Diagnóstico baseado na influência local conforme para os modelos de regressão Birnbaum-Saunders e Senh-Normal.

Abstract

The method of local influence proposed by Cook (1986) is an important tool in the analysis of the joint influence of the observations in the results of a regression fit. This technique has as main goal to evaluate hanges in the results of the analysis when small perturbations are incorporated the model and/or to the data. If these perturbations would cause non proportion effect it can be indication of that the model is badly fitted or that possible departures from the assunptions made for this model an exist. Although the method based on normal curvature proposed by Cook (1986) has been demonstrated to be very useful, this possesses some issues. For example, the normal curvature may take any value and it is not invariant under a uniform hange of scale. In this work, we study the conformal local influence, proposed by Poon & Poon (1999) in regression models, that has as objective to address these issues. More specifically , we apply this technique of diagnostic for log-Birnbaum-Saunders regression models and we derive matrices appropriate to obtain the conformal local influence in the estimated parameters of a more general model, the senh-normal regression model. Finally, we consider empiritical examples with real data to illustrate the theory developed.