Bayesian nonparametric mixed random utility models

We propose a mixed multinomial logit model, with the mixing distribution assigned a general (nonparametric) stick-breaking prior. We present a Markov chain Monte Carlo (MCMC) algorithm to sample and estimate the posterior distribution of the model’s parameters. The algorithm relies on a Gibbs (slice) sampler that is useful for Bayesian nonparametric (infinite-dimensional) models. The model and algorithm are illustrated through the analysis of real data involving 10 choice alternatives, and we prove the posterior consistency of the model.