Which VIF value is commonly cited as a threshold indicating potential multicollinearity concerns?

Which VIF value is commonly cited as a threshold indicating potential multicollinearity concerns?

• A. Greater than 1
• B. Greater than 5
• C. Greater than 10
• D. Greater than 20

Answer: (C)

VIF shows how much the variance of a regression coefficient is inflated because that predictor is related to the others. It’s calculated by examining how well a predictor can be predicted from the rest: the closer to a perfect linear relationship, the higher the R-squared in that auxiliary regression, and the higher the VIF (VIF = 1 / (1 − R²)). A value commonly cited as a threshold for potential multicollinearity concerns is greater than 10. When VIF exceeds 10, the standard errors of the affected coefficients tend to be large, making it hard to determine the unique contribution of that predictor and leading to unstable estimates. This is a conservative rule of thumb that signals you should investigate further. In practice, some researchers flag moderate concern at lower cutoffs, such as VIFs above 5, but the typical benchmark for “serious” multicollinearity is a VIF above 10. If you see VIFs above this level, you might consider removing or combining predictors, or using techniques like regularization or principal component approaches to mitigate the issue.

VIF shows how much the variance of a regression coefficient is inflated because that predictor is related to the others. It’s calculated by examining how well a predictor can be predicted from the rest: the closer to a perfect linear relationship, the higher the R-squared in that auxiliary regression, and the higher the VIF (VIF = 1 / (1 − R²)).

A value commonly cited as a threshold for potential multicollinearity concerns is greater than 10. When VIF exceeds 10, the standard errors of the affected coefficients tend to be large, making it hard to determine the unique contribution of that predictor and leading to unstable estimates. This is a conservative rule of thumb that signals you should investigate further.

In practice, some researchers flag moderate concern at lower cutoffs, such as VIFs above 5, but the typical benchmark for “serious” multicollinearity is a VIF above 10. If you see VIFs above this level, you might consider removing or combining predictors, or using techniques like regularization or principal component approaches to mitigate the issue.