Multicollinearity Test for Panel Data

[Nathalie Ruth]

Hi everyone.
I am new to Stata and currently using it for my research.

I am trying to run a multicollinearity test for my panel data.
Should I use to VIF command or the estat vce, corr, or correl command? What are the differences?

Thank you for your kind help.

[Joro Kolev]

There are no multicollinearity tests neither for panel, nor for any other data. The things you mention are diagnostics.

You can use such diagnostics in panel data if you think they are useful. The random and the fixed effects models are simply OLS on transformed variables, so all the diagnostics should work just as in standard OLS.

[Nathalie Ruth]

Hi, thank you for your response. I tried those 3 commands, and these are the results. What are the differences between these results? Thanks again.

. vif

Variable | VIF 1/VIF
-------------+----------------------
bm | 1.35 0.742879
size | 1.34 0.744481
roe | 1.24 0.805383
beta | 1.04 0.957648
turnover | 1.03 0.972894
-------------+----------------------
Mean VIF | 1.20

. estat vce, corr

Correlation matrix of coefficients of regress model

e(V) | turnover beta roe bm size _cons
-------------+------------------------------------------------------------
turnover | 1.0000
beta | -0.1231 1.0000
roe | 0.0397 0.0906 1.0000
bm | 0.0650 -0.0152 0.2741 1.0000
size | 0.1022 -0.1617 -0.2194 0.3554 1.0000
_cons | -0.1478 -0.0636 0.0637 -0.3382 -0.9549 1.0000

. correl return turnover beta roe bm size
(obs=462)

| return turnover beta roe bm size
-------------+------------------------------------------------------
return | 1.0000
turnover | 0.1016 1.0000
beta | 0.1611 0.1129 1.0000
roe | 0.1451 -0.0617 -0.0495 1.0000
bm | -0.1971 -0.0096 -0.0267 -0.3832 1.0000
size | 0.0685 -0.0830 0.1316 0.3458 -0.4396 1.0000


.

[Joro Kolev]

Like I told you the multicollinearity comes with the model that you are estimating.

It does not make sense to check whether the levels of variables are multicollinear, if you are going to apply first difference on the variables, or the within transformation, or quasi-demeaning (random effects model).

If you want to check for multicollinearity, choose what panel data model you want to use, and then check in this model.

[Nathalie Ruth]

Hi, if I'm using Fixed Effect Model, does that mean I need to run xtreg .., fe first and then run the estat vce, corr command?

Thanks!

[Carlo Lazzaro]

Nathalie:
yes, that's the sequence.
That said, when non-perfect (and then Stata chimes in by default omitting one of the culprits), multicollinearity is kinda of an oversold issue (see if interested chapter 23 of https://www.hup.harvard.edu/catalog....=9780674175440).

[Joro Kolev]

Hi, if I'm using Fixed Effect Model, does that mean I need to run xtreg .., fe first and then run the estat vce, corr command?

Thanks!

Yes, if any of the diagnostics you mentioned are available post -xtreg- apply then post xtreg.

If they are not available post xtreg, you can manually apply the within transformation -- there is a command for that -xtdata-.-- and then apply the standard diagnostics on the transformed variables.