MerQuR+Statistics Glossary

Statistics Glossary → Generalized eta squared (η²G)

Generalized eta squared (η²G)concept

A proportion-of-variance effect size designed to make effects comparable across between-subjects, within-subject and mixed designs.

Partial eta squared (η²p) divides an effect's sum of squares by the sum of that effect and its own error term. In within-subject designs the error term is small, because individual differences have already been removed, so η²p can make the same effect look much larger than it would in a between-subjects design. Generalized eta squared, proposed by Olejnik and Algina (2003), also adds to the denominator the variability due to subjects and to measured (not experimentally assigned) factors.

As a result, η²G gives similar values when the same effect is studied with different designs, which makes it better suited to meta-analysis and to comparisons across studies. In a fully between-subjects one-factor design, η²G equals both η² and η²p. Its value is usually smaller than η²p. The calculation requires stating correctly which factors are 'measured' (e.g. sex) and which are 'manipulated', because that choice changes the result.

An everyday example

Two researchers study the same caffeine effect: one splits 40 people into caffeine and no-caffeine groups, the other gives 20 people both conditions on different days. Because individual differences are removed in the second study, partial eta squared might come out at 0.40 instead of 0.10; that does not mean the effect is four times stronger. Generalized eta squared gives a similar value in both studies and makes the comparison fair.

In MerQur

It is requested with the 'Generalized η²' checkbox in Mixed-Design ANOVA and is also among the effect sizes in Repeated Measures ANOVA.

Analyses where this concept appears

Repeated Measures ANOVAMixed-Design ANOVA, Split-Plot

Related concepts

Eta squared (η²) and partial eta squared (η²ₚ)Within-subject (repeated-measures) factor