Date created: | Last Updated:

Category: Software

Description: Statistically speaking, an interaction effect is observed whenever the impact of a factor changes based on the levels of another factor. The interaction effect is of special interest in psychology, as typical experimental designs involve the comparison between more than one factor, like groups or conditions. When a statistically significant interaction is found, it is usually followed by post-hoc pairwise comparisons performed via t-test or univariate Anova. Despite both procedures have been criticized as erroneous methods to explain an interaction effect, they are still largely used in psychological research. In the present paper, we aim to highlight the weaknesses of post-hoc pairwise comparisons and discuss the use of two alternative methods to empower the interpretation and visualization of interaction effects. The first, is an explorative method based on the estimation approach, consisting in the comparison of confidence intervals of the estimated marginal means. The second, is a hypothesis-driven approach based on the analysis of specific ordering of means via Bayesian Informative hypothesis, which can overcome the limitations of planned comparisons.