ANOVA versus Regression Analysis of ATI Designs: An Empirical Investigation

Abstract

The theory indicating that general linear model methods can be superior to ANOVA analogs is well known. However, the magnitudes of differences induced by different analysis choices has not been empirically investigated. The paper reports a Monte Carlo study of these differences over nine combinations of three sample sizes and three population parameter effect sizes. Generally, ANOVA methods tended somewhat to overestimate smaller effect sizes and to underestimate larger effect sizes.