Comparação entre algoritmos MCMC para Inferência Bayesiana em modelos dicotômicos da TRI

• Ano de defesa: 2019
• Autor(a) principal: Gabriel Oliveira Assunção
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; Programa de Pós-Graduação em Estatística; UFMG
• Palavras-chave em Português: Estatística; Teoria de resposta ao item; Teoria bayesiana de decisão estatística; Markov, Processos de
• Link de acesso: http://hdl.handle.net/1843/36648

This study aims to compare MCMC algorithms for Bayesian Inference in the 3-parameter TRI model. We consider four different algorithms already proposed inthe literature, which differ basically in relation to the use of auxiliary variables.The main objective is to investigate which algorithm is computationally moreefficient to return a sample of the (same) distribution to textit posteriori. Thecomparison is made based on computational time and effective sample size ofrelevant statistics. The comparison is made in different scenarios with respect tosample size (including items). Through it, one can see that the performance ofthe algorithms varies as the sample size increases. An extension of the Gonçalveset al. (2018) algorithm for the 4-parameter model is also presented and appliedto an Enem database.