Análises Bayesiana para o modelo de regressão Birnbaum-Saunders com zeros ajustados

Abstract

Modeling based on the Birnbaum-Saunders distribution has received considerable attention in recent years. In this work we consider the reparametrized Birnbaum-Saunders distribution with zero-adjusted (ZARBS) (SANTOS-NETO et al., 2012). This distribution admits the occurrence of zeros with positive probability, considering a discretecontinuous mixing model that is constructed using a probability mass at zero and a continuous component. ZARBS generalizes at least seven reparametrized Birnbaum- Saunders regression models. In this context, the main contribution of this dissertation is to study ZARBS under a Bayesian approach using the BAMLSS package developed in the R software, as well as to derive influence diagnoses for the model. Diagnostic methods have been important tools in regression analysis to detect anomalies, such as breaking assumptions in the stochastic part of the model, presence of outliers and influential observations. We assess the local influence on parameter estimates considering a perturbation scheme. To verify the potential of the proposed methodology, an application to a set of real data will be considered.