Discrete choice models with latent variables using subjective data

This chapter proposes a method for incorporating latent qualitative factors in travel-demand models. The framework for the proposed method is composed of a linear structural equation model and a discrete choice model. The linear structural equation model describes the process of the latent attributes generating subjective psychometric indicators of various aspects of travel attributes, and the relationship between the latent attributes and the observable objective variables. The discrete choice model expresses the observed choices explained by observable variables and the latent attributes. A practical calibration method of unknown parameters included in this system employs a two-step sequential maximum-likelihood estimation. First, a LISREL like program estimates the structural equation model to calculate the fitted values of the latent variables. Then, a discrete choice model, such as legit or probit with the calculated latent explanatory variables, is estimated. The practicality of the proposed method is demonstrated by an empirical analysis. The case study uses survey data of inter-city travel mode choice between rail and car. The data include six subjective indicators of travel attributes from which two latent attributes, ride comfort and convenience, are identified. The fitted values of these two latent attributes had outstanding explanatory power in the mode choice model. The proposed method is characterised by constructing the latent variables using objective variables, such as travel attributes and socioeconomic variables. This implies that the method can be used for predicting demand in conjunction with policy changes, because the predicted values of the latent variables can be calculated from objective variables.