Centered VIF or uncentered VIF?

[Kevin Traen]

Dear all,

I want to check for multicollinearity with 'estat vif' command, but the uncentered version gives high results for 2 variables (age and net assets). With the centered version, there is no problem. How should I interpret this and should I be worried?

Is this a sign that centering my independent variables could give better p-values?

Thanks in advance!

[Justin Niakamal]

estat vif, uncentered should be used for regression models fit without the constant term. So if you're not using the nocons option in your regression then you shouldn't even look at it.

[Kevin Traen]

Thank you!

[Jeff Wooldridge]

I'll go a step further: Why are you looking at the VIFs, anyway? I doubt that your standard errors are especially large, but, even if they are, they reflect all sources of uncertainty, including correlation among the explanatory variables. Until you've studied the regression results you shouldn't even think about multicollinearity diagnostics. Are the estimates too imprecise to be useful? If you're confidence intervals on key variables are acceptable then you stop there.

If none of your explanatory variables appear as squared or in interactions, centering your explanatory variables will only affect the intercept. Nothing else will be affected.

[Kevin Traen]

Dear Mr Woolridge,
As some of the variables are not significant, I was thinking on centering with the result that they could become significant.

[Jeff Wooldridge]

Right. That won’t help. Are the variables insignificant because the effects are small? Have you made sure to first discuss the practical size of the coefficients?

[ALKEBSEE RADWAN]

Right. That won’t help. Are the variables insignificant because the effects are small? Have you made sure to first discuss the practical size of the coefficients?

Hello Dear Jeff,
Actually I was looking for explanation for the VIF but Here I saw your answer and I would like to be more specific as follows:
one referee of my paper asked me to check the VIF and I did and got the highest value is 1.989. should I be worried about this results?
one more question:
Another paper of mine, I have regressed my model and I got a Coefficient of 16.23. My question is :
I have checked the VIF as well and the highest value of my independent variables is 2.14. So, what do you think the issue ?
Is that possible to have a coefficient over 1? if yes, How to translate such a coefficient to the economic magnitude " as i am doing research in accounting)?
thank you in advance