Course contentsShow
← Data Science
Lesson 582 of 2,14515. Regression DiagnosticsFree lesson

Variance Inflation Factor (VIF)

Calculate and interpret VIF values to quantify the severity of multicollinearity for each predictor.

Variance Inflation Factor (VIF)

What You'll Learn

You'll learn how to calculate and interpret VIF values to measure exactly how much multicollinearity is inflating the uncertainty in each predictor's coefficient.

The Core Idea

When predictors in a regression model are correlated with each other (multicollinearity), it becomes harder to estimate their individual effects precisely. The Variance Inflation Factor (VIF) quantifies this problem by measuring how much the variance of a coefficient estimate is "inflated" due to correlation with other predictors.

Think of it like trying to credit individual team members for a group project when they always work together. The more they're inseparable, the harder it is to measure each person's unique contribution—and the less confident you can be in your assessment.

Calculating VIF

For each predictor X_j, the VIF is calculated as:

VIF_j = 1 / (1 - R²_j)

Where R²_j is the R-squared from regressing X_j on all other predictors (not including the outcome variable).

  • If X_j is completely independent of other predictors, R²_j = 0, so VIF = 1 (no inflation)
  • If X_j is perfectly predicted by other predictors, R²_j approaches 1, so VIF approaches infinity

Interpreting VIF Values

  • VIF = 1: No correlation with other predictors (ideal)
  • VIF = 1–5: Moderate correlation (usually acceptable)
  • VIF = 5–10: High correlation (concerning, investigate further)
  • VIF > 10: Severe multicollinearity (action needed)

A VIF of 5 means the variance of that coefficient is 5 times larger than it would be if that predictor were uncorrelated with others.

Key Takeaway: VIF quantifies multicollinearity severity for each predictor, with values above 5–10 indicating problematic correlation that inflates coefficient uncertainty and reduces model reliability.