Title: Bayesian inference in censored linear regression model with AR(p) errors
Authors: Rodney Sousa - Universidade de Aveiro (Portugal) [presenting]
Abstract: Censored linear regression models are a class of models which allow the dependent variable to be censored. The problem of estimating a censored linear regression model with autocorrelated errors, CLR-AR, may arise in many environmental and social studies. The likelihood function for this model is very complex, hindering the use of Maximum Likelihood methods in real problems, especially when the number of censored observations in the sample is large. We consider a Bayesian approach to estimate the CLR-AR. We propose a Gibbs sampler with Data Augmentation algorithm in which each augmented datum is, in fact, the mean of multiple simulated values, GDA-MMS. The performance of the algorithm is analyzed in a simulation study. The results indicate that the estimates show good accuracy and Bayesian consistency, even in scenarios where the proportion of censored observations is large. The approach is also illustrated in an empirical dataset regarding cloud ceiling height, where 41\% of the observations are censored.