Estimação Bayesiana em modelos de regressão T de student com erros nas variáveis, respostas multivariadas e censuras

Abstract

We propose an extension of the usual normal regression model where both the vector of responses and the covariate are possibly censored. We assume that the jointly distribution of covariate and errors is Student-t, which is an alternative to the normal distribution, but with heavy tails. A Gibbs-type algorithm is proposed to carry out Bayesian estimation of the parameters in the model. Three simulation studies are conducted, showing that the proposed model is more flexible than the normal one when fitting data with censoring pattern and heavy tails, in addition to an application with real data.