|Friday, February 21|
|CS02 Interpreting Analyses||
Fri, Feb 21, 9:15 AM - 10:45 AM
Information Value Statistic and Predictors for Logistic Regression (302714)*Bruce Stephen Lund, Marketing Associates, LLC
Keywords: informative value statistic, IV, predictor variable binning, logistic regression
In preparing predictor variables for binary logistic regression, it is good practice to collapse the levels of a predictor X to achieve parsimony while maintaining predictive power. In the first section of the talk, an algorithm is given as well as a SAS® macro for collapsing the levels of X. If numeric, the ordering of X can be maintained by collapsing only adjacent levels of X. Otherwise, all pairs of levels are considered for collapsing. The modeler has the choice of two criteria for collapsing: (a) maximizing information value (IV) or (b) maximizing log likelihood (LL). Stopping guidelines are provided. Several examples are given. In one example, the algorithm is used to find interactions of two predictors. In the final section of the talk, the IV statistic is discussed. Familiar, but perhaps mysterious, guidelines for deciding if the IV of a predictor X is high enough to use in modeling are given in many textbooks. To provide insight into IV guidelines, the IV is compared to other measures of predictive power.