Análise de diagnóstico em modelos de regressão normal e logística

Abstract

Presented some techniques for analyzing the diagnostic quality of fit in regression models and logistic normal. Gave special attention to the method of local influence. This method was initially proposed by Cook (1986) and has shown how powerful analysis tool for diagnosis. The proposal to Cook is to assess the normal curvature of a surface based on the likelihood measure displacentement under a small pertubation in the model presented the approach of Billor and Loynes (1993), who applied a measure, the expulsion likelihood, whose first derivative is not annihilated, except in trivial cases, thus the maximum slope is used as a measure of local influence. Developed the approach Billor and Loynes (1993), used only in the normal model, for the logistic model. Presented two applications in wich techniques were used to analyze the diagnostic of quality of fit, through graphics and the two proposed local influence. In the case of simple linear normal model, the maximum normal curvature and slope showed is the same sensitivity in the identification of influential observations.