Modeling tourist opinions : from exploration to confirmation

Abstract

The purpose of this dissertation is to model tourist attitudes and opinions through the use of a latent factor model. Our aim shall be facilitated by the study of relevant theories needed for exploratory and confirmatory factor analysis in order to produce a meaningful model. A general factor analytic model is given, followed by an in depth study of parameter estimation, by considering two techniques: the maximum likelihood and the minimum residual. Parceling is subsequently practiced once the unidimensionality of the factor to be parceled is known. Since we are dealing with ordinal data, based on a Likert scale, the polychoric correlations are discussed. The problem of missing data requires statistical attention and is taken care of by studying the multiple imputation method. Hence, a detail of the Expectation Maximization algorithm which is a multiple imputation technique is presented. Finally, the factor structure of a set of observed variables forming part of a tourist "Traveller Survey" is investigated and verified by performing confirmatory factor analysis. A study of model fit is summarised. Since, we are dealing with large samples, problems with the chi-square were evident. As in other reported similar studies in the literature, reasons for this are due to model misspecification and the power of tests used. Hence, these two problems are discussed. One must mention that, other fit indices, do indicate a model which gives an acceptable specification in many ways.