Bayesian generalized errors in variables (GEIV) models for censored regressions

Errors in variables models in linear regression are an old and important theoretical topic, but rather neglected in applied statistics. This is especially true for Bayesian statistics where the numerical diiculties for the estimation of the models are even greater. Using a Bayesian approach and modern numerical integration techniques (the Gibbs sampler) we derive the necessary full conditional distributions for the simulation procedure to obtain the complete posterior distribution. The multivariate errors in variable model is considered with a correlation structure between dependent and independent variables. Furthermore, we show how this GEIV approach can be extended to a censored regression (the Tobit GEIV) model.