Power analysis
Also known as: A priori power analysis, Sample size calculation
Planning sample size to give a study a real chance of finding a true effect.
What it means
A power analysis is the calculation that links a study's sample size, the significance threshold, the expected effect size, and its statistical power — the probability of detecting an effect that genuinely exists. Conducted before data collection, it determines how many participants are needed to keep the false-negative rate acceptably low for a plausible effect. Skipping it produces chronically underpowered studies that both miss real effects and, through selection on significance, exaggerate the ones they report. Honest power analysis depends on a realistic estimate of the smallest effect worth detecting, which is often the hardest and most consequential input.
Examples
Calculating that detecting a small correlation with 80% power and alpha 0.05 requires roughly 780 participants, not the 50 originally planned.
A product team runs an A/B test on 400 users, sees no difference, and calls the new checkout a failure — with that sample they could only ever have detected a huge jump in conversion.
A gym asks whether a new class reduces dropout; the planner works backwards from the smallest drop worth caring about and finds the study needs a year of sign-ups, not a month.
First described in Jacob Cohen (1969, 1988).