Integrating psychometric indicators in latent class choice models

Abstract

Latent class models are a convenient and intuitive way to introduce taste heterogeneity in discrete choice models by relating attributes of the decision makers with unobserved behavioral classes, hence allowing for a more accurate market segmentation. Estimation and specification of latent class models can be improved with the use of psychometric indicators that measure the effect of unobserved attributes in the individual preferences. This paper proposes a method to introduce these additional indicators in the specification of integrated latent class and discrete choice models, through the definition of measurement equations that relate the indicators to attributes of the decision maker. The method is implemented for two mode-choice case studies and compared with alternative methods to introduce indicators. Results show that the proposed method generates significantly different estimates for the class and choice models and provide additional insight into the behavior of each class.