This thesis develops a new estimation method for the multinomial logit model, which is applied in modeling choice probability in marketing and econometrics. Based on the connection of the mixture model and the logit model capturing both parameter and consideration set heterogeneity, we examine a particular sampling method of Bayesian estimation, so-called the Gibbs sampler weighted Chinese restaurant process, to evaluate posterior quantities. In the posterior analysis of the simulation algorithm, consideration set probabilities and parameter estimates of the logit model are our main concern. The simulation study indicates that the parameter estimates depend on the choice of prior parameters. Nevertheless, the resulting estimates of posterior probabilities is consistent with those in the...[ Read more ]

This thesis develops a new estimation method for the multinomial logit model, which is applied in modeling choice probability in marketing and econometrics. Based on the connection of the mixture model and the logit model capturing both parameter and consideration set heterogeneity, we examine a particular sampling method of Bayesian estimation, so-called the Gibbs sampler weighted Chinese restaurant process, to evaluate posterior quantities. In the posterior analysis of the simulation algorithm, consideration set probabilities and parameter estimates of the logit model are our main concern. The simulation study indicates that the parameter estimates depend on the choice of prior parameters. Nevertheless, the resulting estimates of posterior probabilities is consistent with those in the literature.