ABSTRACT
To increase the predictive power of a model beyond that provided by its components, data analysts create an interaction variable, which is the product of two or more component variables. This chapter explores chi-squared automatic interaction detection (CHAID) as an alternative data mining method for specifying a model, thereby justifying the omission of the component variables under certain circumstances. Database marketing provides an excellent example of the proposed data mining method. The chapter illustrates the alternative method with a response model case study. The popular strategy for modeling with interaction variables is the principle of marginality, which states that a model including an interaction variable should also include the component variables that define the interaction. A significance test, which requires a unique partitioning of the dependent variable variance regarding the interaction variable and its components, is not possible due to the multicollinearity.
