On Heterogeneity in Discrete Choice Modeling for Transportation

We discuss methodological advances in modeling user heterogeneity within the discrete choice framework and their implications for transportation research. Beyond the conventional dichotomy of observable and latent heterogeneity, we argue for a third dimension, decision-mechanism heterogeneity, in which population segments differ in the decision rule that maps attributes into a choice rather than in the value of their taste parameters. We make this precise through a perturbed-utility representation: observable and latent heterogeneity vary the taste parameters under a fixed perturbation function, whereas decision-mechanism heterogeneity is heterogeneity in the perturbation function itself, which gives an operational and falsifiable distinction. We survey four methodological strands shaping the field, namely classical parametric models, distribution-free choice modeling, neural-embedded discrete choice models, and tree-structured choice attribution, and synthesize them in a comparison table organized around this axis. We then set out three future research directions framed within the data-driven turn now reshaping transportation research, in which the traditional specify-then-estimate approach is giving way to a learn-from-data paradigm. Transportation, with its multimodal, electrified, and emerging air-mobility complexity, is the natural testbed for these methodological advances.