Effect Size Calculator
Calculate the standardized effect size for a difference between two groups from their means and standard deviations, or from a reported t statistic. The calculator gives Cohen’s d, Hedges’ g and Glass’s Δ with exact confidence intervals, and converts the result into a correlation, an odds ratio and the probability of superiority so you can report the measure your field expects.
Cohen’s d
95% CI −0.10 to 1.10
A medium effect by Cohen’s benchmarks: the group means differ by 0.50 standard deviations, but the interval includes 0.
- Cohen’s d
- 0.501
- 95% CI −0.099 to 1.096
- Hedges’ g (bias-corrected)
- 0.492
- 95% CI −0.097 to 1.076
- Glass’s Δ (control SD)
- 0.421
- Pooled SD
- 1.5967
- Standard error of d
- 0.3046
- df = 43
- Correlation r (point-biserial)
- 0.243
- equal group sizes
- r² (variance explained)
- 5.9%
- Odds ratio equivalent
- 2.48
- exp(dπ/√3)
- Probability of superiority
- 63.8%
- common-language effect size
- Cohen’s U3
- 69.2%
- group 1 above the group 2 mean
- Overlap of distributions
- 80.2%
- NNT (Kraemer)
- 3.6
Report it like this
Group 1 (M = 5.2, SD = 1.1, n = 20) scored higher than group 2 (M = 4.4, SD = 1.9, n = 25), Cohen’s d = 0.50, 95% CI −0.10 to 1.10 (Hedges’ g = 0.49).
How to use the calculator
- Means and SDs: enter the mean, standard deviation and sample size of each group. Put the treatment or exposed group first, so a positive d means it scored higher.
- t statistic: use this when a paper reports t and the group sizes but not the SDs, for example to include it in a meta-analysis. Choose “Paired” for a within-subject t-test.
- Convert: enter a d, a correlation r or an odds ratio to see the equivalent values of the other measures.
For the full t-test with its p-value from the same numbers, use the t-test calculator. To find the sample size needed to detect an effect of a given size, use the power analysis calculator.
Which effect size measure?
- Cohen’s d divides the difference in means by the pooled standard deviation. It is the most widely reported measure for two groups.
- Hedges’ g multiplies d by a small-sample correction, J = 1 − 3/(4df − 1). It removes the slight upward bias of d in small samples and is the usual choice in meta-analysis. With 20 per group, g is about 2% smaller than d.
- Glass’s Δ divides by the control group’s SD only. Use it when the treatment is expected to change the spread of scores as well as the mean.
- Cohen’s dz is the paired version: the mean difference divided by the SD of the differences. It is usually larger than a d based on the SD of the scores, so name it explicitly.
- r and the odds ratio express the same effect on other scales, which helps when combining studies that report different measures.
Formulas
spooled = √[((n₁ − 1)s₁² + (n₂ − 1)s₂²) / (n₁ + n₂ − 2)]
d = t × √(1/n₁ + 1/n₂) (paired: dz = t / √n)
The confidence interval for d is found by inverting the noncentral t distribution: the calculator finds the noncentrality values for which the observed t sits at the 2.5th and 97.5th percentiles, then rescales them to d (Cumming and Finch 2001). This is the method used by the R packages MBESS and effectsize, and it is more accurate in small samples than the normal approximation SE = √[(n₁ + n₂)/(n₁n₂) + d²/(2(n₁ + n₂))], which the calculator also reports.
Conversions assume normal distributions and equal group sizes: r = d/√(d² + 4); odds ratio = exp(dπ/√3) (Chinn 2000); probability of superiority = Φ(d/√2) (McGraw and Wong 1992); U3 = Φ(d); NNT = 1/(2Φ(d/√2) − 1) (Kraemer and Kupfer 2006).
Worked examples
- Two groups: 5.2 (SD 1.1, n = 20) vs 4.4 (SD 1.9, n = 25). Pooled SD = √[(19 × 1.21 + 24 × 3.61)/43] = 1.597, so d = 0.8/1.597 = 0.50 (95% CI −0.10 to 1.10), Hedges’ g = 0.49 and Glass’s Δ = 0.8/1.9 = 0.42. The interval includes 0, so the study cannot rule out no effect.
- From t: t(58) = 2.50 with 30 per group gives d = 2.50 × √(2/30) = 0.65 (95% CI 0.12 to 1.16).
