I have a linear regression model that has no multicolinearity problem with low VIF scores. However, when I include the interaction term, this interaction term and its components get very high VIF scores. Can I ignore the multicolinearity problem and high VIF scores of the interaction term in this case?


The short answer is yes. Interaction terms tend to be collinear with the original variables involved. That is why post-hoc interaction tests are often underpowered.

Interaction that is unaccounted for renders the estimate wrong, while inflated variance inflates p-value. If the interaction terms are already statistically significant, inflation of variance is no longer a problem.

| cite | improve this answer | |
  • $\begingroup$ Thank you very much for your reply. Then reversely, if the interaction terms are not statistically significant, the VIF is a problem? $\endgroup$ – Eric Apr 18 '17 at 17:55
  • $\begingroup$ If the interaction terms are not significant then the reduced model with the interaction term is usually preferred. For that reason VIF became irrelevant. $\endgroup$ – Penguin_Knight Apr 19 '17 at 17:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.