How to categorize individuals on the basis of underlying attitudes? A discussion on latent variables, latent classes and hybrid choice models

Abstract

Classifying individuals into distinctive groups in discrete choice modelling work has become a common procedure since latent class models were introduced in the field. However, latent classes have certain shortcomings regarding the interpretation of the identified classes. On the other hand, hybrid choice models incorporating latent variables and psychometric data, are a powerful tool to treat and identify some underlying attitudes affecting behaviour; however, the treatment of the latent variables into the utility function has not been analysed in sufficient depth. Latent variables accounting for attitudes resemble socio-economic characteristics and, therefore, both systematic-taste-variations and categorizations may also be considered. We examine different ways to categorize individuals based on latent characteristics, and explain why this may be convenient. In particular, we propose a direct categorization of individuals based on underlying latent variables and conduct theoretical analyses contrasting this method with existing approaches. Based on this analysis, we conclude that some of the methods used in the past exhibit certain shortcomings that can be overcome by relying on a direct categorization. Then, we show some of the analytical advantages of the approach with the aid of two illustrative examples. Furthermore, the proposed approach allows bridging the gap between latent classes and latent variable models.