Modelo beta inflacionado de zeros e uns: análise bayesiana de notas no curso de estatística aplicada à engenharia

Abstract

This dissertation discusses an implementation of Bayesian inference on parameters of a Zero and One Inflated Beta Distribution (zoib). Data used to illustrate the proceeding were grades in Applied Statistics from Engeneering Students in second semester of 2015. Algorithms were tested in a simmulation study with different sample sizes and combinaions of parametric values. This study has shown routines are unbiased and consistent. Samples from the joint posterior distribution were drawn using Gibbs Sampler with a taylor made Gaussian random walk to yield better sampling properties for the full conditional distributions of parameters of the continuous part of the model (m and f ). The actual data analysis was carried out in 13 out of the 14 items with zeroes and ones inflation. Analysis can be done routinely to examples alike and can generate usefull insight for researchers in the applied area (educational testing).