ANOVA (Analysis of Variance) is a statistical test that checks whether the average scores of three or more groups differ by more than you'd expect from random chance — and you use it when you're comparing a numeric outcome across multiple groups or conditions. If you've ever stared at three group means, wondered if they're "really" different, and worried that running a bunch of t-tests feels wrong, ANOVA is the tool you're looking for.
Key Takeaways
- ANOVA compares the means of three or more groups to test whether at least one group differs significantly from the others.
- Use ANOVA instead of multiple t-tests — running many t-tests inflates your chance of a false positive, while ANOVA controls that error with a single test.
- A significant ANOVA tells you that a difference exists, not where — you need post-hoc tests (like Tukey's HSD) to find which specific groups differ.
- The key output is the F statistic and its p value; report effect size (η²) alongside them for APA 7 compliance.
- One-way ANOVA handles one grouping variable; two-way ANOVA handles two and can test their interaction.
What does ANOVA actually mean?
ANOVA stands for Analysis of Variance, and the name is a little misleading — you're comparing means, but you do it by analysing variance. The logic is simple: ANOVA compares how much your group averages differ from each other (variation between groups) against how much scores bounce around within each group (variation within groups).
If the difference between groups is large relative to the noise inside each group, that ratio — the F statistic — gets big, and you conclude the groups genuinely differ. If the between-group differences are small compared to the natural scatter, F stays near 1 and you can't rule out chance.
When should I use ANOVA?
Use ANOVA when you have one continuous outcome variable and want to compare its mean across three or more independent groups or conditions. Classic examples:
- Comparing test scores across three teaching methods.
- Comparing reaction times across four dosage levels of a drug.
- Comparing anxiety scores across five treatment conditions.
If you only have two groups, you don't need ANOVA — use an independent-samples t-test instead (a t-test is mathematically just a two-group ANOVA). If you're deciding between them, see our guide on the t-test.
Why not just run lots of t-tests?
Running multiple t-tests inflates your false-positive rate, and ANOVA is designed to prevent exactly that. Each t-test carries roughly a 5% chance of a false positive at α = .05. Compare four groups with separate t-tests and you'd run six comparisons — pushing your combined error rate to around 26%. ANOVA answers the whole question in one test at your intended 5% threshold.
What are the assumptions of ANOVA?
Before trusting an ANOVA result, check these:
- Independence — observations aren't linked (e.g. not the same person measured repeatedly, unless you use repeated-measures ANOVA).
- Normality — the outcome is roughly normally distributed within each group.
- Homogeneity of variance — the spread of scores is similar across groups (Levene's test checks this).
If normality is badly violated, the Kruskal–Wallis test is the non-parametric alternative. If variances differ, Welch's ANOVA is a robust fallback. Tools like StatRyx run these assumption checks automatically and flag which version of the test your data actually supports — so you don't accidentally report the wrong one.
The two main types of ANOVA
| Type | Grouping variables | Use when… | Example |
|---|---|---|---|
| One-way ANOVA | 1 | You compare means across levels of a single factor | Test scores across 3 teaching methods |
| Two-way ANOVA | 2 | You test two factors and their interaction | Scores by teaching method × gender |
| Repeated-measures ANOVA | 1 (within-subjects) | The same participants are measured multiple times | Anxiety before, during, after therapy |
The most common by far is one-way ANOVA, so that's what we'll walk through next.
A worked example: ANOVA with real numbers
Imagine a study of 45 psychology students split evenly into three study-technique groups (15 each): flashcards, re-reading, and practice testing. Their exam scores (out of 100) are compared.
Running a one-way ANOVA, we get:
F(2, 42) = 4.31, p = .019, η² = .17
Here's what each piece means:
- F(2, 42) — the F statistic is 4.31. The 2 is the between-groups degrees of freedom (3 groups − 1); the 42 is the within-groups degrees of freedom (45 participants − 3 groups).
- p = .019 — there's about a 1.9% chance of seeing group differences this large if the techniques truly had no effect. Since p < .05, the result is statistically significant.
- η² = .17 — eta squared is the effect size. It means 17% of the variance in exam scores is explained by study technique. By convention, η² of .01 is small, .06 is medium, and .14 is large — so this is a large effect.
What counts as significant?
A result is statistically significant when p is below your chosen threshold, almost always .05. In this example, p = .019 < .05, so we reject the null hypothesis that all three techniques produce equal average scores. But that only tells us at least one group differs — not which ones.
Finding which groups differ: post-hoc tests
A significant ANOVA doesn't tell you where the difference lies, so you run a post-hoc test to compare groups pair by pair. Tukey's HSD is the standard choice because it controls the error rate across all comparisons. In our example, Tukey's HSD might reveal that practice testing (M = 82.4) significantly outperformed re-reading (M = 71.2, p = .014), while flashcards fell in between and didn't differ significantly from either.
How do I report ANOVA in APA 7 format?
APA 7 wants the F statistic, both degrees of freedom, the exact p value, and an effect size. Here's a clean write-up of our example:
A one-way ANOVA revealed a significant effect of study technique on exam scores, F(2, 42) = 4.31, p = .019, η² = .17. Post-hoc comparisons using Tukey's HSD indicated that practice testing (M = 82.4, SD = 9.1) produced significantly higher scores than re-reading (M = 71.2, SD = 8.7), p = .014.
Notes on the notation: italicise F, p, M, and SD; drop the leading zero on p and η² (they can't exceed 1); and always include degrees of freedom in parentheses. Getting these details wrong is one of the most common reasons theses get sent back — StatRyx generates this exact APA sentence for you, formatting and all.
Running ANOVA without the SPSS headache
SPSS can run ANOVA, but you're navigating nested menus, decoding dense output tables, and hand-typing the results into APA format — and an SPSS licence runs roughly $100+ per year for individuals. StatRyx takes your dataset, checks the assumptions, picks between standard ANOVA, Welch's, or Kruskal–Wallis based on what your data supports, runs the post-hoc tests, and hands you a copy-paste APA 7 write-up.
Stop calculating this by hand — run it free in StatRyx → Try StatRyx
Frequently Asked Questions
What is the difference between ANOVA and a t-test?
A t