The sample size question, answered before it hurts
23 October 2026 · Cohira team
An underpowered study is not a smaller version of a good study. It is a study designed to be inconclusive.
Four numbers decide it
The effect size worth detecting, the outcome's variability, the alpha level, and the power you want. Everything else is arithmetic. The hard part is the first number, which is a clinical judgement, not a statistical one.
Where studies go wrong
Powering for an implausibly large effect to keep the sample convenient; ignoring clustering; forgetting attrition. Each inflates the false reassurance of a formal calculation.
What a reviewer wants to see
A justified effect size with a citation, the formula or software named, and the assumptions stated plainly, before the data was collected. A sample size argued after the results is an argument lost.