Effect size: what it is and how to report it in APA 7 (with examples)
Picture two theses. In the first, with 2,000 respondents, the satisfaction gap between two sites is significant (p < .001), but it’s 0.05 points on a 1 to 5 scale. In the second, with 30 people, the gap is a full point and p = .08. If you only look at p, the first one seems more important. It isn’t.
The p-value depends heavily on sample size: with thousands of cases, almost everything is significant. Effect size doesn’t depend on sample size and answers the question that matters: how big is the difference or relationship?
What APA 7 says
The seventh edition of the APA manual asks you to report effect size and, where possible, its confidence interval next to every hypothesis test. Many journals and more and more graduate schools require it. A thesis that reports only p-values gets the comment sooner or later.
Which measure goes with each test
- Independent-samples t-test: Cohen’s d, or Hedges’ g with small groups.
- Paired t-test: Cohen’s dz.
- Mann-Whitney U and Wilcoxon: r = Z / √N.
- ANOVA: eta squared (η²) or, better, omega squared (ω²).
- Kruskal-Wallis: epsilon squared (ε²).
- Friedman: Kendall’s W.
- Chi-square test of independence: Cramér’s V (phi in 2 × 2 tables).
- Correlation: r or rho itself is the effect size.
- Regression: R² and adjusted R².
How to interpret it
The most cited cutoffs are Cohen’s (1988). They’re general guides, not rules: in some fields a “small” effect can matter a lot.
- Cohen’s d: 0.2 small, 0.5 medium, 0.8 large.
- r: 0.1 small, 0.3 medium, 0.5 large.
- η²: 0.01 small, 0.06 medium, 0.14 large.
- Cramér’s V: depends on table size. In a 2 × 2 table the cutoffs are 0.1, 0.3 and 0.5; in larger tables they drop.
One way to explain d in words is the probability of superiority: with d = 0.5, a person from the first group has a 64% chance of scoring higher than one from the second.
How to write it in APA 7
It goes in the same sentence as the test, after the p-value. Statistics are italicized (t, F, p, d, r) with no leading zero when the value can’t exceed 1 (p, r, η²).
- t-test: Private school teachers reported more stress (M = 3.8, SD = 0.7) than public school teachers (M = 3.4, SD = 0.8), t(85) = 2.49, p = .015, d = 0.53, 95% CI [0.11, 0.96].
- Mann-Whitney: The two groups differed significantly, U = 612.5, Z = −2.31, p = .021, r = .25.
- ANOVA: Workplace climate differed across the three sites, F(2, 117) = 4.12, p = .019, η² = .07.
- Chi-square: Intention to quit was associated with shift, χ²(2, N = 150) = 9.60, p = .008, V = .25.
If you can, add the confidence interval for the effect size. A d of 0.53 with an interval from 0.11 to 0.96 says the effect could be anywhere from small to large: with a small sample the estimate is imprecise, and it’s better to say so in the discussion than to hide it.
After the number, interpret it in words: “a medium effect”. That’s what the committee reads first.
If your software doesn’t compute it
Since version 27, SPSS computes Cohen’s d in the t-test and eta squared in ANOVA if you tick the option. For r from a Mann-Whitney or Wilcoxon Z, or Cramér’s V from an existing table, you can get it from the numbers already in your output.
Enter your means or your test result and get the effect size with a ready-made sentence:
Effect size calculator (free)Common mistakes
- Confusing significant with important. An effect can be significant and trivial at the same time.
- Reading the sign of d as size. The sign only tells you which group has the higher mean.
- Reporting SPSS’s η² as plain η² in multi-factor designs: SPSS gives partial η².
- Using Cohen’s cutoffs without thinking about your field. If earlier studies used the same instrument, compare your effect with theirs.
If you’re still unsure which test fits:
Which statistical test to use in your thesis