Effect sizes, confidence intervals, and confidence intervals for effect sizes

摘要

The present article provides a primer on (a) effect sizes, (b) confidence intervals, and (c) confidence intervals for effect sizes. Additionally, various admonitions for reformed statistical practice are presented. For example, a very important implication of the realization that there are dozens of effect size statistics is that authors must explicitly tell readers what effect sizes they are reporting. With respect to confidence intervals, when interpreting a 95% interval, we should never say that we are 95% confident that our interval captures the estimated population parameter. It is explained that effect sizes should be reported even for statistically nonsignificant effects. And, most importantly of all, it is emphasized that effect sizes should not be interpreted using Cohen's benchmarks. Instead, we ought to interpret our effects in direct and explicit comparison against the effects in the related prior literature. (c) 2007 Wiley Periodicals, Inc.