Help with collinearity

#1
Hi,
I have a collinearity problem with my data. How do I deal with this? the variable which has collinearity is important, so I can't remove it from my equation.

Any help would be appreciate. Thanks
 

Dragan

Super Moderator
#2
Hi,
I have a collinearity problem with my data. How do I deal with this? the variable which has collinearity is important, so I can't remove it from my equation.

Any help would be appreciate. Thanks
Try centering your independent variables (so that they have means of zero) --- see if this helps.
 

jpkelley

TS Contributor
#4
In addition to Dragan's suggestion, one question we should ask if how you identified collinearity? VIF? A gut feeling? Even if collinearity exists, you should be careful before removing one of the variables. As I mentioned in a different post:

Which of the colinear variables to eliminate should be considered carefully. I would first look for ones that are positively correlated and that ultimately measure similar attributes of the system. Beware eliminating one variable in a negatively correlated pair of variables, unless you can show they're a proxy for the same feature of the system at large (i.e. both directly increase cortisol activity in the cell such as an increase in receptors and a decrease in binding globulins). Use your knowledge of all 30 of the independent variables in conjunction with variance inflation factors. Concerning variance inflation factors (VIFs), I would be aware of an unsettling trend in the literature towards acceptance of increased VIFs. That is, I've seen some papers suggesting that VIFs as high as 10 pose no issue at all. Ugh.