Experiment Design Studio

Sample sizes for t-tests and ANOVA, covering two groups, paired designs, and one-way and repeated-measures ANOVA, are found by computing exact power from the noncentral F distribution, the approach used by G*Power; one-sided t-tests use the upper tail of the noncentral t distribution. The values reproduce G*Power’s results for standard cases: for example, 64 per group for d = 0.5, 34 pairs for dz = 0.5 and 53 per group for f = 0.25 with three groups.

For three or more conditions measured in the same units, the power calculation uses the correlation between repeated measurements and assumes sphericity. Where sphericity is doubtful, a sample slightly larger than the number shown is advisable.

When a rank-based test such as the Mann–Whitney or Wilcoxon test is planned, the sample size is increased using an asymptotic relative efficiency of 0.864, the lowest value over all distributions. The resulting number is therefore conservative.

Sample sizes for correlations use the Fisher z approximation, those for two proportions the normal approximation with pooled variance, and those for time-to-event outcomes Schoenfeld’s formula for the number of events with equal allocation. These approximations can differ slightly from exact methods.

The review applies a set of rules to the answers given. The rules are drawn from published standards for rigorous design and reporting: randomisation, blinding, sample-size justification and data handling, the experimental unit, sex as a biological variable, pre-registration and trial registration.

The studio’s calculations have been checked against an independent statistical implementation across more than 120,000 test cases.

All answers are stored only in the browser and are not sent to GloNeuro or to any other party. The File menu saves a copy of a plan or opens a saved one.