Re: st: One question about XTOVERID

[Tingting Tong <pumpkinting@gmail.com>]

Dear Kit,

Thank you for your quick reply.

After doing the FE and RE, I need to do the PMG (Pool mean group) and
FMOLS (Fully modified OLS) since I have a large T and large N panel.

However, both PMG and FMOLS is based on Fixed effect model.

So I wish my test result would be fixed effect model. I cannot perform
Hausman test and robust hausman test due to my data problem. (I have
time variant and individual invariant variables.) My XTOVERID result
indicates a RE model.

Furthermore, the intercept, coefficients and S.D. of my FE and RE
model results are quite similar. I think this furture proves that RE
is better.

With all evidence, I am not sure about should I insist that FE is
better in my case?
Could you please give me some guidance or reference?
Thank you very much.

Tingting Tong



On Wed, Dec 28, 2011 at 12:01 PM, Christopher Baum <kit.baum@bc.edu> wrote:
> <>
>
>
> I saw a lot of introduction about XTOVERID on internet. But i still
> have one simple question.
>
> When using XTOVERID determine FE and RE, can anybody tell me what is
> the null hypothesis?
>
> For instance, in my case, when running XTOVERID, my p-value is bigger than 0.1.
>
> Does it mean FE model is better?
>
>
>
> As the help file for Schaffer & Stillman's -xtoverid- (from SSC) explains, the "Hausman test" for FE vs RE can
> also be cast as a test of the additional overidentifying restrictions that RE imposes. FE (xtreg, fe) is consistent
> iff X \perp \epsilon, where X contains regressors and \epsilon is the idiosyncratic error. FE does not require
> that X \perp u, where u is the fixed effect. RE does require that X \perp u, so there are additional overidentifying
> restrictions associated with RE. The null of either the Hausman form of the test or of the test performed by
> xtoverid is that RE is consistent. A large value of the test statistic (or a small p-value of the statistic) is a rejection
> of that null, saying that RE is inconsistent. If you are getting a p-value of 0.15 or 0.20, then the evidence
> against RE is not that strong, and you can get away with the RE assumptions on the error process.
>
> Kit
>
>
> Kit Baum   |   Boston College Economics & DIW Berlin   |   http://ideas.repec.org/e/pba1.html
>                             An Introduction to Stata Programming  |   http://www.stata-press.com/books/isp.html
>  An Introduction to Modern Econometrics Using Stata  |   http://www.stata-press.com/books/imeus.html
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/