Evaluating mixed models of discrete response using standardized visual predictive check

• Ano de defesa: 2019
• Autor(a) principal: Avena, Lina Maria Acosta
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Discrepâncias preditivas; Discrete response; Mixed models; Modelos mistos; Predictive discrepancies; Resposta discreta; Standardized visual predictive check; Verificação visual preditiva padronizada
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/45/45133/tde-20082025-211306/

Logistic linear mixed model, Poisson and negative binomial log linear mixed models are increasingly used to analyze longitudinal discrete data that arising in many areas of applied science. Despite their popularity, these models involve analytical challenges that lead to a series of implications that may compromise inferential conclusions. This work aims to provide a tool that allows to assessment the goodness of _t of the mixed models of discrete response, essentially, to allow us to perform this task in the practical applications. Its methodology involves the use of the predictive capabilities of the model, so it covers some of the inferential objectives when regression models are used. In this approach with Standardized visual predictive check (svpc), the inspections are performed visually over time or any other covariate, which represents an important potential for use in relation to the overall diagnosis of a model. We investigate the performance of our proposal through simulations studies and four data sets from di_erent areas of the science. The procedures were implemented of the computing environment R.