ANOVA Calculator

ANOVA (analysis of variance) answers a simple question: are the differences between group averages bigger than the ordinary scatter inside the groups would explain? Enter grouped numbers for a one-way design, or a two-factor grid for a two-way design, and this calculator builds the full ANOVA table with sums of squares, the F statistic, exact p values and effect sizes, plus Tukey pairwise comparisons for one-way designs.

How to use this calculator

  1. Choose the design. One-way: each observation belongs to one group of a single factor (three fertilisers, four teaching methods). Two-way: observations sit in a grid of two factors (fertiliser by field), with the same number of replicates in every cell.
  2. One-way: type each group’s numbers into its box, separated by commas, spaces or new lines. Name the groups if it helps you read the results. Add or remove groups as needed (2 to 10).
  3. Two-way: set the numbers of levels and replicates, press Build grid, then type each cell’s replicate values separated by commas. Every cell must hold exactly the chosen number of replicates; two-way ANOVA without replication is a different, weaker design and is not offered here.
  4. Choose the significance level (0.05 is the usual convention) and press Calculate. You get the group summary, the ANOVA table, a plain-language verdict for each effect, effect sizes, and, for one-way designs, Tukey pairwise comparisons.

The method: splitting the variation

ANOVA works by splitting the total variation in the data into parts. In a one-way design there are two parts: the variation between the group means, and the variation within the groups (the scatter of individual values around their own group mean).

SST = SSB + SSW
F = (SSB / (k − 1)) / (SSW / (N − k)) = MSB / MSW

Here k is the number of groups and N the total number of observations. If the group means truly differ, the between variation swells relative to the within variation and F grows. The p value is the probability that an F statistic this large, with these degrees of freedom, would occur if all group means were equal; it is computed here from the exact F distribution, not read from a table.

A two-way design splits the total further: a part for Factor A, a part for Factor B, a part for their interaction (the effect of A changing across levels of B), and the leftover error variation inside the cells. Each effect gets its own F test against the error mean square. Interaction is the part people miss: a significant interaction means the effect of one factor depends on the level of the other, so the main effects must be read with care.

After a significant one-way result: Tukey HSD

A significant F says at least one group mean differs, not which ones. Tukey’s honestly significant difference method compares every pair of means while keeping the family error rate at the chosen level. This calculator uses the Tukey-Kramer form, which handles unequal group sizes: a pair is declared different when its mean difference exceeds q × √(MSW/2 × (1/nᵢ + 1/nₒ)), with q from the studentized range distribution for k groups and the within degrees of freedom.

Effect sizes

Alongside p values, the results give η² (eta squared, SSB/SST: the share of total variation the factor accounts for) and ω² (omega squared, a less biased version) for one-way designs, and partial η² for each effect in two-way designs. A tiny p value with a tiny effect size means a real but small difference; both numbers are needed to judge practical importance.

Worked examples

Example 1 (one-way): three feed blends

Weight gains under three feeds, four animals each. A: 4, 5, 6, 5 (mean 5.00). B: 6, 7, 8, 7 (mean 7.00). C: 9, 10, 11, 10 (mean 10.00). The sums of squares are SSB = 50.67 and SSW = 6.00, so MSB = 25.33, MSW = 0.67 and F(2, 9) = 38.00, with p = 0.00004: the feeds differ far beyond chance scatter. Effect sizes are large: η² = 0.89 and ω² = 0.86. Tukey’s HSD at the 0.05 level is 1.61, and every pair clears it (B−A = 2, C−A = 5, C−B = 3), so all three feeds differ from each other. This data set is the Load worked example button in the one-way panel.

Example 2 (two-way): two factors, 2 by 2, two replicates

Cells (A1B1: 4, 6; A1B2: 8, 10; A2B1: 6, 8; A2B2: 12, 14). The decomposition is SSA = 18.00, SSB = 50.00, interaction SS = 2.00, error SS = 8.00, total = 78.00. With error MS = 2.00, the F statistics are F(A) = 9.00 (p = 0.0399), F(B) = 25.00 (p = 0.0075) and F(interaction) = 1.00 (p = 0.3739). At the 0.05 level both main effects are significant and the interaction is not: the factors act additively here, and B is the stronger effect. This is the Load worked example button in the two-way panel.

Reading your results

The verdict line states, for each effect, whether the differences are statistically significant at your chosen level. “Significant” means the pattern is unlikely under the null hypothesis of no effect; it does not by itself mean the effect is large or important (that is what the effect sizes are for), and it is never proof of the size of any single group’s advantage.

