Seleção de variáveis via Backward em Modelo Linear Normal Assimétrico

Abstract

This work aims to select variables using the backward method, via Akaike Information Criterion (AIC), in a linear regression model with errors following a Skew normal distribution. We investigated the effect of multicollinearity of the explanatory variables and the lack of correlation between the explanatory variables dependent on the fit of the final model (parsimony) process variable. The purpose of this research is to verify the quality of the selection method proposed in this monograph. For this, we investigated four strategies for selecting the final model. The first strategy is to investigate the goodness of fit of the final model without using any prior exploratory analysis variable selection using the backward method, via AIC. In the second strategy will be phased out problems of lack of correlation between the dependent variable and the explanatory variables before using the selection method. In the third strategy will be removed problems of multicollinearity of the explanatory variables before using the method studied. Finally, the fourth strategy will be removed the two problems mentioned above before selection. A simulation study is conducted to demonstrate the best strategy for model selection. Finally, the proposal of this monograph is applied to a real data set.