A proposal for examining and reporting stepwise regressions

Abstract

A common mode of regression analysis is a stepwise regression. One of the problems associated with the stepwise algorithm is that the data analyst may wrongly identify the selected variables as the important variables. We propose a simple table as an aid in examining and reporting stepwise regression analyses. The key to our proposal is recognition of 'close alternatives', that is, those variables excluded from selection by the inclusion of another variable. This presentation should encourage the analyst and reader to think in terms of sets of variables that, together, represent factors associated with the outcome.