What does the variance inflation factor measure?

What does the variance inflation factor measure?

• A. A measure of multicollinearity
• B. A measure of central tendency
• C. A test for normality
• D. A statistic for model fit

Answer: (A)

The variance inflation factor shows how much the variance of a regression coefficient is inflated because predictors are correlated with each other. In practical terms, it tells you how much less precise the estimate of a predictor’s effect becomes due to multicollinearity among the predictors. For each predictor, VIF is derived from regressing that predictor on all the other predictors; it equals 1 divided by 1 minus that model’s R-squared. If a predictor isn’t related to the others, its VIF is about 1. If it’s highly predictable from the other predictors, its VIF becomes large, indicating inflated standard errors for that coefficient. This is a diagnostic for multicollinearity, not a measure of central tendency, normality, or overall model fit. Thresholds like 5 or 10 are often used to flag potential problems. If you see high VIFs, you might consider removing or combining predictors, or using regularization.

The variance inflation factor shows how much the variance of a regression coefficient is inflated because predictors are correlated with each other. In practical terms, it tells you how much less precise the estimate of a predictor’s effect becomes due to multicollinearity among the predictors.

For each predictor, VIF is derived from regressing that predictor on all the other predictors; it equals 1 divided by 1 minus that model’s R-squared. If a predictor isn’t related to the others, its VIF is about 1. If it’s highly predictable from the other predictors, its VIF becomes large, indicating inflated standard errors for that coefficient.

This is a diagnostic for multicollinearity, not a measure of central tendency, normality, or overall model fit. Thresholds like 5 or 10 are often used to flag potential problems. If you see high VIFs, you might consider removing or combining predictors, or using regularization.