Comparação das funções de ligação logit e probit em regressão binária considerando diferentes tamanhos amostrais

• Ano de defesa: 2013
• Autor(a) principal: Freitas, Leillimar dos Reis
• Orientador(a): Martins Filho, Sebastião
• Banca de defesa: Almeida, Fernanda Maria de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Viçosa
• Programa de Pós-Graduação: Mestrado em Estatística Aplicada e Biometria
• Departamento: Estatística Aplicada e Biometria
• País: BR
• Palavras-chave em Português: Regressão logística; Regressão probit; Simulação de dados
• Palavras-chave em Inglês: Logistic regression; Probit regression; Data simulation
• Área do conhecimento CNPq: CNPQ::CIENCIAS EXATAS E DA TERRA::PROBABILIDADE E ESTATISTICA::ESTATISTICA
• Link de acesso: http://locus.ufv.br/handle/123456789/4062

It was considered a binary regression analysis with the logit and probit link function in order to verify the link functions robustness in sample size variation. These link functions apply, respectively, the cumulative distributions logistics and normal, and the probabilistic main difference values of independent variable extremes. Then, simulations were performed with 500 replicates using 10 different sizes samples, from 10 to 91, with 9 successively units between the samples. Performance convergence percentage measures, general probabilistic average squared error, specific probabilistic average squared error and coefficients Wald test were used to establish a specific use recommendation for the two different link functions just when data were generated with logit and probit use and analyzed with the both link functions in each sample size. It was concluded that the work aim was achieved by establishing a recommendation for the logit link function use for sizes below 20 due to higher convergence rate, ie, it was verified with logit link function that there is a greater number of samples which was possible to estimate the binary regression parameters. For larger sample sizes, using other performance measures, both, the logit and probit, were similar, as there were no significant differences between these two different functions.