The Abuse of Stepwise Regression, or Experiments in Fishing.
Briefly discusses the inappropriate evaluation of stepwise regression results. The author employs a simple argument revealing the misleading significance levels that are customarily reported when using a stepwise regression program. Using such a program and applying standard statistical tests, experience has indicated that random numbers can produce, in the majority of cases, significant correlations at the 0.05 level. Until the development of better test statistics, a more prudent use of stepwise regression is suggested. 7 pp.