MerQuR+Statistics Glossary

Statistics Glossary → Hedges' g

Hedges' gconcept

A standardized mean difference that removes the small-sample upward bias of Cohen's d by means of a correction factor.

Cohen's d divides the difference between two means by the pooled standard deviation, expressing the effect in standard-deviation units. In small samples, however, d slightly overestimates the true population effect on average. Hedges' g removes this bias with the correction factor J: g = d · J. J is computed exactly with the gamma function, and a common approximation is J ≈ 1 - 3 / (4·df - 1), where df is n₁ + n₂ - 2 for two independent groups.

J is always below 1 and approaches 1 as the sample grows, so g and d differ in small samples but are almost identical in, say, studies with more than 40 participants in total. It is interpreted like Cohen's d (roughly 0.2 small, 0.5 medium, 0.8 large, although these benchmarks vary by field). In meta-analyses that pool studies of different sizes, g is preferred to d. In paired designs the correction is applied to d_z with n - 1 degrees of freedom. It should not be confused with Glass's Δ, which divides the difference by the control group's standard deviation only.

An everyday example

A teacher tries a new reading method in two small classes of 8 pupils each and finds Cohen's d = 0.60. With classes this small, d slightly exaggerates the effect; with the factor J ≈ 1 - 3 / (4·14 - 1) ≈ 0.945, Hedges' g ≈ 0.57. Had the same difference been found in classes of 200 pupils, the gap between d and g would have all but vanished.

In MerQur

It is requested with the 'Hedges g' checkbox in the one-sample, independent-samples and paired t-tests. The Permutation Test gives Hedges g = d·J together with Cohen d, and the Effect Size Calculator computes Hedges g with the exact correction factor.

Analyses where this concept appears

One-Sample t-TestIndependent-Samples t-TestPaired t-TestPermutation TestEffect Size Calculator

Related concepts

Cohen's d (effect size)Glass's Δ