Course contentsShow
← Mathematics
Lesson 3080 of 3,12160. Statistics — ModelingFree lesson

Multicollinearity: Detection and Consequences

Identify multicollinearity using variance inflation factors (VIF); understand its effect on coefficient estimates and stability.

Multicollinearity: Detection and Consequences

You'll learn how to spot when predictor variables are too correlated and why that creates problems in multiple regression.

What Is Multicollinearity?

Multicollinearity occurs when two or more predictor variables in a multiple regression model are highly correlated with each other. Think of it like trying to determine whether rain or humidity causes people to carry umbrellas—these two factors move together so closely that it's nearly impossible to separate their individual effects.

When predictors are strongly correlated, the regression struggles to isolate each variable's unique contribution to predicting the outcome. The mathematical matrix operations become unstable, like trying to balance on a wobbling table.

Detecting Multicollinearity: Variance Inflation Factor (VIF)

The Variance Inflation Factor quantifies how much the variance of a regression coefficient is inflated due to multicollinearity. For each predictor:

VIF = 1 / (1 - R²ⱼ)

where R²ⱼ is the R² from regressing predictor j on all other predictors.

Rule of thumb:

  • VIF = 1: No correlation with other predictors
  • VIF > 5: Moderate multicollinearity (concern)
  • VIF > 10: Severe multicollinearity (serious problem)

Consequences of Multicollinearity

  1. Inflated standard errors: Coefficient estimates become less precise
  2. Unstable coefficients: Small changes in data can produce wildly different estimates
  3. Misleading p-values: Individual predictors may appear non-significant even when the overall model is significant
  4. Unreliable interpretation: Can't trust which variables truly matter

Importantly, multicollinearity doesn't bias your predictions—the model can still predict well—but it destroys your ability to interpret individual coefficients reliably.

Key Takeaway: Use VIF to detect multicollinearity (VIF > 5–10 signals trouble); high multicollinearity inflates standard errors and makes coefficient estimates unstable and untrustworthy, though predictions may remain accurate.