Effect size

Cohen's d calculator

Calculate the standardized mean difference between two independent groups. Enter means, SDs and sample sizes, or a Student's t with the two group sizes, and get Cohen's ds, Hedges' g, d* and Glass's delta, each labeled.

  • Hedges' g with the small-sample correction
  • From summaries or from t
  • Each variant named, as journals ask

Sources checked on November 21, 2026. Results computed by the site's engine and tested against R and SciPy.

Input

Standardized mean difference

Cohen's ds−2.1312
Hedges' g−2.0577
Correction J0.9655

Hedges' g of −2.06 is above the large benchmark in Cohen's rough conventions (0.2 small, 0.5 medium, 0.8 large).

Pooled SD3.7538
d* (average variance)−2.1909
Glass's delta (SD 2)−1.9124
Glass's delta (SD 1)−2.6423
Reproduce this result in R or Python

R

sp <- sqrt(((10 - 1) * 3.0276503541^2 + (14 - 1) * 4.1833001327^2) / (10 + 14 - 2))
d <- (5.5 - 13.5) / sp
g <- d * (1 - 3 / (4 * (10 + 14) - 9))
c(d = d, g = g)

Python (SciPy)

m1, s1, n1, m2, s2, n2 = 5.5, 3.0276503541, 10, 13.5, 4.1833001327, 14
sp = (((n1 - 1) * s1**2 + (n2 - 1) * s2**2) / (n1 + n2 - 2)) ** 0.5
d = (m1 - m2) / sp
g = d * (1 - 3 / (4 * (n1 + n2) - 9))
print(d, g)

What Cohen's d measures

Cohen's d expresses a difference between means in standard deviation units, so results from different scales can be compared. A d of 0.5 means the group means are half a standard deviation apart. Unlike a p-value, it does not grow with sample size: a tiny difference can be highly significant in a huge study and still have a small d.

Worked example

The calculator opens with two small groups from R's t.test example: the numbers 1 to 10 and the numbers 7 to 20. Their means are 5.50 and 13.50, with SDs 3.0277 and 4.1833. The pooled SD is 3.7538, so Cohen's ds = −8.00 / 3.7538 = −2.1312. The small-sample correction J = 0.9655 gives Hedges' g = −2.0577.

Switch to "From t and n" and enter Student's t = −5.1473 with n = 10 and 14: ds = t × √(1/n1 + 1/n2) gives the same −2.1312. Our test suite checks that the two routes agree to twelve decimals.

The variants and when to use each

MeasureDivides the difference byUse it when
Cohen's dsThe pooled SDTwo independent groups, a standard default
Hedges' gThe pooled SD, then multiplies by JSmall samples; the usual choice for meta-analysis
d* (average variance)The square root of the mean of the two variancesYou prefer not to weight by group size, for example alongside Welch's test
Glass's deltaThe control group's SD onlyA treatment changes the spread as well as the mean
Cohen's dzThe SD of the paired differencesPaired data; shown by the paired calculator

Cohen's ds, Hedges' g, Glass's delta and dz are all set out in Lakens (2013), the clearest single reference; d* simply swaps the pooled SD for the square root of the average of the two variances, so unequal group sizes do not change the denominator. For this example Glass's delta with the first group as control is −2.6423, larger in size than g because the first group's SD is the smaller one.

How big is big?

Cohen suggested 0.2, 0.5 and 0.8 as small, medium and large, and he called them rough guides for when nothing better is known. A d of 0.2 in a cheap public health measure may matter more than a d of 0.8 in a lab task. Compare your effect with others in your field where you can. The calculator places your g against the conventions in plain words; in this example −2.06 is above the large benchmark.

Effect sizes with Welch's test

Welch's t does not pool the variances, so d from a Welch t is not well defined by the formula above. The usual practice is to report Welch's t, df and p for the test, and Hedges' g or d* for the effect, naming which. The independent samples calculator shows all of them together.

What is not here yet

A confidence interval for d needs the noncentral t distribution. It is planned for a later version; until then, report the confidence interval of the raw mean difference, which every t-test result on this site includes.

Sources