Variance Inflation Factor (VIF) calculator

Calculate VIF values to diagnose multicollinearity in multiple regression model predictors online.

What it does

Measures how much a predictor's variance is inflated by its correlation with the other predictors — how much of it is already explained by the rest of the model.

When to use it

After fitting a multi-predictor regression, when coefficients look unstable, implausibly large, or flip sign as you add variables.

Worked example

You fit a model predicting salary from age, years of experience and years since graduating — then notice the coefficient for age is huge and negative, which makes no sense. VIF tells you why: those three predictors carry nearly the same information, so the model cannot separate their effects.

Cautions

• Common thresholds (5, 10) are conventions, not tests. Judge them against how precisely you need each coefficient.
• Collinearity inflates standard errors; it does not bias the coefficients. If prediction is your aim, a high VIF may not matter at all.
• Dummy variables from one categorical predictor are collinear by construction — that is expected, not a problem to fix.

Alternatives

• Ridge / LASSO regression — the predictors are genuinely collinear and you want a model that copes rather than a diagnostic

How to read the output

What VIF measures

How much a coefficient's variance is inflated by its correlation with the other predictors. VIF = 1 means none; VIF = 9 means the standard error is three times what it would otherwise be. It is about the standard error, not about bias. Collinearity does not bias the coefficients — it makes them imprecise and unstable, so they can swing wildly when a single row changes.

The thresholds are conventions

Five and ten are the usual cutoffs, and neither has any theoretical basis. A VIF of 12 on a control variable you never intended to interpret is harmless. A VIF of 4 on the exposure you are there to estimate may not be. Judge by which coefficient is affected and whether its interval is still usable.

Dropping variables is not automatically the fix

Removing a collinear predictor changes what the remaining coefficients mean. If the dropped variable was a genuine confounder, the survivors are now biased — a well-estimated wrong number in place of an imprecise right one. Combining the variables, or accepting the imprecision and saying so, is often better.

How this calculator is validated · Which statistical test should I use?