Statistics and replication experiments: Comparing interaction effects

A method is presented that allows the comparison and combination of interaction effects obtained in two or more experiments. Using an analysis of variance model, we first evaluated the interaction effects found in experiments conducted by Alloy and Abramson (1979, Experiment 2) and Bryson, Doan, and Pasquali (1984). Both experiments examined the same independent (level of depression and outcome frequency) and dependent (judgment of control) variables. Whereas Alloy and Abramson reported a significant depression ×outcome frequency interaction for female subjects, Bryson et al. failed to find any significant interaction effects. The latter authors challenged the replicability of the Alloy and Abramson results. Our reanalyses of the results from these two studies indicated that, for females, Bryson et al. produced an interaction effect not different from that of Alloy and Abramson. Our next task was to compare the conclusions reached from our analytical approach with those based on standard meta-analytic methods (Rosenthal, 1984). Finally, we evaluated the general utility of the two analytical approaches and made some general comments about replication experiments.We thank Margaret Murray for her contributions to early stages of this project. The comments and advice of Rebecca Warner, Paul Sweeney, William Stine, Carolyn Mebert, Ellen Cohn, Vernon Padgett, David Sugarman, and several anonymous reviewers are appreciated.