VIF analysis in Panel data

[eljayski]

You can actually test for multicollinearity based on VIF on panel data. lets say the name of your equation is eq01, so type "eq01.varinf" and then click enter. then you will get centered (with constant) vif and uncentered (without constant) vif. Keep in mind, if your equation dont have constant, then you will only get the uncentered.

this is kool info for me! but, what is the syntax? that is, where, or what screen, within eviews do i type in the name of my model, followed by .varinf?

help gratefully appreciated! eljayski

[EconometricAarhus]

You can actually test for multicollinearity based on VIF on panel data. lets say the name of your equation is eq01, so type "eq01.varinf" and then click enter. then you will get centered (with constant) vif and uncentered (without constant) vif. Keep in mind, if your equation dont have constant, then you will only get the uncentered.

When having a constant in a regression does it then only make sense to look at the centered? It seems like there is a huge difference. I have correlation up to 70%. Does it make most sense to ommit some variables or just be in mind of possible multicolinarity?

Best regards.

[Max]

You can actually test for multicollinearity based on VIF on panel data. lets say the name of your equation is eq01, so type "eq01.varinf" and then click enter. then you will get centered (with constant) vif and uncentered (without constant) vif. Keep in mind, if your equation dont have constant, then you will only get the uncentered.

Does anyone know if this is appropriate to use that method for panel data? I am confused because in 2008 Eviews Gareth said

ok, VIFs shouldn't be available for panel equations

I was wondering if the output I have when using eq01.varinf makes sense now under Eviews 8 ?

Thanks

[Lalan_dk]

The reason i test for multicollinearity is to check whether there is correlation between my variables.. is there any test that i can use for testing multicollinearity in panel data? or, we are assuming no multicollinearity in panel data?

Why do you care whether there is correlation between your variables? There almost certainly is correlation, but that's what OLS is built to sort out.

Hey startz, can you give references about ignoring multicollinearity?

[startz]

The reason i test for multicollinearity is to check whether there is correlation between my variables.. is there any test that i can use for testing multicollinearity in panel data? or, we are assuming no multicollinearity in panel data?

Why do you care whether there is correlation between your variables? There almost certainly is correlation, but that's what OLS is built to sort out.

Hey startz, can you give references about ignoring multicollinearity?

Wooldridge, section 3.4 writes

Since multicollinearity violates none of of our assumptions, the "problem" of multicollinearity is not really well defined.

The best reference is Goldberger (1991), but I'm afraid my copy is under lockdown in my office.