Interaction: additivity plus nonlinearity?

Abstract

Whether or not there is an interaction between two factors in their effects on a dependent variable is often a central question. This paper proposes a general mechanism by which an interaction may arise: (a) the two factors are the same thing — or, at least, have a dimension in common — in the sense that it is meaningful to add (or subtract) them; (b) the sum of them (or the difference between them) is what determines the dependent variable; and (c) the relation between the sum (or difference) and the dependent variable is nonlinear. For example, if several factors contribute to arousal in an additive manner, and the relationship of performance score to arousal is inverted-U, the factors will appear to interact in their joint effect on performance. Psychological, medical, and other scientists are likely to be unfamiliar with the (nonlinear) equations used to express this type of theory. Consequently, the task of promoting and interpreting such ideas will fall to the mathematician and statistician.