Bayesian reciprocal bridge composite tobit quantile regression

Abstract

A composite tobit quantile regression approach is proposed for Bayesian simultaneous covariate selection and estimation in the setting of left censored regression. The proposed approach uses prior distributions for the regression coefficients that are scale mixtures of inverse uniform priors on the coefficients and independent Gamma priors on their mixing parameters. The proposed method was illustrated using simulation examples. Results show that the proposed method performs very well compared to the existing method.