estat vce, correlation

[Colin McAweeney]

Hey everybody,
After running xtreg, fe vce(robust) I have examined the correlation between the independent variables using the estat vce, correlation command.

My question, in economics and other social sciences what is an acceptable level to stay below concerning the correlation between independent variables? I have heard .4, but I wanted to get a second opinion.

Much appreciated,
Colin

[Carlo Lazzaro]

Colin:
welcome to the list.
If you're concerned about quasi-extreme multicollinearity, you may want to take a look at http://statisticalhorizons.com/multicollinearity.
However, as far as I know, there's no an unquestionable threshold-value that you can invoke to start worrying about that issue..

[Colin McAweeney]

Hi Carlo,
Thanks for the reply. Although VIF is used post reg command and not xtreg, fe, would you consider low VIF values to be justification for limited collinearity between independent variables?

[Carlo Lazzaro]

Colin:
I would also take a look at the mean VIF and consider the risk of the so called "tipping effect" (see http://statisticalhorizons.com/multicollinearity).
As it often happens in statistics, it's your educated call (supported by the literature in your research field) to consider a given VIF value to be negligible for your analysis.

[Colin McAweeney]

Thank you very much for the advice Carlo! It is much appreciated.


Colin