How to test for assumptions when FE regression model is used for panel data?

[Christina Adnett]

Hi there,

I am new to Stata and have run a FE regression on my data using: xtreg dep indep, fe
I now wish to run tests: VIF for multicollinearity, hettest for heteroscedasticity, Jarque-Bera for normality and Durbin-Watson for Auto-correlation.
However, from what I have read, these are only possible after the command function regress. Is there any way for me to test the panel data without reverting to an OLS estimation?
Thanks

[Carlo Lazzaro]

Christina:
welcome to the list.
The first thing I would do i investigating whether -xtreg, fe- outperforms pooled OLS.
If the F-test appearing at the foot of the outcome table of -xtreg, fe- lacks of statistical significance, pooled OLs is the way to go.
Assuming that your result is opposite, you can test for heteroskedasticity and/or autocorrelation by comparing -xtreg, fe- output with and without robustified/clustered standard errors.

As an aside, you post highlight the importance of posting what one typed and what Stata gave back (as reminded by the FAQ) and/or posting an excerpt/example of your dataset (anonymized, if covered by undisclosure agreement) via the user-written programme -dataex- (type -search dataex- from within Stata to install it). Thanks.

[Phil Bromiley]

To add to Carlo's answer, if the tests Carlo suggests demonstrate you need fixed effects and you really want vif's, you can simply put in dummies for each panel (i.panel) in an OLS. OLS with i.panel is the same as xtreg,fe if you make the same error structure assumptions.