Monte Carlo simulation methods are used to confirm the identifiability of discrete choice models in which unobserved heterogeneity is specified as a random effect and modelled using the nonparametric mass-points approach. This simulation analysis is also used to examine alternative strategies for the estimation of such models by using a quasi-Newton maximum-likelihood estimation procedure, given the apparent sensitivity of model identification to choice of starting values. A mass-point model approach is then applied to a dataset of repeated choice involving household shopping trips between three types of retail centre, and the results from this approach are compared with those obtained from a conventional cross-sectional multinomial logit choice model as well as to results from a model in which a parametric distribution (the Dirichlet) is used to model the unobserved heterogeneity.
Linear IV Regression Estimators for Structural Dynamic Discrete Choice Models