ABSTRACT
Interactions play an important role in modeling. They are used when the levels of one predictor infl uence the response differently according to the levels of another predictor. For example, consider the continuous predictor age and a binary risk factor, sex. The response is death. If we model death on age and sex, we assume that the coeffi cients for both age and sex (male/female) are independent of one another. Moreover, the assumption of such a model, also called a main effects model, is that the coeffi cient, or slope, of age is constant across all values of age in the data, that is, the difference in odds or risk between males and females does not change with age. We know, though, that women tend to live longer than men do, so the odds of death at a certain age should be different between the two levels of sex. Creating an interaction of age and sex aims to refl ect this relationship.
