A variance inflation factor (VIF) measures how much multicollinearity among predictors increases the variance of an estimated regression coefficient. section 5.2.2
Use VIF to guide predictor selection in regression models:
- Treat VIF values of 1 or more as evidence that some predictor collinearity exists. section 5.2.2
- Prioritize remediation when one or more predictors has a VIF of 10 or more. section 5.2.2
- Calculate a predictor’s VIF by regressing it on the remaining predictors, obtaining R², and using that result in the VIF calculation. section 5.2.2
- Consider discarding offending predictors and refitting a reduced model. section 5.2.2
Raisins logistic regression
Regressing MinorAxisLength on the other predictors produces R² = 0.9751 and VIF = 40.21. Automatic calculation returns high VIFs for Area, MajorAxisLength, MinorAxisLength, and ConvexArea. section 5.2.2
