Establishing behavioral correlates: The MMPI as a case study

Abstract

Two methods have been used to determine MMPI behavioral correlates: the whole-sample and split-sample techniques. The latter technique, which splits a sample into half samples, is currently in use because it ostensibly controls more adequately for Type I errors by requiring significance in both half samples. By simply adjusting the level of significance for the whole-sample approach, however, it can control for Type I errors as well as the split-sample technique. Furthermore, the whole-sample approach appears to control more adequately for Type II errors than the split-sample technique. Finally, data are presented suggesting that as an indicator of the average result obtained by repeated random splittings of a sample, results from the whole-sample approach are preferable to those from a single split of a sample. The whole-sample method was thus recommended; however, the probability of a Type I error for each correlate tested should be set at values smaller than those currently used.