Repeated Measures ANOVA & Mixed Model Calculator
Repeated-measures ANOVA with Greenhouse-Geisser correction and linear mixed models for unbalanced designs with missing values, alongside t-tests, one-way ANOVA, Tukey, Kruskal-Wallis and Dunn. Exact p-values, free, runs in your browser.
How it works
Two-sample and paired t-tests, one-way ANOVA, Tukey-Kramer HSD, Kruskal-Wallis and Dunn are computed with exact p-values from the relevant distributions rather than lookup tables. The tool reports the test statistic, degrees of freedom and p-value together, since a p-value without its statistic and df cannot be independently checked. Two designs most bench software skips are included: repeated-measures ANOVA, for when the same passage or animal is measured under every condition, with Mauchly sphericity checked and the Greenhouse-Geisser correction applied; and a linear mixed-effects model with a random intercept, for unbalanced designs where a subject is missing a measurement and a repeated-measures test would otherwise force you to discard that subject entirely.
Frequently asked questions
When should I use a t-test instead of Mann-Whitney?
Use a t-test when your data are approximately normally distributed and you are comparing means; it has more power under those conditions. Use Mann-Whitney when the data are clearly skewed, ordinal, or contain outliers you cannot justify removing — it compares distributions rather than means and makes no normality assumption. With very small samples, normality is hard to assess, and the non-parametric test is the more conservative choice.
What is the difference between paired and unpaired t-tests?
A paired test is for measurements that come in natural pairs — the same sample before and after treatment, or matched littermates. It tests whether the mean of the within-pair differences is zero, which removes between-subject variability and gives more power. An unpaired test compares two independent groups. Using an unpaired test on paired data discards the pairing and loses sensitivity; using a paired test on independent data is simply invalid.
Which test should I use when the same cells are measured under every condition?
A repeated-measures ANOVA, not a one-way ANOVA. When one passage of cells is treated three ways and the experiment repeats on separate days, the three measurements within a day are not independent — a slow-growing day moves all of them together. One-way ANOVA assumes independence, so it pushes that shared day-to-day swing into the error term, inflating it and hiding real treatment effects. Repeated measures removes the day as its own source of variation, so the F ratio is tested against error that no longer contains it. Same data, correct partition, usually a smaller p — not because the test is more permissive, but because what it divides by was never noise.
What do I do if one sample is missing a measurement?
Use a linear mixed-effects model rather than dropping the sample. A repeated-measures ANOVA requires every subject in every condition, so a single failed well forces you to discard all of that subject's other measurements. A mixed model treats the subject as a random effect, so it contributes whatever data it has while the shared subject-level offset is still removed from the error term. That is the whole reason mixed models exist, and it is why unbalanced designs are not a reason to throw away data.