Your VIF is too high. What do you do about multicollinearity?

Two or more predictors in your regression carry much the same information, so the model cannot tell their contributions apart.

Reading the number

VIF measures how much a coefficient's variance is inflated by correlation with other predictors. VIF of 1 means no correlation.

Common thresholds are 5 and 10. Neither is a law. They are conventions, and which you use should depend on what you are doing.

When it actually matters

It matters if you are interpreting individual coefficients. Multicollinearity makes them unstable: large standard errors, coefficients that flip sign when you add a variable, and predictors that are jointly significant while none is individually significant.

It matters much less if you are only predicting. Overall model fit and predictions are largely unaffected. If you only need accurate predictions, a high VIF may not be a problem worth solving.

The fixes

Drop one of the correlated predictors. Simplest and usually best. Keep the one that is easier to interpret or measure. You lose very little, because it was duplicating information.

Combine them. Height and weight becoming BMI, or several related items becoming a mean score. Often more interpretable than either alone.

Centre your predictors before creating interaction terms. Interactions are correlated with their components almost by construction, and centring removes most of that. This is a common and easily fixed cause of alarming VIFs.

Use PCA to reduce correlated predictors to uncorrelated components, at the cost of interpretability.

Collect more data, if that is available to you. Multicollinearity is partly a sample-size problem.

What not to do

Do not drop predictors purely on VIF while ignoring your research question. If a variable is central to your hypothesis, keep it and report the instability honestly.

In Plotara

VIF sits in Diagnostics alongside residual diagnostics, and runs against the same model you fitted in Linear Models.