ANOVA Help: Compare Groups With Confidence
Analysis of Variance (ANOVA) is one of the most widely used techniques in research. Any time you compare the means of three or more groups, or examine how two or more factors combine to influence an outcome, you are in ANOVA territory.
It is also one of the most frequently misapplied methods: the wrong design selected, assumptions left unchecked, post-hoc tests chosen at random and interactions misinterpreted. As a statistics professor helping students since 2014, I make sure your ANOVA is the right model, correctly run and clearly reported.
What ANOVA Actually Does
ANOVA tests whether the variability between group means is large relative to the variability within groups. The resulting F-ratio tells you whether at least one group mean differs from the others. A significant F does not tell you which groups differ; that is the job of planned contrasts or post-hoc tests.
Why not just run several t-tests? Because each test carries its own chance of a false positive. With three groups and three t-tests at α = .05, the familywise error rate rises to roughly 14%. ANOVA controls that risk.
Which ANOVA Do You Need?
| Design | When to use it | Example |
|---|---|---|
| One-way ANOVA | One categorical factor, 3+ independent groups | Exam scores across three teaching methods |
| Factorial (two-way) ANOVA | Two or more factors, including their interaction | Effect of teaching method and gender on scores |
| Repeated-measures ANOVA | Same participants measured 3+ times | Anxiety at baseline, 6 weeks and 12 weeks |
| Mixed ANOVA | Between-subjects and within-subjects factors together | Treatment vs. control measured over time |
| ANCOVA | Group comparison adjusting for a continuous covariate | Post-test scores adjusted for pre-test scores |
| MANOVA | Two or more related dependent variables | Treatment effect on anxiety and depression |
Not sure which applies? Try the free Which Test to Use? guide.
Assumptions We Check Together
- Independence of observations: guaranteed by design, not by a test
- Normality of residuals: Q-Q plots and Shapiro-Wilk, interpreted in light of sample size
- Homogeneity of variance: Levene’s test, and Welch’s ANOVA with Games-Howell when it fails
- Sphericity (repeated measures): Mauchly’s test, with Greenhouse-Geisser or Huynh-Feldt corrections
- Homogeneity of regression slopes (ANCOVA)
- Outliers and their influence on group means
When assumptions are violated, I help you choose a defensible alternative, such as Welch’s ANOVA, Kruskal-Wallis, Friedman, robust methods or a mixed-effects model, and justify it in your write-up.
Post-Hoc Tests and Planned Contrasts
- Tukey HSD: all pairwise comparisons with equal variances
- Games-Howell: unequal variances or unequal group sizes
- Bonferroni and Holm: a small number of planned comparisons
- Dunnett: comparing every group to a single control
- Simple effects analysis: unpacking a significant interaction
Effect Sizes and APA Reporting
A p-value alone is not enough. I show you how to report η² or partial η² (and ω² when bias matters), along with confidence intervals and descriptive statistics, in correct APA 7 format, for example:
A one-way ANOVA revealed a significant effect of teaching method on exam scores, F(2, 87) = 6.42, p = .002, η² = .13.
ANOVA in Your Software
I support ANOVA in SPSS (GLM Univariate and Repeated Measures),
R (aov, car, afex, emmeans), Stata
(anova, margins), Python (statsmodels, pingouin) and
Excel (Data Analysis ToolPak).
Get Expert Help With Your ANOVA
Whether you are preparing for an exam, analyzing experimental data or writing your dissertation results chapter, book a free 15-minute consultation. Tell me your design, your variables, your software and your deadline, and you will receive a clear plan and quote within 24 hours.
Prefer email or chat? Write to [email protected] or message me on WhatsApp at +1 (425) 999-3993.