ANOVA Explained: How to Make, Read and Report It
ANOVA (analysis of variance) tests whether the means of three or more groups differ by comparing variation between groups with variation within groups. This overview covers the logic, a worked one-way example, post hoc testing and reporting.
Why not several t-tests?
Each additional test increases the chance of a false positive. With three groups there are three pairwise tests, and the familywise error rate rises above the nominal .05. ANOVA first asks one overall question.
The logic
SStotal = SSbetween + SSwithin F = MSbetween / MSwithin
MS is the sum of squares divided by its degrees of freedom: dfbetween = k − 1 and dfwithin = N − k. If the group means are equal, F is expected to be near 1. Assumptions: independent observations, approximately normal residuals, and similar variances across groups (homogeneity of variance). ANOVA is fairly robust with equal group sizes; Welch's ANOVA is an option when variances differ.
Original example (illustrative)
Weekly study hours were compared under three tutoring formats, n = 4 per group. A: 4, 5, 6, 5. B: 6, 7, 8, 7. C: 8, 9, 10, 9.
Group means are 5, 7 and 9, and the grand mean is 7. SSbetween = 4 × (4 + 0 + 4) = 32. Within-group sums of squares are 2, 2 and 2, so SSwithin = 6 and SStotal = 38. With df = 2 and 9: MSbetween = 16, MSwithin = 0.667, so F(2, 9) = 24.00, p < .001. Effect sizes: η² = 32/38 = .84, and the less biased ω² = (32 − 2 × 0.667)/(38 + 0.667) = .79.
Tukey's HSD (Tukey, 1949) for equal n is q × √(MSwithin/n) = 3.95 × 0.408 = 1.61, using q(.05; 3, 9) ≈ 3.95. All mean differences (2, 2 and 4) exceed 1.61, so every pair differs. These numbers are deliberately clean; real data have much more overlap.
How to make it in StatFAS
Open the ANOVA Calculator. Use a group column with an outcome column, or one column per group. StatFAS has separate posts for one-way, two-way, factorial and repeated-measures ANOVA; link them after verifying their URLs.
APA 7 reporting
A one-way ANOVA showed a significant effect of tutoring format on weekly study hours, F(2, 9) = 24.00, p < .001, ω² = .79. Tukey post hoc tests indicated that all three formats differed from one another (all p < .05).
Common mistakes and alternatives
Skipping the post hoc step or interpreting a significant F as "all groups differ", using ANOVA for repeated measures, ignoring unequal variances, and reporting only η² from small samples (it is biased upward). If assumptions fail, consider Kruskal-Wallis. With two groups, ANOVA is equivalent to the t-test (F = t²).
Conclusion
ANOVA provides a single omnibus test; the useful scientific information comes from the effect size and the targeted comparisons that follow.
1.
statfas-anova-interface.png, alt: "StatFAS ANOVA Calculator interface"2.
statfas-anova-example-data.png, alt: "Example data entered in the StatFAS ANOVA Calculator"3.
statfas-anova-settings.png, alt: "ANOVA Calculator settings in StatFAS"4.
statfas-anova-result.png, alt: "ANOVA Calculator result generated in StatFAS"References
- Fisher, R. A. (1925). Statistical methods for research workers. Oliver & Boyd.
- Tukey, J. W. (1949). Comparing individual means in the analysis of variance. Biometrics, 5(2), 99–114.