Bayesian empirical likelihood and variable selection for censored linear model with applications to acute myelogenous leukemia data

This paper develops the Bayesian empirical likelihood (BEL) method and the BEL variable selection for linear regression models with censored data. Empirical likelihood is a multivariate analysis tool that has been widely applied to many fields such as biomedical and social sciences. By introducing two special priors to the empirical likelihood function, we find two obvious superiorities of the BEL methods, that is (i) more precise coverage probabilities of the BEL credible region and (ii) higher accuracy and correct identification rate of the BEL model selection using an hierarchical Bayesian model, vs. some current methods such as the LASSO, ALASSO and SCAD. The numerical simulations and empirical analysis of two data examples show strong competitiveness of the proposed method.