitivity moderated by this interest in such a way that the negative effect of price diminishes. Also, we further explore these culture-interested tourists by a segmentation analysis, identifying five segments with different price sensitivities—one of them even showing certain high-price proneness.

Appendix: Applying the Bayesian procedure to this empirical application

Following Train (2003), the likelihood L of observed choice y _n for an individual n conditional on parameters b and W (average and variance of β _n , respectively) is expressed as:

$$ L(y_{n} /b,W) = {\frac{{{\text{e}}^{{X_{n} \beta_{n} }} }}{{\sum\nolimits_{j = 1}^{J} {{\text{e}}^{{X_{n} \beta_{n} }} } }}}\varphi (\beta_{n} /b,W) $$

where ϕ is the function of Normal distribution.

Let k(b,W) be the prior distribution of parameters b and W. In general, it is assumed that b has a Normal distribution and W an Inverted Gamma distribution (or Inverted Wishart distribution in the case of multi-variation) of type f(W) = W ^−(v+1)/2e^−vs/2W with v being the degrees of freedom and s a parameter of scale to be estimated. Bayes’ rule allows the analyst to obtain the posterior distribution K(b,W,β _n /Y) for the group of choices Y of the sample individuals (n = 1,…,N) as:

$$ K(b,W,\beta_{n} /Y) \propto \prod\limits_{n = 1}^{N} {L(y_{n} /b,W)k(b,W)} $$

The posterior distribution has three parameter types to estimate θ = {b,W,β _n }: the average b, the variance W, and the parameters of each individual β _n , from which we obtain the conditional indirect utility functions of each individual and, therefore, the preference structure. The estimation of the parameters is obtained through the following expression

$$ \hat{\theta } = \int\limits_{\theta } {\theta K(\theta /Y){\text{d}}\theta } $$

This integral has no closed solution, which leads the researcher to use a procedure of estimation by simulation. Therefore, θ is estimated as the average of the simulated drawings. However, the posterior distribution K(θ/Y) does not always take the form of a known distribution from which one could immediately take draws. Train (2001a), in the case of choice models, suggests the use of Monte Carlo Markov Chains; specifically, the sample simulation algorithms of Gibbs and Metropolis-Hasting for the draws of the density function (the parameter estimates for the model are based on 14,000 draws obtained after discarding the first 4,000 iterations (which are used for burn-in) and the prior values for parameters comes from the maximum likelihood sample estimates). Train (2001b) also demonstrates that the estimator of the simulated average of the posterior distribution is consistent, asymptomatically normal, and equivalent to the estimator of maximum likelihood.