High VIF Values with Interaction Term in Probit Regression

[Serena Menny]

Dear members,

In my probit regression analysis, the average VIF was 1.15 (highest individual vif= 1.43) without interaction terms. After including an interaction term, the average VIF rose to 11.11, with the interaction term VIF at 63 and the next highest values at 55.

Is such an increase in VIF normal when adding interaction terms, or does it indicate a problem with multicollinearity in my model?

Thank you,

[Jeff Wooldridge]

You should be computing something like the average marginal effects and they won’t be affected much by this kind of collinearity. But what I would do is center the two variables about their means before interacting them. The AMEs won’t be affected but the main effects coefs will be more meaningful and precisely estimated.

Essentially the same issue arises in linear models.

[Serena Menny]

You should be computing something like the average marginal effects and they won’t be affected much by this kind of collinearity. But what I would do is center the two variables about their means before interacting them. The AMEs won’t be affected but the main effects coefs will be more meaningful and precisely estimated.

Essentially the same issue arises in linear models.

Thank you so much for your answer!