One-way ANOVA calculator

The line between a one-way ANOVA and a two-way ANOVA is simply the number of grouping factors your design contains. A one-way design has a single independent variable, such as comparing exam scores across three teaching methods. A two-way design crosses two factors at once, for example teaching method and gender, and it earns its keep by testing each factor separately while also testing their interaction effect, the way the influence of one factor shifts across the levels of the other. That interaction is usually the most interesting result, because it tells you the factors do not act independently.

Choose the structure that mirrors how you actually collected the data. If every participant appears in only one cell of the design, a between-groups ANOVA applies; if the same participants are measured under several conditions, you need the repeated-measures version of the analysis instead. When a single two-group comparison is all you have, an ANOVA reduces to a t-test, and the relationship between the t-test and ANOVA shows why the F equals t squared in that case.

The ANOVA summary table reports a row for the between-groups effect and a row for the within-groups error, and reading it in order makes the logic clear. Each row lists a sum of squares, its degrees of freedom, and the mean square that results from dividing one by the other. The F ratio in the effect row is the between-groups mean square over the within-groups mean square, so it asks whether the spread among the group means is large relative to the ordinary scatter inside each group. An F near 1 is what you expect when the means are genuinely equal.

Significance is decided by comparing the F against its critical value at your chosen alpha, which the reported p-value already encodes: a value below .05 says the means are unlikely to be equal. Always read the F beside the eta-squared effect size, because a large sample can push a trivial difference past significance. For a wider walkthrough of the output panels, see the box-by-box SPSS output guide before you paste numbers into your results chapter.

An ANOVA rests on three assumptions: the observations are independent, the residuals are approximately normal within each group, and the groups share a similar variance, a condition called homogeneity of variance and usually checked with Levene's test. The procedure is fairly robust to mild departures from normality when group sizes are equal, but unequal variances paired with unequal group sizes can distort the F badly. If normality is in doubt, run a formal check for normality rather than eyeballing a histogram.

When the assumptions fail and a transformation does not fix them, the Kruskal-Wallis test is the nonparametric counterpart to a one-way ANOVA. It ranks all observations and compares the rank sums across groups, so it needs no normality assumption and tolerates ordinal data and outliers. The trade is a small loss of power when the data really are normal, and the conclusion shifts from a difference in means to a difference in distributions. The choice between parametric and nonparametric tests lays out the full decision.

A significant omnibus F tells you the group means are not all equal, but it stays silent on which pairs differ. That is the job of a post hoc test, run only after the ANOVA is significant so you do not fish through comparisons that the data never justified. Tukey's HSD is the standard choice when you want every pairwise comparison while holding the family-wise error rate at your alpha, and the Bonferroni correction offers a simpler, more conservative alternative when you have a small set of planned contrasts.

The right procedure also depends on whether the variances were equal. When homogeneity of variance holds, Tukey's HSD is efficient; when it does not, the Games-Howell test is safer because it does not pool the variances. Whatever you pick, report the adjusted comparisons with their confidence intervals and an effect size for each pair, so the reader sees both where the differences sit and how large they are, not just that some difference exists somewhere in the design.

APA style packs the ANOVA result into a single sentence built around the F statistic. The format is F(df between, df within) = value, p = exact value, followed by the effect size, for example F(2, 87) = 5.42, p = .006, η² = .11. Report the F to two decimals, give the exact p-value rather than only "p < .05", and italicise the test letters. The two degrees of freedom go in parentheses in the order between-groups then within-groups.

After the omnibus sentence, summarise the group means and standard deviations, then describe the significant post hoc comparisons in words so the reader knows which groups drove the effect. For the surrounding conventions on signs, decimals, and table layout, follow the guide to reporting statistics in APA style, or hand the write-up to our APA sentence builder to get the wording right the first time.