- Paired: t(7) = 3.99 from 8 patients gives dz = 3.99/√8 = 1.41 (95% CI 0.39 to 2.39).
- Converting: r = .30 corresponds to d = 0.63 and an odds ratio of 3.13; an odds ratio of 2 corresponds to d = 0.38.
How big is a small, medium or large effect?
Cohen (1988) suggested 0.2, 0.5 and 0.8 as small, medium and large effects when nothing better is available. On the other scales they correspond to:
| Label | d | r | Odds ratio | Probability of superiority |
|---|---|---|---|---|
| Small | 0.2 | .10 | 1.44 | 56% |
| Medium | 0.5 | .24 | 2.48 | 64% |
| Large | 0.8 | .37 | 4.27 | 71% |
Cohen himself warned against using these labels mechanically. A d of 0.2 can matter a great deal for a cheap intervention applied to millions of people, and effects in some fields are routinely smaller than in others. Compare your effect with those found in similar studies, and always report the confidence interval, which shows how precisely the effect was estimated (Lakens 2013).
Participants in the intervention group scored higher (M = 5.20, SD = 1.10, n = 20) than controls (M = 4.40, SD = 1.90, n = 25), Cohen’s d = 0.50, 95% CI −0.10 to 1.10 (Hedges’ g = 0.49).
Questions
- How do I calculate an effect size?
- For two groups, subtract one mean from the other and divide by the pooled standard deviation. That gives Cohen’s d. With means of 5.2 and 4.4 and a pooled SD of 1.6, d = 0.8 ÷ 1.6 = 0.50, a medium effect.
- What is the difference between Cohen’s d and Hedges’ g?
- Both divide the mean difference by a pooled SD. Hedges’ g multiplies d by a correction factor, 1 − 3/(4df − 1), because d slightly overestimates the population effect in small samples. With large samples the two are almost identical. Meta-analyses usually report g.
- How do I calculate an effect size from a t value?
- For two independent groups, d = t × √(1/n₁ + 1/n₂). For a paired t-test, dz = t ÷ √n, where n is the number of pairs. For example, t(58) = 2.50 with 30 people per group gives d = 0.65.
- How do I convert r to Cohen’s d?
- Use d = 2r ÷ √(1 − r²), which assumes two groups of equal size. A correlation of .30 corresponds to d = 0.63. In the other direction, r = d ÷ √(d² + 4).
- How do I convert an odds ratio to Cohen’s d?
- Multiply the natural log of the odds ratio by √3/π (about 0.551). An odds ratio of 2 corresponds to d = 0.38. This logistic approximation (Chinn 2000) is widely used to combine binary and continuous outcomes in a meta-analysis.
- Should I report the effect size if the result is not significant?
- Yes. The effect size and its confidence interval show what the data are compatible with, which a p-value above .05 does not. A wide interval that includes 0 and large effects means the study was too small to decide, not that there is no effect.
Sources
- Cohen J. Statistical Power Analysis for the Behavioral Sciences. 2nd ed. Hillsdale, NJ: Lawrence Erlbaum; 1988. doi:10.4324/9780203771587
- Hedges LV. Distribution theory for Glass’s estimator of effect size and related estimators. J Educ Stat 1981;6(2):107–128. doi:10.3102/10769986006002107
- Lakens D. Calculating and reporting effect sizes to facilitate cumulative science: a practical primer for t-tests and ANOVAs. Front Psychol 2013;4:863. doi:10.3389/fpsyg.2013.00863
- Chinn S. A simple method for converting an odds ratio to effect size for use in meta-analysis. Stat Med 2000;19(22):3127–3131.
- McGraw KO, Wong SP. A common language effect size statistic. Psychol Bull 1992;111(2):361–365. doi:10.1037/0033-2909.111.2.361
- Cumming G, Finch S. A primer on the understanding, use, and calculation of confidence intervals that are based on central and noncentral distributions. Educ Psychol Meas 2001;61(4):532–574. doi:10.1177/0013164401614002
- Kraemer HC, Kupfer DJ. Size of treatment effects and their importance to clinical research and practice. Biol Psychiatry 2006;59(11):990–996. doi:10.1016/j.biopsych.2005.09.014
Calculations run in your browser; nothing you enter is sent or stored. Results are checked against SciPy and statsmodels in our test suite. Last updated 22 November 2026.