Analyzing the impact of a fatality index on a discrete, interurban mode choice model with latent safety, security, and comfort

Abstract

In discrete choice analysis, travel mode decisions are typically modeled as a function of observable attributes such as time and cost. Other relevant factors that are harder to conceptualize and measure – such as safety and security – are often omitted, producing bias. The leading modeling approach to integrating these subjective factors considers an econometric system of discrete choice with endogenous latent attributes. Whereas a full information estimator has been derived, it is still possible to find work that uses a two-stage estimator with poorer statistical properties. Furthermore, little attention has been devoted to how to forecast with the joint model. In this paper we use data on interurban trips with latent attributes accounting for safety, security, and comfort to compare the full and limited information estimators. In particular, we analyze the effect of alternative normalizations of the measurement scale of the subjective attributes not only for estimation but also for forecasting. In fact, we derive the choice probability elasticity with respect to a marginal change in a fatality index to measure the impact of safety and security on choice. We conclude that normalization does not affect statistical inference when the full information MLE is used. In contrast, the two-stage estimator is biased and produces important differences when different normalizations are used.