Determining which variables in a model are its most important predictors (in ranked order) is a vital element in the interpretation of a linear regression model. A general rule is to view the predictor variable with the largest standardized regression coefficient as the most important variable; the predictor variable with the next largest standardized regression coefficient as the next important variable, and so on. This rule is intuitive, easy to apply and provides practical information for understanding how the model works. Unknown to many, however, is that the rule is theoretically problematic. Thus, the purpose of this chapter is twofold: first, to discuss why the decision rule is theoretically amiss, yet works well in practice; second, to present an alternative measure — the
predictive contribution coefficient
— which offers greater utile information than the standardized coefficient, as it is an assumption-free measure founded in the data mining paradigm.