What this answers

This calculator answers "how much does overlap between my predictors inflate each coefficient's standard error, compared to what it would be if the predictors were completely unrelated to each other?" High collinearity does not bias coefficients, but it does make them noisy and unstable.

How it is calculated

For each predictor, this calculator regresses that predictor on every other predictor in the set and records the R-squared from that auxiliary regression. The variance inflation factor is 1 divided by 1 minus that R-squared (equivalently, 1 divided by the tolerance). A VIF of 1 means no overlap at all; a commonly cited rule of thumb flags VIF values above 5 or 10 as concerning, though these thresholds are conventions, not universal statistical laws.

Worked example

For two predictors built from an orthogonal 2 by 2 factorial contrast, completely uncorrelated with each other by construction, both predictors get a VIF of exactly 1, the best possible result, confirming there is no overlap between them to inflate either coefficient's variance.

Assumption audit

Calculated from your data: each predictor's auxiliary R-squared, tolerance, and VIF.

Evidence to review: which specific predictors carry the highest VIF values; a high VIF on one predictor often points to a natural pairing or near-duplicate measurement worth reconsidering, combining, or dropping from the model.

You must verify: that a high VIF actually threatens your specific research question; if you only care about the model's overall predictive accuracy rather than interpreting individual coefficients, moderate collinearity is often tolerable.

What this result does not mean

A high VIF does not mean your model is wrong or your predictions are bad; it specifically means individual coefficient estimates are less precise than they would be with less overlapping predictors. VIF thresholds are conventions, not automatic deletion rules, no predictor should be dropped from a model based on VIF alone without also considering whether it is theoretically important.

Common mistakes

Dropping every predictor with a VIF above a fixed cutoff, without asking whether that predictor is theoretically important, is a common way collinearity screening does more harm than good; a well-justified predictor with moderate collinearity is often still worth keeping. A second mistake is checking VIF once and never again after adding new predictors to a model, since collinearity is a property of the whole predictor set together, not a fixed property of any one variable.

Limitations

This calculator requires at least 2 predictors and reports the standard VIF, which assumes a linear relationship among predictors; it does not detect more complex, nonlinear forms of redundancy between predictors.