On Stepwise Discriminant Analyses Applied to Physiologic Data

ABSTRACT Recently widespread interest has arisen in the use of stepwise discriminant function techniques with evoked physiologic data. The purpose of this presentation is to discuss in non‐mathematical terms how such analyses can easily be misused and misinterpreted, pointing out specific pitfalls to be avoided. The various statistical tests of hypotheses used with stepwise discriminant functions, predominantly in the BMDO7M computer program, are discussed in relationship to the law of the iterated logarithm. This law clearly shows how highly significant differences will almost always be found when a stepwise procedure is used even though there may be no real differences between groups. As an empirical illustration, a series of physiologic data (EKG and EEG) was collected under stimulus and no stimulus conditions. Stepwise analyses consistently produced significant differences under no stimulus conditions whereas nonstepwise procedures did not. Under stimulus conditions, stepwise analyses consistently exaggerated the extent of differences detected by nonstepwise analyses.