Reducing the randomness of latent variables using the evaluative space grid: Implementation in a hybrid choice model

Abstract

The study of latent variables, and in particular of attitudes, contributes to a better understanding of individual preferences and behavior and it is now common practice within transportation literature. However, the procedure of attitude measurement is still not optimal. Two major issues are the misspecification of the attitude itself and the number of suitable items used for defining the psychological factor. The incorrect measurement entails a poor representation of individuals on the latent continuum and a less precise definition of the latent variable itself.These issues become even more relevant when a Likert scale is used. Indeed, the neutral point of this scale is selected by both individuals having an ambivalent and an indifferent attitude, and the poor representation makes impossible to distinguish these categories. Nevertheless, such a distinction can be very profitable for policy reasons. To overcome this issue and to suggest more effective policies, we propose using the Evaluative Space Grid (ESG), which is a single-item measure of positivity and negativity, to collect attitudinal variables. This tool can distinguish between individuals with indifferent and ambivalent attitudes, as well as those with positive and negative inclinations.This paper models the ESG using a pair of ordered logit regressions and suggests a procedure to include this approach in the framework of hybrid choice models. Furthermore, it endeavors to shed light on the preferences of individuals having indifferent and ambivalent inclinations in a transportation context, showing the hypothesis that their preferences are different for commuting trips.