Influence Diagnostics in Linear Censored Regression Models with Skew Scale Mixtures of Normal Distributions

• Ano de defesa: 2024
• Autor(a) principal: Guzman, Daniel Camilo Fuentes
• 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: Algorithm; Algoritmo; Assimetria; Censoring; Censura; Diagnóstico; Diagnostics; Distribuições; Distributions; EM; Influence; Influência; Linear; Mean-Shift; Misturas de escala; Mixtures of normal; Modelos; Models; Normais; Outlier; Regressão; Regression; Scale; Skew; SSMN
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/104/104131/tde-30072024-110832/

In this research, we conducted studies on local and global influence diagnostics for Censored Linear Regression Models with Skew Scale Mixtures of Normal Distributions (SSMN-CR), proposed by Guzman, Ferreira and Zeller (2020). Initially, we discussed methods for generating censored data, specifically presenting methods to generate randomly censored data with both unilateral and interval censoring. Subsequently, we addressed case deletion and local influence diagnostics based on the Q function, inspired by the findings of Zhu et al. (2001) and Zhu and Lee (2001). To analyze the sensitivity of the maximum likelihood estimators of the SSMN-CR model parameters to small perturbations in assumptions and/or data, we considered various perturbation schemes, such as case weighting, explanatory variables, response variables, and perturbations in scale and skewness parameters. To illustrate the usefulness of the proposed methodology, we presented the analysis of a real dataset and three simulation studies.