Enter each group's measurements

Paste or type the measurements for three or four independent groups, one group per box. Numbers can be separated by commas, spaces, semicolons, or new lines. Each group needs at least two values so its variance is defined; the fourth group is optional for a three-group ANOVA.

02

Name your groups

Replace the default 'Group 1, 2, 3' labels with the conditions you're comparing — 'Placebo', '10 mg', '20 mg', and so on. Labels carry through to the result card, the ANOVA table, the bar chart, and every pairwise comparison row.

03

Pick an alpha level

Alpha is the false-positive risk you're willing to accept. 0.05 (5%) is the conventional default in most fields; pick 0.01 for higher-stakes decisions where a false significance is costly, or 0.10 for exploratory work. The critical F value and the reject/fail-to-reject decision update with your choice.

04

Read the F-statistic and check the post-hoc tab

If F exceeds the critical F (equivalently, if p < alpha) the means differ more than random sampling alone would explain — reject the null hypothesis of equal means. Then open the Post-Hoc Comparison tab to see which specific group pairs drive the difference; rows highlighted in green are significant after Bonferroni correction.

F-statistic (one-way ANOVA)

F = MS_between / MS_within where MS_between = SS_between / (k - 1) and MS_within = SS_within / (N - k)

F compares the variance explained by group membership (mean square between) against the residual variance within groups (mean square within). A large F means the group means are far apart relative to the spread inside each group, so the difference is unlikely to be due to chance.

Sum-of-squares partition

SS_total = SS_between + SS_within; SS_between = sum n_i * (x_i_bar - x_bar)^2; SS_within = sum sum (x_ij - x_i_bar)^2

Total variability decomposes additively into the variance between groups (each group mean's deviation from the grand mean, weighted by sample size) and the variance within groups (each observation's deviation from its own group mean). The same partition underpins R-squared in regression.

Eta-squared (effect size)

eta^2 = SS_between / SS_total

Eta-squared is the proportion of total variability explained by group membership. It ranges from 0 (no group effect) to 1 (groups explain everything). Cohen's rough benchmarks: 0.01 small, 0.06 medium, 0.14 large. Report eta-squared alongside p — a tiny p with a tiny eta-squared often signals a large N rather than a meaningful effect.

Bonferroni-corrected pairwise t (post-hoc)

t_ij = (x_i_bar - x_j_bar) / sqrt(MS_within * (1/n_i + 1/n_j)); p_adj = min(1, p_raw * k*(k-1)/2)

Once ANOVA rejects, the post-hoc tab tests every pair using the pooled within-groups variance for the standard error and the within-groups degrees of freedom. Multiplying each raw p-value by the total number of comparisons (Bonferroni) controls the family-wise error rate so the overall false-positive risk stays at alpha.

F-statistic The ratio of between-groups mean square to within-groups mean square. Values near 1 are consistent with equal-mean groups; larger values indicate the group means are spread out far relative to the noise inside each group.

Between-groups sum of squares (SS_between) The weighted sum of squared deviations of each group mean from the grand mean. It captures variation that can be attributed to group membership — the signal that ANOVA is trying to detect.

Within-groups sum of squares (SS_within) The sum of squared deviations of every observation from its own group mean. It captures variation that is unexplained by group membership — the residual noise that the F-ratio compares the signal against.

Degrees of freedom df_between equals k minus 1 (one fewer than the number of groups). df_within equals N minus k (total observations minus the number of group means estimated). The F-distribution that gives the p-value is indexed by this pair, written F(df_between, df_within).

Critical F The smallest F value that triggers rejection at your chosen alpha. If your observed F exceeds the critical F, p is less than alpha; if not, p exceeds alpha and you fail to reject the null. The critical F depends on alpha and both degrees of freedom.

Eta-squared Proportion of total variance explained by group membership: SS_between divided by SS_total. It's the ANOVA analog of R-squared in regression. Always report it — a tiny p with eta-squared below 0.01 typically means you found a real but trivially small effect in a very large sample.

Post-hoc comparison A follow-up test that runs after a significant omnibus ANOVA to identify which specific pair of groups differs. Bonferroni correction guards against false positives by multiplying raw pairwise p-values by the number of comparisons.

Type I error Rejecting the null hypothesis when it's actually true (a false positive). Alpha is the long-run rate at which this happens — choosing alpha = 0.05 means 1 in 20 comparisons over equal-mean populations will falsely flag a difference.

🌾

Agricultural researcher

Crop yields under three fertilizer regimes

Group A (Control) 48, 49, 50, 49 Group B (Fertilizer X) 47, 49, 48, 48 Group C (Fertilizer Y) 49, 51, 50, 50 Alpha level 0.05

Grand mean is 49 bushels per plot; SS_between = 8 and SS_within = 6 give F(2, 9) = 6.0 with p around 0.022, just inside the 5% threshold. Eta-squared = 0.57 says fertilizer choice explains 57% of the yield variation. Post-hoc reports Group B vs Group C as the only significant pair after Bonferroni correction — Fertilizer Y outperforms Fertilizer X but neither significantly beats the control at this sample size.

💊

Clinical trial coordinator

Dose-response comparison across four arms

Placebo 12, 14, 11, 13, 15 10 mg 18, 20, 17, 19, 21 20 mg 22, 25, 23, 24, 26 40 mg 28, 30, 27, 29, 31

Four arms of n=5 produce monotonically increasing means (13, 19, 24, 29). F(3, 16) is roughly 116 with p far below 0.0001 — a hugely significant dose effect, and eta-squared is about 0.96, putting almost all the response variation on dose. Every Bonferroni-corrected pair flags as significant, so each dose step adds measurable response. A trial reporter would still flag that ANOVA assumes equal variances and independent observations; if either is doubtful, run a Welch's ANOVA or a non-parametric Kruskal-Wallis alternative.

One-way ANOVA tests whether the means of three or more independent groups differ by more than random sampling variation alone. This calculator computes the F-statistic, its p-value, eta-squared as an effect size, and Bonferroni-corrected pairwise comparisons so you can identify which specific groups drive any significant overall difference.

How ANOVA works

ANOVA partitions the total variability in your data into two pieces: variability between groups (how far each group mean sits from the grand mean) and variability within groups (how far individual observations sit from their own group mean). If group membership matters, the between-groups piece should dwarf the within-groups piece. The F-statistic is exactly that ratio — between-groups mean square divided by within-groups mean square — and the F-distribution gives the probability of seeing an F that large by chance if every group truly had the same mean. A small p-value (smaller than your chosen alpha) means that's unlikely, so you reject the null hypothesis of equal means.

Why ANOVA instead of multiple t-tests

Running pairwise t-tests across k groups means k(k-1)/2 separate comparisons, each at alpha = 0.05. With four groups that's six tests, and the chance of at least one false positive climbs to roughly 26%. ANOVA performs a single omnibus test that holds the family-wise error rate at alpha for the overall question of whether any means differ. Only after that test rejects do you drop into pairwise comparisons — and even then you apply a multiplicity correction like Bonferroni so the overall false-positive risk stays controlled.

Limits and edge cases

One-way ANOVA assumes the groups are independent, the residuals are roughly normally distributed within each group, and the groups have roughly equal variances (homoscedasticity). The F-test is fairly robust to mild violations, but if variances differ by a factor of more than 3 or 4, use Welch's ANOVA. If the data are highly skewed or ordinal, use the non-parametric Kruskal-Wallis test instead. ANOVA also can't tell you about repeated measurements on the same subjects — that calls for a repeated-measures or mixed-effects model. Finally, eta-squared is the simplest effect-size measure but slightly biased upward in small samples; report omega-squared for publication-quality work.

What is ANOVA and what does it test?+

Analysis of variance (ANOVA) tests whether three or more independent groups have different mean values. The null hypothesis is that every group is drawn from a population with the same mean; rejecting it means at least one group's mean differs, though ANOVA itself doesn't say which one — that's what the post-hoc tab is for.

When should I use ANOVA instead of a t-test?+

Use a t-test when you have exactly two groups; use one-way ANOVA when you have three or more independent groups and want a single test of whether any means differ. Running multiple pairwise t-tests across many groups inflates the false-positive rate, so an omnibus ANOVA followed by corrected pairwise comparisons is the right pattern.

How do I interpret the F-statistic?+

F is the ratio of between-groups variance to within-groups variance. Values near 1 mean the groups don't separate any more than random noise would predict. The further F sits above 1 the stronger the evidence of a group effect. Always pair F with its p-value and an effect size like eta-squared — a giant F in a tiny sample says less than a moderate F in a large sample.

What's the difference between one-way and two-way ANOVA?+

One-way ANOVA tests the effect of a single categorical factor (this calculator's setup). Two-way ANOVA tests two factors at once and lets you measure their interaction — for example, drug dose AND patient age group, with the question of whether dose response depends on age. Two-way ANOVA isn't supported here; for that, use a dedicated stats package.

What does the p-value mean?+

The p-value is the probability of observing an F-statistic at least as extreme as the one you computed if the null hypothesis (all group means equal) were true. A small p — typically below your alpha threshold of 0.05 — is grounds for rejecting the null. A p of 0.03 means there's a 3% chance you'd see this much group separation by random sampling alone.

ANOVA Calculator

Group data

Group summary

ANOVA summary table

Pairwise comparisons (Bonferroni-corrected)

How to Use This Calculator

Enter each group's measurements

Name your groups

Pick an alpha level

Read the F-statistic and check the post-hoc tab

Formula & Methodology

Key Terms Explained

Real-World Examples

Agricultural researcher

Clinical trial coordinator

Understanding the ANOVA Calculator

How ANOVA works

Why ANOVA instead of multiple t-tests

Limits and edge cases

Frequently Asked Questions

Source	SS	df	MS	F	p	F-critical
Between groups	—	—	—	—	—	—
Within groups	—	—	—	—	—	—
Total	—	—	—	—	—	—

ANOVA Calculator

Group data

Group summary

ANOVA summary table

Pairwise comparisons (Bonferroni-corrected)

How to Use This Calculator

Enter each group's measurements

Name your groups

Pick an alpha level

Read the F-statistic and check the post-hoc tab

Formula & Methodology

Key Terms Explained

Real-World Examples

Understanding the ANOVA Calculator

How ANOVA works

Why ANOVA instead of multiple t-tests

Limits and edge cases

Frequently Asked Questions

Keep Exploring

Related calculators

Guides & articles