Publication Details

A note on controlling significance levels in stepwise regression

Type: article

Author(s): Kupper, L. L.; Stewart, J. Richard; Williams, K. A.

Pages: 13-15

Url: http://dx.doi.org/10.1093/oxfordjournals.aje.a112200

Publication Date: Jan-1976

Journal: American Journal of Epidemiology

Volume: 103

Issue: 1

Doi: 10.1093/oxfordjournals.aje.a112200

Pmid: 1247008

Abstract: Stepwise regression is being used more and more as an exploratory data analysis tool in the health and social sciences. While it may be generally recognized that the probability of finding at least one significant regressor variable when actually there are none increases as the number of possible regressor variables increases, the actual extent of this increase may not be fully appreciated. This note investigates the relationship between this probability (the true significance level α*) and the so-called nominal significance level of α which is typically used to control entry into the regression equation of one of k candidate independent variables.