One-way ANOVA (Analysis of Variance) is a parametric hypothesis test that compares the means of three or more independent groups to determine whether at least one group mean is significantly different from the others. Use it when you have one categorical independent variable with 3+ levels and one continuous dependent variable. Example: Does fertilizer type (A, B, C) significantly affect crop yield? ANOVA answers this without inflating Type I error the way multiple t-tests would.

The p-value is the probability of obtaining an F-statistic as large as — or larger than — the one observed, if all group population means are truly equal (i.e., if the null hypothesis H₀ is true). A p-value below your chosen alpha (typically 0.05) means you reject H₀ and conclude that at least one group mean differs. Note: p does not tell you which groups differ — use post-hoc tests (Tukey HSD) for that. Also note: p is not the probability that H₀ is true.

Statistical significance (p < 0.05) only tells you the result is unlikely to be due to random chance alone. It says nothing about the size or real-world importance of the difference. With very large samples, trivially tiny differences can become statistically significant. This is why you must always report an effect size alongside p. An η² = 0.01 (small effect) in 10,000 participants may be statistically significant but practically irrelevant, while η² = 0.14 (large effect) in a small sample has far more practical weight.

Eta-squared (η²) is the proportion of total variance in the dependent variable that is accounted for by group membership. It ranges from 0 to 1. Cohen's (1988) benchmarks: η² = 0.01 is a small effect, η² = 0.06 is a medium effect, and η² ≥ 0.14 is a large effect. For example, η² = 0.18 means your grouping factor explains 18% of the variance in the outcome — a large, practically meaningful effect. Omega-squared (ω²) is a bias-corrected alternative preferred for small samples.

One-way ANOVA assumes: (1) Independence — each observation must be independent of others; (2) Normality — residuals within each group should be approximately normal (check with Shapiro-Wilk test or Q-Q plot; less critical if n ≥ 15 per group); (3) Homogeneity of variances — variances across groups should be roughly equal (check with Levene's test). If normality is violated with small n, use the Kruskal-Wallis H test. If variances are unequal, use Welch's ANOVA with Games-Howell post-hoc.

A common rule of thumb is at least 15–20 observations per group. For 80% power to detect a medium effect (Cohen's f = 0.25) at α = 0.05 with 3 groups, you need approximately 52 total observations (~17 per group). For a small effect (f = 0.10), you would need ~200 total. Small groups (n < 10 per group) make ANOVA unreliable and greatly increase the risk of failing to detect a real effect (Type II error). Always plan your sample size with a power analysis before data collection.

The ANOVA F-test is an omnibus test — it tells you only that at least one group mean differs, without specifying which pairs of groups differ. Post-hoc tests like Tukey HSD perform all pairwise comparisons while controlling the family-wise error rate (keeping the overall Type I error at α). Tukey HSD is the most widely recommended post-hoc for equal or roughly equal group sizes. Use Bonferroni for a more conservative correction, or Games-Howell if variances are unequal.

APA 7th edition format: "A one-way ANOVA was conducted to compare [DV] across [IV] groups. There was a [significant/non-significant] effect of [IV] on [DV], F(df_between, df_within) = [F], p = [p], η² = [η²]." Always include group means and SDs in a table or in text. Report post-hoc results as: "Tukey HSD post-hoc tests revealed that [Group A] differed significantly from [Group B], p = [p], 95% CI [LL, UL]." See the "How to Write Your Results" section above for five fully auto-filled templates.

Yes — this tool is designed for educational and exploratory analysis and is suitable for coursework, dissertations, and preliminary research. For peer-reviewed publication, always cross-verify with validated software: aov() + TukeyHSD() in R, scipy.stats.f_oneway() in Python, or SPSS Analyze → Compare Means → One-Way ANOVA. Cite this tool as: STATS UNLOCK. (2025). One-way ANOVA calculator. Retrieved from https://statsunlock.com/tools/one-way-anova-calculator/

A non-significant result (p > α) does not prove the null hypothesis is true — it only means your data did not provide sufficient evidence to reject it. Common causes include: (a) insufficient sample size (low power), (b) large within-group variability masking real differences, or (c) the true effect being smaller than expected. Check your power: with n = 10 per group, power for a medium effect may be only 50–60%. Consider running a Bayesian ANOVA (Bayes Factor) to quantify evidence in favour of H₀ rather than simply failing to reject it.