Presença de dados missing em modelos de regressão logística

Abstract

In this work we present a detailed study of the logistic regression model with missing data in the independent variables. Several techniques are considered such as Complete Case, Mean Imputation and Corrected Complete Case. We present a new estimator, denoted EMVGM, given by the combination between the Complete Case estimator and the ML-estimator with the use of Gaussian quadrature. A simulation study is carried out to evaluate the performance of the ML-estimators obtained in each technique above mentioned. In general, the alternative estimador, EMVGM, presents a better performance taking into account the variance, the bias and the mean quadratic error.