In PROC HPREG diagnostics, a predictor has VIF = 18. What does this indicate and what is the typical remediation?
Select an answer to reveal the explanation.
Short Explanation
Here's the deal — a VIF of 18 means the predictor's coefficient variance is being badly inflated by multicollinearity. You're way past the usual 5-to-10 warning zone, so you should drop one correlated predictor, combine them, or use ridge regression.
Full Explanation
In PROC HPREG diagnostics, VIF measures how much a coefficient's variance is inflated by multicollinearity. VIF = 18 far exceeds the common threshold of 5 to 10, indicating severe multicollinearity. Standard remediation includes removing one correlated predictor, creating composites, or applying ridge regression. The claim that VIF = 18 means 18% unexplained variance and no action is needed is wrong because VIF is not a percentage. The claim that VIF = 18 is below a threshold of 20 and shows no concern is wrong because VIF above 10 is widely considered severe. The claim that VIF = 18 indicates heteroscedasticity and requires weighted least squares is wrong because VIF diagnoses multicollinearity, not heteroscedasticity.