Identifying the Limitation of Stepwise Selection for Variable Selection in Regression Analysis

In application, one major difficulty a researcher may face in fitting a multiple regression is the problem of selecting significant relevant variables, especially when there are many independent variables to select from as well as having in mind the principle of parsimony; a comparative study of the limitation of stepwise selection for selecting variables in multiple regression analysis was carried out. Regression analysis in its bi-variate and multiple cases and stepwise selection (forward selection, backward elimination and stepwise selection) was employed for this study comparing the zero-order correlations and Beta (β) weights to give a clearer picture of the limitation of stepwise selection. Subsequently, from the comparisons, it was evident that including the suspected predictor (suppressor) variable that was not significant in the bi-variate case as suggested by the stepwise selection improved the beta weight of other predictors in the model and the overall predictability of the model as argued.