ANOVA Calculator

About the ANOVA Calculator

Compare the means of three or more groups with a one-way ANOVA, or study two factors and their interaction with a two-way ANOVA, with or without replication. Paste raw data — one group per line, columns from a spreadsheet, a spreadsheet’s two-factor grid, or one row per value — or type each group’s n, mean and standard deviation when that is all you have.

You get the full ANOVA table (sums of squares, degrees of freedom, mean squares, F and the p-value), the critical F, the effect sizes η², ω² and partial η², and Tukey’s HSD comparisons of every pair of means with adjusted p-values, confidence intervals and a letter display (the Tukey–Kramer form when group sizes differ). One-way results also include Welch’s ANOVA, which does not assume equal variances, and the Brown–Forsythe test of equal variances. Unbalanced two-way designs get Type III sums of squares (the default of SPSS and SAS) or Type I and II on request. Every result comes with the working, a chart, a plain-language conclusion and an APA-style sentence.

Common uses

• Testing whether three or more teaching methods, fertilisers, diets, machines or ad versions give different average results.
• Finding which pairs of groups differ after a significant ANOVA, with Tukey’s HSD and a letter display for a table or chart.
• Analysing a two-factor experiment and checking whether the factors interact.
• Re-checking an ANOVA from a paper when only group sizes, means and standard deviations are reported.
• Getting an APA-style sentence and a Word report for coursework, a thesis or a lab report.

How one-way ANOVA works

ANOVA splits the total variation into the part between the groups and the part within them. With group means ȳᵢ, sizes nᵢ and standard deviations sᵢ, and the grand mean ȳ:

• SS between = Σnᵢ(ȳᵢ − ȳ)², with k − 1 degrees of freedom;
• SS within = Σ(nᵢ − 1)sᵢ², with N − k degrees of freedom;
• each mean square is SS / df, and F = MS between / MS within.

If every group has the same mean, F follows the F distribution with (k − 1, N − k) degrees of freedom, and the p-value is the chance of an F this large or larger. These are the formulas of Fisher’s Statistical Methods for Research Workers and of the NIST/SEMATECH e-Handbook, whose examples the calculator reproduces. Effect sizes: η² = SS between / SS total, and ω² = (SS between − (k − 1)·MS within)/(SS total + MS within), which corrects η²’s upward bias.

Tukey’s HSD test

A significant F says that the means are not all equal, not which ones differ. Tukey’s honestly significant difference test compares every pair while keeping the chance of any false difference at α: two means differ when |ȳᵢ − ȳⱼ| / SE exceeds the critical value q(α; k, N − k) of the studentized range distribution, with SE = √(MS within / 2 · (1/nᵢ + 1/nⱼ)) — the Tukey–Kramer form, which gives each pair its own standard error when the groups differ in size. The calculator computes the studentized range distribution itself, by numerical integration of its definition, and checks it against the published tables of q.

Unequal variances: Welch and Brown–Forsythe

Classic ANOVA assumes the groups have the same variance. Welch’s ANOVA weights each group by nᵢ/sᵢ² and adjusts the denominator degrees of freedom, so it keeps the right error rate when variances and group sizes differ; when the two tests disagree, Welch’s is the safer one. The Brown–Forsythe test checks the assumption itself: it runs an ANOVA on each value’s distance from its group’s median (Levene’s test with the median), which stays reliable when the data are skewed.

Two-way ANOVA and the types of sums of squares

With two factors A and B, the variation splits into the main effect of A, the main effect of B, their interaction (does the effect of A depend on the level of B?) and the error within each combination (NIST/SEMATECH e-Handbook). When every combination has the same number of values, the split is unique. When they differ, it depends on the order in which the effects are fitted: Type I fits A, then B, then the interaction; Type II tests each main effect after the other; Type III tests each effect after all the others, in sum-to-zero coding, which is what SPSS and SAS report by default. With one value per combination, the interaction cannot be separated from the error, so it serves as the error term and the factors are assumed not to interact.

Assumptions

• Independent observations, within and between groups.
• Roughly normal data in each group (or combination). ANOVA is robust to moderate skew with similar group sizes; with small, skewed groups consider a rank test such as Kruskal–Wallis.
• Equal variances, unless you use Welch’s ANOVA. Unequal variances matter most when the group sizes differ too.

Limitations

• One-way ANOVA takes up to 40 groups; two-way ANOVA up to 12 levels per factor; up to 200,000 values in all.
• Two-way ANOVA needs at least one value in every combination of the two factors; Tukey comparisons of the main effects are shown for balanced designs.
• Both factors are fixed effects; random-effects, repeated-measures and three-way designs are not covered.
• Standard deviations must be sample standard deviations (n − 1 in the denominator), as software reports them.
• The calculator does not test normality or remove outliers: look at the data, especially with small groups.

Privacy

On this device: nothing is sent

Everything happens in your browser. What you enter or open here is not uploaded or stored by MySmartCoPilot.