UTILIZACIÓN DEL BMA EN EL MODELO DE REGRESIÓN LOGÍSTICA Y SU COMPARACIÓN CON OTROS CRITERIOS DE SELECCIÓN DE MODELOS

Abstract

The identification of risk factors associated with a disease is an important part on the model construction for public health. The investigators frequently use Logistic Regression (LR) as a statistical tool to achieve such end. LR it is commonly used with stepwise selection methods to select risk factors associated with a particular disease. Nonetheless stepwise methods for model selection tend to select over estimated models. This happens because standard selection methods don’t take into account the inherent uncertainty on model selection. A criteria that takes into account the model selection uncertainty is the Bayesian Model Average (BMA) recently applied on LR for case and study controls. This paper presents the application utility of the BMA criteria and illustrates its use in a study about the presence of preeclampsia during pregnancy on Guerrero Mexico, said application is performed on the R software it also compares the results obtained with other selection criteria.