Variance Inflation Factor (VIF)
Also known as: VIF, Variance Inflation Index, Multicollinearity Inflation Factor, Varyans Enflasyon Faktörü
The Variance Inflation Factor (VIF) is a scalar diagnostic statistic proposed by Donald Marquardt (1970) that quantifies how much the variance of an estimated regression coefficient increases due to linear dependence—multicollinearity—among the predictors in an ordinary least squares model. It is routinely applied in econometrics, social science, and biomedical research whenever analysts suspect that two or more independent variables move together closely enough to destabilize coefficient estimates.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Apply VIF after fitting an OLS or generalized linear model whenever two or more continuous or dummy-coded predictors may be linearly related. Prerequisites include a linear model specification, continuous or binary predictors, and residuals that at least approximately satisfy OLS assumptions. VIF is not suited to pure interaction or polynomial terms without centering, and it does not address nonlinear dependencies. When VIF signals severe multicollinearity, consider ridge regression, principal component regression, variable selection, or collecting more data as remedies.
Strengths & limitations
- Simple to compute and interpret: a single number per predictor summarizes collinearity severity.
- Model-agnostic in spirit: applicable to OLS, logistic, and other GLM contexts with appropriate extensions.
- Directly linked to coefficient variance inflation, giving it a clear statistical interpretation.
- Widely implemented in all major statistical software, ensuring reproducibility across teams.
- Requires an a priori fitted regression model; it is a post-fit diagnostic, not a pre-screening tool.
- Sensitive to model specification: including irrelevant variables or omitting important ones distorts VIF values.
- Conventional thresholds (5 or 10) are rules of thumb with no universal theoretical justification.
- Does not identify which subset of predictors causes the collinearity—only that a given predictor is involved.
Frequently asked
What VIF value indicates a problem?
There is no universally agreed threshold. A VIF of 1 indicates no collinearity; values between 1 and 5 are generally considered acceptable; values between 5 and 10 suggest moderate collinearity that may warrant attention; values above 10 are widely treated as evidence of severe multicollinearity that can meaningfully inflate standard errors and destabilize coefficient estimates. Some strict disciplines use 5 as the cutoff for concern.
Does a high VIF mean my regression coefficients are wrong?
No. Multicollinearity does not bias OLS coefficient estimates under the classical assumptions—the estimates remain unbiased and consistent. What high VIF signals is that the standard errors of the affected coefficients are large, reducing statistical power and making individual coefficients statistically fragile. The overall model fit (R², F-statistic) can still be valid and reliable even when individual VIFs are high.
How does VIF relate to the condition index?
Both diagnose multicollinearity but from different vantage points. VIF is predictor-specific: it quantifies how much variance inflation affects one coefficient at a time. The condition index is matrix-level: it examines the eigenvalues of the scaled X'X matrix to detect near-linear dependencies that may involve several predictors simultaneously. Using both together gives a more complete picture of collinearity structure than either alone.
Sources
- Marquardt, D. W. (1970). Generalized inverses, ridge regression, biased linear estimation, and nonlinear estimation. Technometrics, 12(3), 591–612. DOI: 10.1080/00401706.1970.10488699 ↗
How to cite this page
ScholarGate. (2026, June 2). Variance Inflation Factor (VIF). ScholarGate. https://scholargate.app/en/econometrics/variance-inflation-factor
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Condition IndexEconometrics↔ compare
- OLS RegressionEconometrics↔ compare
- Ridge RegressionMachine learning↔ compare