Skip to content
K Knidox Search…
Statistics · Inference

Effect Size Calculator

Cohen’s d for two independent groups, using the pooled standard deviation.

Group 1
Group 2
Try an example
Cohen’s d
0.333Small

Small effect. Pooled SD = 15; the group means differ by 5 on that scale.

The two groups as normal curves — more overlap means a smaller effect
Two overlapping normal distributions: group 1 centred at 105 and group 2 centred at 100; the more the curves overlap, the smaller Cohen’s d.40165
Group 1Group 2

Cohen’s d is (M₁ − M₂) ÷ pooled SD. With means 105 and 100, both SDs 15, and 30 per group, the pooled SD is 15, so d = 5 ÷ 15 = 0.33 — a small-to-medium effect. It scales the difference between two group means into standard-deviation units, so magnitude is comparable across studies.

What effect size measures

A p-value tells you whether a difference is likely real; an effect size tells you how big it is. Cohen’s d expresses the gap between two group means in standard-deviation units, so a d of 0.5 means the groups differ by half a standard deviation regardless of the original scale. Because it strips out the units, d lets you compare and combine results across studies — the reason it anchors meta-analysis and power analysis.

d = (M₁ − M₂) ÷ s_pooled

s_pooled = √( ((n₁−1)s₁² + (n₂−1)s₂²) ÷ (n₁+n₂−2) )

Worked example

Two groups: means 105 and 100, both SDs 15, 30 participants each.

  1. 1
    Pool the two standard deviations. s_pooled = √( ((30−1)·15² + (30−1)·15²) ÷ (30+30−2) ) = √( (6525 + 6525) ÷ 58 ) = √225 = 15.
  2. 2
    Subtract the means. M₁ − M₂ = 105 − 100 = 5.
  3. 3
    Divide by the pooled SD. d = 5 ÷ 15 = 0.333.
  4. 4
    Interpret the magnitude. |d| = 0.33 sits between the 0.2 (small) and 0.5 (medium) benchmarks, so it is a small-to-medium effect.

Cohen’s benchmarks for d

Rules of thumb, not hard cutoffs — interpret in the context of your field.

|d|LabelRough meaning
0.2Smallmeans differ by one-fifth of an SD
0.5Mediumvisible to the naked eye
0.8Largemeans differ by four-fifths of an SD
≥ 1.2Very largegroups barely overlap

Effect size versus p-value

Significance is not magnitude. A tiny, unimportant difference can be highly significant with a large enough sample, while a big, meaningful difference can miss significance in a small study. Report both: the p-value for whether an effect exists, and d for how large it is.

Sign and direction. The sign of d follows M₁ − M₂; only its absolute value determines the magnitude label. Swap the groups and the sign flips but the size does not.

Assumptions. The pooled-SD version assumes the two groups have similar variances. When spreads differ sharply, prefer Glass’s Δ (using one group’s SD) or Hedges’ g, which corrects the slight upward bias of d in small samples.

What is the difference between effect size and a p-value?
A p-value indicates whether a difference is likely real; effect size (Cohen’s d) indicates how large it is. With a big sample, a trivial difference can be statistically significant, so always report both.
What counts as a large effect?
By Cohen’s benchmarks, |d| ≈ 0.2 is small, 0.5 is medium, and 0.8 is large; values of about 1.2 or more are often called very large. Treat these as rough guides, not strict cutoffs.
Why is the pooled standard deviation used?
Pooling combines both groups’ variances into one estimate weighted by sample size, giving a common yardstick for the difference in means. It assumes the two groups have roughly equal variances.
Does the sign of d matter?
The sign just shows which group scored higher (it follows M₁ − M₂). The magnitude label depends only on the absolute value, so swapping the groups flips the sign but not the size.
When should I use Hedges’ g or Glass’s Δ instead?
Use Hedges’ g for small samples, since it corrects d’s slight upward bias, and Glass’s Δ when the group variances differ a lot, since it divides by one group’s SD rather than the pooled value.
How large a sample do I need?
Each group needs at least two observations to estimate a standard deviation, but small samples give unstable d estimates. Larger, balanced groups produce more reliable effect sizes.