Variation Inflation Factor

[Jessica Berrett]

What should the variation inflation factor (VIF) be no higher than? And does anyone have a good reference for this? I'm reading up on this, but I'm seeing a wide range of recommendations. Thank you!

[Ken Chui]

It is called variance inflation factor, not variation.

10 has been the most commonly cited number (see http://www.how2stats.net/2011/09/var...actor-vif.html for some possible starting references). Those "no higher than" rules is just approximation, the conceptual relationship between the variables should also be considered as well. In addition, because each independent variable has their own VIF, it is also reasonable to focus more on the main predictor's VIF.