More Reliable Outcome Measures Can Reduce Sample Size Requirements

Abstract

In the design of a clinical trial, considerations of statistical power primarily involve the evaluation of prospective sample sizes. Another strategy for increasing statistical power that is rarely used focuses on the selection of the outcome measure. When an outcome measure is selected, its reliability and validity must be carefully evaluated. Here the relationship between reliability and statistical power is explored empirically. We show that as the number ofrelateditems in an outcome scale increases, the internal consistency reliability of the scale also increases. As a consequence, the within-group variability decreases and, in turn, the between-group effect size increases and sample size requirements decrease. As a result, sample size requirements can be reduced and research costs decreased. We recommend careful consideration of the psychometric properties of outcome measures prior to sample size determination in any statistical power analyses.