Mark,
Thanks for the fast reply. As you point out my goal is to decide whether the
fixed or random effects model is appropriate.
As the model indicates that both firm and time effects are present I use
-xtreg, cluster()- (i.e. -cluster(id)- for the firm effect) with addtitional
time dummies. In this context I encountered the discussion about the
appropriateness of the Hausman test.
That said, just to rephrase: implementing an artifical regression as below
gives me a more robust Hausman test that works in the described setting?
I was just wondering if I get too complicated or overlook something, because
an official Stata command seems to be missing (other than -xttest0- for
random effects) and I assume that the setting described is not so uncommon?
- Tom
-----Urspr�ngliche Nachricht-----
Von: [email protected]
[mailto:[email protected]] Im Auftrag von Schaffer, Mark
E
Gesendet: Donnerstag, 18. Januar 2007 17:41
An: [email protected]
Betreff: st: RE: Panel re vs. fe model
Thomas,
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of
> Thomas Erdmann
> Sent: Thursday, January 18, 2007 3:21 PM
> To: [email protected]
> Subject: st: Panel re vs. fe model
>
> Hi -
>
> The drawbacks of Hausman test regarding its asymptotic
> assumptions or in a cluster setting are discussed at several
> points in the Statalist.
>
> e.g. http://www.stata.com/statalist/archive/2005-08/msg00741.html
> or several others by e.g. searching for "fe re cluster
> Hausman" in the archives.
>
> My conclusion would be that using -suest- would be a good
> alternative but is not possible because -xtreg, fe- does not
> produce scores. I found another alternative described here,
> http://www.stata.com/statalist/archive/2005-08/msg00853.html
> where Vince Wiggins suggests using an artificial regression
> (find below).
>
> Two questions:
>
> (1) Regarding the code below, could you just add the
> -cluster()- option as further option and would the artificial
> regression still make sense?
Yes.
> (2) Are there yet other alternatives to the Hausman test in
> this setting?
The difference between the FE and RE specification is that the latter
uses more orthogonality conditions, and the Hausman test in this context
is just a GMM test of these extra orthogonality conditions. If you want
to test something else, then I suppose you would use an alternative
test, but I'm not sure what the something else would be.
Cheers,
Mark
> Thanks for your feedback,
> Tom
>
>
> ---------------------------------- BEGIN --- artreg.do ---
> CUT HERE ------- local id idcode local depvar ln_wage local
> varlist ttl_exp union white
>
> iis `id'
> xtreg `depvar' `varlist' , re
>
> local i 0
> foreach var of varlist `varlist' {
> egen mymean`++i' = mean(`var') if e(sample), by(`id') }
>
> xtreg `depvar' `varlist' mymean1-mymean`i' , re
>
> testparm mymean1-mymean`i'
> ---------------------------------- END --- artreg.do ---
> CUT HERE -------
>
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