The ANOVA table is the audit trail: SS (sum of squares), df (degrees of freedom), MS (mean square, SS/df), F and p for every source of variation. In one-way output the rows are Between groups, Within groups and Total; in two-way output they are Factor A, Factor B, Interaction, Error and Total.

The Tukey table lists every pair of groups with its mean difference, the Tukey-Kramer threshold for that pair, and whether the pair differs significantly. Read it before claiming any particular group beat another; the overall F alone does not license pairwise claims.

Group summaries (n, mean, standard deviation) are printed so a surprising result can be sanity checked against the raw pattern: one wild value can manufacture a “significant” result, and the summary is where you spot it.

Assumptions and common mistakes

  • ANOVA assumes roughly normal data within each group and similar variances across groups. Mild departures are tolerated, especially with balanced designs; heavily skewed data or one group far more spread out than the others can distort p values. The calculator does not test these assumptions for you; the group SDs in the summary are a first look.
  • Running many t-tests instead. Testing every pair separately at 0.05 inflates the false-positive rate badly, which is exactly why ANOVA plus Tukey exists. The overall F first, then protected pairwise comparisons.
  • Reading a non-significant result as “no difference”. It means the data do not establish a difference at this sample size. Small samples miss real effects; report the effect size and the group means, not just the verdict.
  • Ignoring a significant interaction in two-way designs. When the interaction is significant, a main effect average can mislead, because the effect of A reverses or grows across B. Describe the cell means, not just the factor verdicts.
  • Using two-way without replication and expecting an interaction test. With one observation per cell there is no separate error estimate, so the interaction cannot be tested. This calculator requires replicates, deliberately.
  • Treating p as the probability the null is true. It is the probability of data at least this extreme if the null were true. The distinction sounds pedantic until a result gets quoted the wrong way round.

Limitations

  • One-way designs: 2 to 10 groups; each group needs at least one value, and the whole design needs at least one group with two or more values so the within variation exists. Two-way designs: 2 to 4 levels per factor, 2 to 4 replicates per cell, balanced (equal replicates) only.
  • Balanced two-way designs only: unequal cell counts make the decomposition non-unique (the Type I/II/III sums of squares question), which this calculator avoids rather than answers silently.
  • No repeated-measures, mixed models, blocking beyond the two crossed factors, covariates (ANCOVA), or non-parametric alternatives (Kruskal-Wallis). Those are separate tools with different assumptions.
  • p values are computed from the exact F distribution and the studentized range distribution by numerical integration; they agree with standard statistical software to the precision shown, but for publication, report the design, the test and the software conventions your field expects.

Frequently asked questions

What does an ANOVA test actually check?

Whether the group means are all equal (the null hypothesis), against the alternative that at least one differs. It does this by comparing the variation between group means with the variation within groups: if the between variation is far larger than chance scatter explains, the F statistic is large and the p value is small.

What is the difference between one-way and two-way ANOVA?

One-way has one factor (one grouping). Two-way has two factors and, with replication, also tests whether the factors interact: whether the effect of one depends on the level of the other. Choose by the design of the study, not by which gives a nicer result.

What does the p value in an ANOVA table mean?

For each effect, it is the probability of an F statistic at least as large as the one observed if that effect were truly absent. Small p (below your significance level) is evidence against “no effect”; it is not the probability that the effect is real, and it says nothing about the effect’s size.

Why do I need Tukey after ANOVA?

The overall test only says some difference exists among the means. Tukey’s method identifies which pairs differ while controlling the error rate across all the pairs at once, which plain repeated t-tests do not.

What is eta squared?

The proportion of the total variation in the data accounted for by the factor: SSB divided by SST in one-way designs. Rough guides treat 0.01 as small, 0.06 as medium and 0.14 as large, but the right yardstick is what matters in your subject, not the label.

Can ANOVA handle unequal group sizes?

One-way ANOVA can, and this calculator accepts them, with Tukey-Kramer adjustments in the pairwise tests. The two-way design here requires balanced cells, because with unequal cells the sums of squares no longer split in one agreed way.

Related calculators

Method: one-way and balanced two-way ANOVA decompositions with F tests from the exact F distribution (regularized incomplete beta); Tukey-Kramer pairwise comparisons with critical values from the studentized range distribution (numerical integration). Checked against independently calculated fixtures from standard statistical software, including F(2, 9) = 38 and a 2 by 2 design. Last reviewed: 21 October 2026.

Disclaimer: This calculator gives estimates for general information only. It is not financial, medical or other professional advice.