Compare the means of three or more independent groups in one test, or analyse two factors at once with two-way ANOVA. Paste each group on its own line or paste columns straight from Excel. The output includes descriptive statistics with confidence intervals, the ANOVA table, η², ω² and Cohen’s f, Levene’s test, Welch’s ANOVA for unequal variances, and your choice of post-hoc test with adjusted p-values, plus an APA sentence.
Two-way ANOVA
Switch to the Two-way tab when your groups are defined by two factors, such as fertiliser and watering level. Paste three columns from a spreadsheet (factor A, factor B and the measurement) with a header row. You get the main effect of each factor, their interaction, partial η² for each term, a table of cell means and an interaction plot. Unbalanced designs are handled with Type III sums of squares, the same as SPSS. For a worked example by hand, read the two-way ANOVA guide.
How to enter your groups
Type one group per line, optionally starting with a name and a colon, for example Lecture: 85 90 78 92. Or copy a block of columns from a spreadsheet, one column per group with the group names in the first row. Groups can have different sizes.
Choosing a post-hoc test
| Post-hoc test | Use when |
|---|---|
| Tukey HSD | Variances look equal. It is the standard choice for all pairwise comparisons and uses the Tukey–Kramer form for unequal group sizes. |
| Games–Howell | Variances differ (Levene’s test is significant), especially with unequal group sizes. |
| Bonferroni / Holm | You want a simple, widely understood correction. Holm is always at least as powerful as Bonferroni. |
| Fisher LSD | Only three groups and the overall F is significant. It makes no correction. |
Formula
F = MS_between ÷ MS_within = [ Σ nᵢ(x̄ᵢ − x̄)² ÷ (k − 1) ] ÷ [ Σ Σ (x − x̄ᵢ)² ÷ (N − k) ]
Assumptions and what to do if they fail
- Independent observations: each participant appears in one group only. For repeated measurements use a repeated-measures design.
- Roughly normal residuals: check the box plots. With skewed data or outliers, use the Kruskal-Wallis test.
- Equal variances: if Levene’s test is significant, report Welch’s ANOVA (shown automatically) and use Games–Howell post-hoc tests.
Effect sizes
η² is the share of total variability explained by group membership: about .01 is small, .06 medium and .14 large. ω² is a less biased version that is better for small samples. Cohen’s f is the effect size G*Power and our power analysis calculator use for planning sample sizes.
Frequently asked questions
Why not just run several t-tests?
With three groups there are three pairwise comparisons, and each one carries a 5% risk of a false positive. Together the risk is about 14%. ANOVA tests all groups at once, and post-hoc tests keep the overall error rate at 5%.
The ANOVA is significant but no post-hoc pair is. Why?
The overall F can pick up a pattern spread across several groups that no single pair shows strongly. It happens most often when the overall p is close to .05. Report both results honestly.