Power Analysis Calculator

Power analysis tells you how many participants you need to have a good chance of detecting an effect, or how much power a study of a given size has. This calculator does what G*Power does, in your browser. It covers two-sample and paired t-tests, one-way ANOVA, two proportions, correlation and chi-square, and draws a power curve so you can see the trade-offs.

The four quantities in a power analysis

Effect size, significance level (α), power (1 − β) and sample size are linked. Fix any three and the fourth is determined. Usually you choose α = 0.05 and power = 0.80, estimate the effect size, and solve for the sample size.

TestEffect sizeSmallMediumLarge
t-testsCohen’s d0.20.50.8
ANOVACohen’s f0.100.250.40
Chi-squareCohen’s w0.10.30.5
Correlationr0.10.30.5

Where should my effect size come from?

  1. Previous studies or a pilot study on the same outcome. This is the best source, though published effects are often inflated.
  2. The smallest effect that would matter in practice, e.g. a 5-point improvement on a 100-point scale with SD 15 gives d = 0.33.
  3. Cohen’s conventions (the table above), only as a last resort.

Method

The t-test, ANOVA and chi-square calculations use the exact noncentral t, F and chi-square distributions, the same method as G*Power. For example, d = 0.5, α = 0.05 (two-tailed) and power 0.80 needs 64 per group. Two proportions use the standard normal-approximation formula, and correlation uses Fisher’s z, which can differ from exact methods by one or two participants.

Worked examples

  • Two groups, medium effect: d = 0.5, power 0.80, α = 0.05 → 64 per group (128 total).
  • Three-group ANOVA, medium effect: f = 0.25 → 53 per group (159 total).
  • 2×2 chi-square, medium effect: w = 0.3, df = 1 → 88 in total.

Frequently asked questions

Why 80% power?

It is a convention suggested by Jacob Cohen. It means accepting a 20% chance of missing a real effect of the size you specified. Grant reviewers increasingly expect 90%, which needs about 34% more participants.

Should I do a power analysis after the study?

“Post-hoc power” calculated from the observed effect adds nothing beyond the p-value and is widely discouraged. Report the confidence interval of the effect instead.