Model Selection in Logistic Regression and Performance of its Predic

Logistic regression studies often have several covariates and asked to cull these covariates to arrive at a parsimonious model. The goal is to maximize predictive power while minimizing the number of covariates in the model. Purposeful selection of covariates does not provide efficient model in case of large number of covariates while mechanical stepwise and best subsets selection procedures still provide a useful and effective model selection tools. Even with moderate number of covariates, stepwise method allows to decrease drastically the total number of models under consideration and to produce the final model on statistical ground. In spite of criticism, stepwise logistic regression has been widely used. Best subsets approach identifies key subsets of covariates on the basis of information criteria and provides predictive model. It is evident that with reasonable number of covariates, best subsets approach is a superior alternative to stepwise logistic regression to opt the predictive model.