| Summary: | Two of the most challenging aspects of teaching regression analysis pertain to model selection and the relative importance of predictors. All-possible-subsets (APS) regression is a particularly useful tool for addressing both of these topics. When using regression analysis to analyze customer satisfaction data in marketing analytics courses, an Excel spreadsheet for implementing APS regression has led to fruitful discussions regarding: (1) which predictor variables should be retained in the regression model, and (2) which predictor variables are most useful for explaining the dependent variable. The spreadsheet, which uses a VBA macro to run APS regression via sweep operations on the correlation matrix, is scalable for up to p = 20 predictors and reports the best subset for all subset sizes on the interval 1 ≤ q ≤ p. The selection of an appropriate value for q is facilitated by R2 and Mallows’ Cp information. Relative predictor importance is established via a general dominance measure obtained from the APS regression.
|
|---|
