Statistical Power
The probability that a study rejects the null when a true effect of a given size exists. In experimental design, power calculations fix the sample size (and, for clustered designs, the number of clusters) needed to detect a hypothesized minimum detectable effect at a chosen significance level — driven by effect size, outcome variance, and (in field experiments) the intracluster correlation. Underpowered studies inflate false negatives and, conditional on “significance,” exaggerate effect magnitudes (the winner’s curse), a key mechanism linking low power to a non-credible literature. Retrospective (“observed”) power computed from the realized estimate is criticized as uninformative.
Relied on by
RCT design and any prospective experimental/quasi-experimental study.