[Sarah Magd]

Dear Carlo Lazzaro
Thanks a lot for your guidance. Could you please explain the difference between the obtained correlation matrix after using "estate vce, corr" and the correlation matrix after using "corr "

[Sarah Magd]

Dear Carlo Lazzaro
Could you please clarify the difference between the correlation matrix obtained from these two codes?:
- estat vce, corr
- corr

[Carlo Lazzaro]

Sarah:
-estat vce, corr- gives back coefficients correlation;
-correlate- (as per its -helpfile-) gives back correlation of variables.

[Clyde Schechter]

What would you recommend for us to test for the mulilcollinearity problem rather than vif?

In nearly all situations I recommend not testing for multicollinearity. The best exposition of this is in Arthur Goldberger's A Course in Econometrics. He devotes an entire chapter explaining why it's just a waste of time. I highly recommend reading that. Not only is it informative, and of real practical use, it is entertainingly written.

That said, if you are unpersuaded, or are working under the command of somebody else who hasn't gotten the message, bear in mind that -vif- depends only on the explanatory variables in the model. So, you can re-run the same model using -regress- instead of whatever other regression command you used, and then run -estat vif- after that.

If I were in charge at StataCorp, I would either get rid of -estat vif- altogether, since I believe it is useless, or, catering to popular demand from customers, I would make it available after any estimation command. I see no rationale for having it available after -regress- and nothing else.

[Carlo Lazzaro]

Sarah:
Clyde is obviously correct.
My previous reply was driven by some recent reviewers' commentsThose guys seem still obsessed with -vif- and multicollinearity no matter the Chapther 23 of . Arthur Goldberger's A Course in Econometrics.