Eta squared (η²) and omega squared (ω²)

η² is the share of the differences in the measured value that group membership explains (0 to 1). ω² is the same idea, bias-corrected for the overestimation in the sample, and comes out slightly smaller. The confidence interval gives the plausible range for η² in the population. For repeated measures: partial η² (ηp²) relates the effect to effect plus its own error; generalized η² (ηG²) includes all error components and is easier to compare between different designs.

Example: F(2, 57) = 67.42, p < .001, η² = .70, 95% CI [.57, .79]: dose explains about 70% of the differences in tooth length; according to these data, 57 to 79% are plausible.

Caution: Cohen's benchmarks: from .01 small, from .06 medium, from .14 large; they apply to η² from group comparisons. With repeated measures Statable therefore classifies by ηG² - ηp² is usually considerably larger there (Bakeman 2005; Lakens 2013). The analysis of variance only shows that at least two groups differ, not which ones. The lower limit of the interval is 0 when the test is not significant.