A property-space perspective for interaction terms of ordinal variables

Summary

It is recommended that before creating an interaction term based on two or more variables, a property-space perspective be employed using a graphic technique such as partial-order scalogram analysis to examine the empirical distribution of the observations in an interaction property space. The empirical space should then be compared with the theoretical description of the interaction. Once this comparative stage of analysis has been completed, the researcher is in a better position to decide which form of interaction term should be used to express the theoretical construct of interaction for specific statistical models. If the interaction property space is close to unidimensional, then an additive or multiplicative term may be adequate; but if the interaction property space is not unidimensional, it is necessary to control for type either statistically or by partitioning the data set according to type. Type, however, must be relevant theoretically.