Date of Award
Doctor of Philosophy in Psychology
Charles E. Collyer
The statistical power of psychological research was examined in three studies. The purpose of Study I was to provide a general assessment of statistical power in psychological research. All of the studies using statistical tests in the 1982 issues of the Journal of Abnormal Psychology and the Journal of Personality and Social Psychology (N = 142) were evaluated and compared to the results of a previous survey. Results indicated that the power of psychological research has increased over the past 20 years but that power must still be characterized as low. These results are limited because accurate estimates of effect size, on which power analysis must be based, are not given in most research reports. The purpose of Studies II and III was to use recently devised techniques of meta-analysis to provide estimates of effect size for subsequent power analyses. Use of this procedure allowed power analyses to be conducted using precise estimates of effect size, rather than a wide range of effect sizes, as in previous power surveys. In study II, a meta-analysis of 48 studies on the spontaneous recovery of verbal learning was conducted. This area of e search was selected because previous studies of spontaneous recovery have produced conflicting results concerning the existence of the phenomenon. Results indicated that the evidence for spontaneous recovery is statistically significant, but that the magnitude of the effect is small. A power analysis was then performed on the spontaneous recovery studies using the effect size information obtained by the meta-analysis. Results indicated that the power of research on spontaneous recovery was low, and implied that contradictory findings may be an artifact of low statistical power. This implication was examined in Study III. Power analyses were conducted on each of 12 published meta-analyses, all of which had been performed on research literatures characterized by inconsistent results. The results of Study II were generally confirmed. For six of the meta-analyses statistical power was similar to the proportion of significant results. In only three of the meta-analyses was power relatively high (> .75), while for the remaining meta-analyses power was very low (< .55). The low power of these research areas is due to the small size of the effects that were investigated. Average effect size (eta2) across all meta-analyses was only .019. For effects of this magnitude only very large sample sizes will yield acceptable levels of statistical power (e.g., .80). It is recommended, therefore, that psychologists increase power by making effect sizes larger, that is by increasing the signal-to-noise ratio of research designs. Ultimately desirable is the construction of "strong" theories, which would enable researchers to make specific predictions concerning the magnitudes of the effects they study.
Rossi, Joseph S., "Statistical Power of Psychological Research: The Artifactual Basis of Controversial Results" (1984). Open Access Dissertations. Paper 962.