APA 7 and every modern journal expect an effect size next to your test result, not just a p value. A p value tells you whether an effect is likely real; an effect size tells you how big it is. The hard part is knowing which effect size belongs with which test and how to format it. This guide is the quick reference, test by test.
Why an effect size, not just a p value
A p value depends on sample size: with a large enough sample, a trivial difference becomes statistically significant. The effect size is the part a reader actually cares about, the magnitude of the difference or relationship, and it is what meta-analyses and power analyses are built on. Reporting it is not optional in APA 7.
Which effect size goes with which test
| Test | Effect size | Format example | Leading zero? |
|---|---|---|---|
| t-test | Cohen's d (or Hedges' g) | d = 0.62 | Keep (can exceed 1) |
| One-way ANOVA | Eta squared or omega squared | eta-squared = .12 | Drop |
| Factorial / two-way ANOVA, ANCOVA | Partial eta squared | partial eta-squared = .09 | Drop |
| Pearson correlation | r (the r is itself the effect size) | r = .34 | Drop |
| Linear / multiple regression | R-squared, or Cohen's f-squared | R-squared = .28 | Drop |
| Chi-square | Cramer's V (or phi for 2x2) | V = .21 | Drop |
| Mann-Whitney U, Wilcoxon | Rank-biserial r, or r = Z / sqrt(N) | r = .30 | Drop |
The formatting rules
- Leading zeros: drop the leading zero for any statistic that cannot exceed 1 in absolute value (r, eta squared, partial eta squared, R-squared, Cramer's V, p). Keep it for statistics that can exceed 1 (Cohen's d, Hedges' g, F, t).
- Decimals: report effect sizes to two decimals.
- Placement: the effect size goes at the end of the result sentence, after the p value: t(58) = 2.41, p = .019, d = 0.63.
- Italics: italicize the symbols (d, r, R, V), but not the Greek names when you spell them out (eta squared, omega squared).
Interpreting small, medium, and large
Cohen offered rough benchmarks. Treat them as a starting point, not a rule, because what counts as "large" depends on your field.
| Effect size | Small | Medium | Large |
|---|---|---|---|
| Cohen's d | 0.20 | 0.50 | 0.80 |
| r (correlation) | .10 | .30 | .50 |
| Eta squared / partial eta squared | .01 | .06 | .14 |
| Cramer's V (df = 1) | .10 | .30 | .50 |
Two cautions. First, partial eta squared is not eta squared: partial eta squared removes other effects from the denominator, so it is usually larger. Name which one you report. Second, in an applied field a "small" effect can still be practically important, so interpret the number against your domain, not just the table.
Mistakes reviewers catch
- No effect size at all. The single most common request in peer review.
- Wrong leading zero. Writing d = .63 (should keep the zero: 0.63) or r = 0.34 (should drop it: .34).
- Confusing eta squared and partial eta squared, or not saying which you report.
- Treating the benchmarks as laws. They are conventions; interpret against your field.
- Reporting a standardized effect size but no means or descriptives, so the reader cannot see the raw picture.
Let KyroStat pick and format the effect size for you
The right effect size depends on the test, and the leading-zero rules catch people out. KyroStat runs your test on real Python or R, computes the matching effect size with a confidence interval where available, and writes it into the APA 7 sentence with the correct formatting. If you are still choosing a test, start with the decision guide. Upload your spreadsheet, and the report is done in seconds.
Frequently asked questions
Does Cohen's d get a leading zero? Yes. Cohen's d can exceed 1, so it keeps its leading zero: d = 0.62. Statistics bounded by 1, such as r and eta squared, drop it.
Is the correlation r an effect size? Yes. For a Pearson or Spearman correlation, r is both the test result and the effect size; you do not need a separate one.
Eta squared or omega squared for ANOVA? Both are acceptable. Omega squared is less biased in small samples, so it is often preferred, but eta squared and partial eta squared are more commonly reported.
Do I need a confidence interval on the effect size? It is increasingly expected, especially for Cohen's d and correlations. Report it when your software provides it.