Quick answer: In jamovi: ANOVA → One-Way ANOVA (or ANOVA → ANOVA for factorial designs), dependent variable in, grouping factor in, and tick descriptive plots under Estimated Marginal Means or Post Hoc Tests → Tukey for pairwise comparisons. ANOVA answers one question — “do ANY group means differ?” — and Tukey then tells you which pairs. Worked example and interpretation below; the t-test version of this workflow is in our jamovi t-test walkthrough.
Worked example: three fertilizers, one yield
45 plots, yield in kg, fertilizer type A/B/C. Setup: ANOVA → One-Way ANOVA → Yield into Dependent Variables, Fertilizer into Fixed Factors. Tick η² (Eta-squared) under Additional Statistics, plus Descriptives table and QQ plot of residuals for assumption checks. Suppose output: F(2, 42) = 6.81, p = 0.003, η² = 0.245.
Reading it: F = variance between groups ÷ variance within groups; p = 0.003 says at least one fertilizer differs (the overall logic: our hypothesis testing guide), and η² = 0.245 says fertilizer explains ~24% of yield variance — a substantial effect. But the F-test never says which pairs differ — that’s Tukey’s job.
Post hoc, plots, and the assumption checks
- Post Hoc Tests → Tukey: adds a pairwise table with corrected p-values. Typical readout: A−B p = .002 (differs), A−C p = .89, B−C p = .041 — now the story is “A beats both, B edges C.”
- Descriptives plots: Estimated Marginal Means → plot the factor; the group means with error bars make the pattern visible in one glance.
- Normality: residuals should be roughly normal (QQ plot) — with equal n per group, ANOVA is quite forgiving.
- Homogeneity (Levene’s): if violated, report Welch’s one-way correction — jamovi offers it under the same menu in newer versions, or Kruskal-Wallis for a rank-based fallback.
ANOVA vs repeated t-tests: why not just three t-tests?
Three pairwise t-tests at α = .05 give a combined false-positive chance near 14% — every extra test inflates it (the mechanism: our p-value explainer). ANOVA’s single overall F-test keeps the error rate at 5%, and Tukey’s correction handles the pairwise follow-ups honestly. If you also measured the same plots across seasons, that becomes a repeated-measures design — ANOVA → Repeated Measures in jamovi. For chi-square style categorical data instead, see the chi-square in jamovi guide.
For one-to-one coaching on factorial designs, interactions and interpreting η², Ampersand Academy teaches statistics with jamovi one-to-one.
Frequently asked questions
How do I run a one-way ANOVA in jamovi?
Analyses ribbon, ANOVA, One-Way ANOVA. Move the numeric outcome into Dependent Variables and the grouping factor into Fixed Factors. Tick effect size and descriptives, then open Post Hoc Tests and choose Tukey for pairwise comparisons.
What does a significant ANOVA result actually tell me?
Only that at least one group mean differs from the others. It does not say which pairs differ – that is what post hoc tests like Tukey’s are for, and jamovi prints them with corrected p-values.
What is eta-squared and how big is a big effect?
Eta-squared is the share of outcome variance explained by the grouping factor. Roughly 0.01 is small, 0.06 medium, 0.14 large – and it makes results comparable across studies in a way raw F values are not.
Can jamovi run repeated-measures ANOVA?
Yes – ANOVA, Repeated Measures ANOVA. Define within-subject factors and their levels, move repeated columns into each cell, and jamovi outputs the full table with sphericity corrections (Greenhouse-Geisser) available.
When should I use Kruskal-Wallis instead of ANOVA?
When the outcome is ordinal or severely non-normal with small groups. It is the rank-based analog of one-way ANOVA, available from jamovi’s One-Way ANOVA menu, with Dunn’s post hoc comparisons as follow-up.