Accuracy and Precision of Generalized Eta Squared: An Empirical Investigation of Research Design and Nonnormality

• In Event: Advances in Statistical Methodology

Abstract

Generalized eta-squared has been proposed as a variance-explained effect size that is comparable across a variety of ANOVA designs, including any number of between- and within-subjects factors. A simulation study was conducted to investigate the accuracy and precision of this effect size across designs, as well as its sensitivity to non-normal population distribution shapes. Factors investigated in the simulation included sample size, distribution shape, population eta-squared value, and ANOVA design characteristics (number and levels of between- and within-subject factors). Results indicate substantial bias in effect size estimates with small samples and greater bias in estimates associated with interaction effects than those associated with main effects. Greater bias was evident in skewed distributions, but kurtosis had little impact on the estimates.

Authors

• Patricia Rodriguez de Gil, University of South Florida

• Patrice S. Rasmussen, University of South Florida

• Thanh Vinh Pham, University of South Florida

• Jeffrey D. Kromrey, University of South Florida

• Anh P Kellermann, University of South Florida

• Jeanine L. Romano, American University of Sharjah

• Yi-Hsin Chen, University of South Florida

• Isaac Li, University of South Florida