ABSTRACT
Unobserved heterogeneity is often captured with random parameters approaches, but such approaches require a distributional assumption (reflecting how the unobserved heterogeneity affected parameters across observations). An alternate way to account for unobserved heterogeneity across observations is to search for groups of observations that share the same unobserved heterogeneity. These groups (or classes) of observations are not known to the analyst and hence these models are often referred to as latent class models or, equivalently, finite mixture models. Unlike random parameters models, which allow each observation to have its own parameter and require a distributional assumption of parameters across observations as mentioned above, latent class models are non-parametric with regard to unobserved heterogeneity, in the sense that no distributional assumption with regard to parameters is needed (although parametric assumptions are still typically made with regard to the model form itself). This chapter presents the derivation and logic behind latent class/finite mixture models and provides examples that show how such models are estimated and their results interpreted. Latent class/finite mixture models have many applications in transportation data analysis, and this chapter provides the fundamentals behind these models.
