A failure to replicate the Bem and Allen study of individual differences in cross-situational consistency

Bem and Allen (1974) purportedly found evidence that, by using self-report measures of cross-situational consistency as moderator variables, it was possible to substantially increase the size of correlation coefficients computed among measures of each of two personality traits. The present study was undertaken to (a) replicate the Bem and Allen finding on a larger set of personality traits, (b) determine if the results hold differentially for self-report, other-report, and objective personality measures, and (c) compare different methods for dividing subjects into high and low consistent groups. One hundred twelve subjects (64 men and 48 women) were divided into high and low consistent groups using three different methods. Within each group correlations were computed among and between 6 self-report measures, 10 other-report measures, and when possible a few objective measures, for each of eight traits. In general, none of the methods of consistency classification replicated the Bem and Allen finding. In cases where the high consistent group had a larger average correlation among the measures than their low consistent counterparts, the size of that difference was negligible. Moreover, there were about as many instances in which the low consistent group actually had larger average correlations. Finally, there was essentially no agreement among the various consistency classification methods in dividing subjects into high and low consistent subsamples.