A new model for visualizing interactions in analysis of variance

In analysis of variance, there is usually little attention for interpreting the terms of the effects themselves, especially for interaction effects. One of the reasons is that the number of interaction-effect terms increases rapidly with the number of predictor variables and the number of categories. In this paper, we propose a new model, called the interaction decomposition model, that allows to visualize the interactions. We argue that with the help of the visualization, the interaction-effect terms are much easier to interpret. We apply our method to predict holiday spending1 using seven categorical predictor variables.