Modeling vehicle-miles of travel accounting for latent heterogeneity

Vehicle use is associated with negative externalities such as traffic congestion, air pollution, and greenhouse gas emissions. Particularly in the U.S. as a car-oriented country, vehicle use — in terms of vehicle-miles of travel (VMT) — has been on the rise and is projected to increase in the future. To curb the VMT growth and mitigate the associated externalities, policy makers can design informed strategies based on VMT predicted by vehicle use models. However, traditional vehicle use models capture merely the observed heterogeneity across vehicle decision making units (e.g., individuals) and ignore the latent or taste heterogeneity sourced in individuals' attitudes and lifestyle preferences, which may cause biased and inconsistent results that mislead implications for policy makers. To address this research gap, the present study introduces a latent class regression model, where a probabilistic multinomial logit component endogenously classifies a sample of vehicle use observations so as to be homogeneous within and heterogeneous across the classes with respect to VMT. At the same time, a finite set of linear regression equations in the number of the latent classes yields class-specific VMT. The model is estimated on a sample dataset from the State of California identifying three latent classes, verifying the hypothesis of positing vehicle use on both observed and unobserved heterogeneity. The estimation results are analyzed to infer implications of potential policies aiming at reducing VMT through increasing fuel cost and switching to telework, and to evaluate the efficiency of resource allocation to policies by targeting different classes with distinctive characteristics.