Effect size calculator: Cohen's d and Hedges' g
Work out Cohen's d and Hedges' g, with their confidence intervals, from each group's mean, standard deviation (SD) and number of people. d is the difference between two means in SDs. g is d corrected for its small bias in small groups.
The calculator runs in your browser. Its intervals come from the noncentral t distribution, as R's effectsize package works them out.
Work out d and g
Example figures. Type your own.
d = 0.48, 95% interval 0.02 to 0.93. g = 0.47, 95% interval 0.02 to 0.92.
| Effect size | Value | 95% interval |
|---|---|---|
| Cohen's d | 0.48 | 0.02 to 0.93 |
| Hedges' g | 0.47 | 0.02 to 0.92 |
Pooled SD 10.50. t = 2.10 with 76 degrees of freedom.
The formulas
| Figure | How it is worked out |
|---|---|
| Pooled SD | √[((n1 − 1) × SD1² + (n2 − 1) × SD2²) ÷ (n1 + n2 − 2)] |
| Cohen's d | (mean1 − mean2) ÷ pooled SD |
| Hedges' g | d × J, where J = Γ(df ÷ 2) ÷ (√(df ÷ 2) × Γ((df − 1) ÷ 2)) and df = n1 + n2 − 2. J is about 1 − 3 ÷ (4 × df − 1). |
| The interval | From the noncentral t distribution: the two noncentralities that put the groups' t at the interval's ends, each times √(1 ÷ n1 + 1 ÷ n2). The interval of g is the interval of d times J. |
d or g
d runs a little high in small groups, and g = d × J corrects it. The difference fades as the groups grow:
| People in all | J | g is smaller than d by |
|---|---|---|
| 10 | 0.903 | 9.7% |
| 20 | 0.958 | 4.2% |
| 40 | 0.980 | 2.0% |
| 100 | 0.992 | 0.8% |
Report g for small groups and for meta-analyses, and say which one you report.
What a d means
For two groups with normal scores and the same SD:
| d | Cohen's name | Share of group 1 above group 2's mean | Chance that a person from group 1 scores higher than one from group 2 |
|---|---|---|---|
| 0.2 | Small | 58% | 56% |
| 0.5 | Medium | 69% | 64% |
| 0.8 | Large | 79% | 71% |
Effect sizes from ConductExperience
- Results gives each number, slider and line scale question, and each score, its n, mean and SD, with n − 1 in the SD as here. Filter Results by group, and type each group's figures here.
- Every download has SPSS syntax and readers for R and Python, for the rest of the analysis.
Limits
- Two independent groups. For the same people measured twice, the d the sample size calculator asks for is dz: the mean of the differences divided by their SD. It is not a two-group d, and it needs each person's two scores.
- The interval assumes scores that are roughly normal, with the same SD in both groups. When the two SDs differ a lot, report them both.
- Cohen's names for the sizes are a last resort. Compare d with the effects in your own field.
Sources
- Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Lawrence Erlbaum Associates.
- Hedges, L. V. (1981). Distribution theory for Glass's estimator of effect size and related estimators. Journal of Educational Statistics, 6(2), 107–128.
Related
Sample size calculator
How many people a study needs for the effect size you expect.
Between and within subjects
When to compare groups of people, and when each person does every condition.
Random assignment generator
Assign people to groups at random, balanced or simple random.
Data export
One zip of tables, with a data dictionary entry for every column.
Planning a study?
Start one now, or tell us what you want to measure.