Métodos de estimação em regressão logística com efeito aleatório: aplicação em germinação de sementes

• Autor(a) principal: Araujo, Gemma Lucia Duboc de
• Data de Publicação: 2012
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: LOCUS Repositório Institucional da UFV
• Texto Completo: http://locus.ufv.br/handle/123456789/4054
• topic: Máxima verossimilhança; Máxima verossimilhança restrita; Aproximação de Laplace; Pinhão manso; Maximum likelihood; Restricted maximum likelihood; Laplace s approximation; Jatropha; CNPQ::CIENCIAS EXATAS E DA TERRA::PROBABILIDADE E ESTATISTICA::ESTATISTICA::REGRESSAO E CORRELACAO

In logistic mixed models with random effect on intercept allows capturing the effects of sources of variation from the particular characteristics of a group (heterogeneity), deflating the pure error and causing a fluctuation in the model intercept. This inclusion brings complexity in estimation methods and also changes the interpretation of the parameters that, originally given by the odds ratio, is then seen from the median odds ratio. The estimation parameters of a mixed model can be made by many different methods with varying performance, as the Laplace s approximation method, maximum likelihood (ML) and restricted maximum likelihood (REML). The objective of this work was to verify in logistic mixed models with random effects on intercept the consequences in interpretation of parameters, in quality of experiment and in classification of treatment via the median odds ratio, and verify the performance of the estimation methods above cited. The analyzes were performed under simulation and after in set of real data from seeds germination experiment of physic nut (Jatropha curcas L.). Considering the logistic mixed model with random effects on intercept, it was verified that the REML estimation method performed better and that the variance of the random effect affects the performance of any of these methods being evaluated inversely proportional. We suggest further studies to determine more properly the influence of the inflexion points and the effective median level in performance methods. In the experiment to evaluate the seeds germination of physic nut involving roll paper, on paper, on sand and between sand substrates, the inclusion of random effects in logistic model showed considerable heterogeneity in seeds germination in different units of the same substrate. The median odds ratio showed the superiority of the substrate between sand over on paper in seeds germination of physic nut, result similar to that obtained by the Tukey s test.