How to Safely Reassess Variability and Adapt Sample Size? A Primer for the Independent Samples t Test

Abstract

When researchers aim to test hypotheses, setting up adequately powered studies is crucial to avoid missing important effects and to increase the probability that published significant effects reflect true effects. Without a priori good knowledge about the population effect size and variability, power analyses may underestimate the true required sample size. However, a specific type of a two-stage adaptive design in which the sample size can be reestimated during the data collection might partially mitigate the problem. In the design proposed in this article, the variability of the data collected at the first stage is estimated and then used to reassess the originally planned sample size of the study while the unstandardized effect size is fixed at a smallest effect size of interest. In this article, we explain how to implement such a two-stage sample-size reestimation design in the setting in which interest lies in comparing means of two independent groups. We investigate through simulation the implications on the Type I error rate (T1ER) of the final independent samples t test. Inflation can be substantial when the interim variance estimate is based on a small sample. However, the T1ER approaches the nominal level when more first-stage data are collected. An R-function is provided that enables researchers to calculate for their specific study (a) the maximum T1ER inflation and (b) the adjusted [Formula: see text] level to be used in the final t test to correct for the inflation. Finally, the desired property of this design to better ensure the power of the study is verified.