An Average Power Criterion for Sample Size Estimation

This is the authors' abstract. We don't add key points for this paper.

The use of an empirical estimate of effect size in the standard sample size formula introduces variability in the power to detect the true underlying effect. A useful criterion for judging the performance of a sample size estimation procedure, when there is uncertainty about effect size, is that the average power supplied to those studies in a discipline seeking a particular level of power should reach the nominal level. The common informal practice of substituting an effect size estimate from an earlier study in the traditional sample size formula fails to meet this criterion, yielding an average power that falls below the nominal level, often by a considerable margin. An alternative method of sample size determination that satisfies the average power criterion is outlined.