A Nonparametric Bayesian Method To Estimate Ordered Probit Model

In this paper, we introduce a random effect variable into the Ordered Probit model and develop a Bayesian Nonparametric approach to deal with it based on Bayesian theory. In order to describe the heterogeneity in real data, we assume the random effect comes from a Dirichlet Process prior and deduce the posterior distribution. Meanwhile, we implement the Bayesian Nonparametric approach in Stata by MCMC algorithm, supplementing the estimate methods for Ordered Probit model in this statistic software. Two simulation are designed to compare our method with the other two estimate methods (maximum likelihood and semi-nonparametric) in Stata. The test results show that our method works better when there is heterogeneity such as bimodal and abnormal. In addition, an empirical analysis is done about the influence factors of job satisfaction using BHPS survey data. The results show that job satisfaction is positively related with absolute income, negatively related with comparison income and it has U shaped age profile.