Multicoliniarity

[David Moore-10]

Dear Listers,

I am hoping that someone might be able to help with a problem I am having.
I am running a few regression models at the moment and an wondering about
multicoliniarty, I believe from what I've read that ideally we need a VIF
<10? this is ok for the first stage of our model however at a second stage
we have entered some interaction terms (created from the first stage of the
model) understandably these correlate highly with some of the other
variables (as they are products of these) resulting in VIF's as high as 40
what does this mean for the analysis, can we continue or is there something
I am missing.

Best
David

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Bruce Weaver]

That kind of multicollinearity will largely go away if you center your variables on some plausible in-range value before computing the product terms.  

Many people center on the mean in knee-jerk fashion.  But you do not have to center on the mean--the value you use is arbitrary.  The mean changes from sample to sample, so if you mean-center, you are centering on a different value every time.  If instead you pick a nice round value near the mean, and use it in every analysis, you have much better comparability across data sets.  (Another option I like is centering on a value near the minimum so that the intercept gives the fitted value for someone near the minimum on that variable.)

David Moore-10 wrote

Dear Listers,

I am hoping that someone might be able to help with a problem I am having.
I am running a few regression models at the moment and an wondering about
multicoliniarty, I believe from what I've read that ideally we need a VIF
<10? this is ok for the first stage of our model however at a second stage
we have entered some interaction terms (created from the first stage of the
model) understandably these correlate highly with some of the other
variables (as they are products of these) resulting in VIF's as high as 40
what does this mean for the analysis, can we continue or is there something
I am missing.

Best
David

=====================
To manage your subscription to SPSSX-L, send a message to
[e-mail] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD