BAYES INFERENCE IN THE TOBIT CENSORED REGRESSION-MODEL

We consider the Bayes estimation of the Tobit censored regression model with normally distributed errors. A simple condition for the existence of posterior moments is provided. Suitable versions of Monte Carlo procedures based on symmetric multivariate-t distributions, and Laplacian approximations in a certain parameterization, are developed and illustrated. Ideas involving data augmentation and Gibbs sampling [cf. Tanner and Wong (1987) and Gelfand and Smith (1990)] are also developed. The methods are compared in two examples with diffuse priors, and various combinations of sample sizes and degrees of censoring.