Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.

# RE: st: hausman test xtivreg(re) vs ivreg

 From "Schaffer, Mark E" To Subject RE: st: hausman test xtivreg(re) vs ivreg Date Fri, 12 Mar 2010 17:37:12 -0000

```Sergio,

I agree with Kit - I don't understand the point of what you're
suggesting.

Let me put it this way.  What if you had no endogenous regressors?  Then
you'd be proposing a comparison of pooled OLS vs. random effects, where
one of your explanatory variables is time-invariant and is identified
only by the cross-sectional variation.

Pooled OLS and the RE estimator both make use of the between and the
within variation, but the RE estimator does so in an efficient way.

Why would you want to make such a comparison?  What would it tell you?

--Mark

> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Kit Baum
> Sent: Friday, March 12, 2010 3:18 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: re: st: hausman test xtivreg(re) vs ivreg
>
> <>
> Sergio said
>
> > I would like to know if it is possible to do a test in
> Stata on xtivreg
> > (with random effects) vs. the same specification using ivreg without
> > random effects.
> > Please note that this is not a FE vs RE comparison. I can't
> run FE on my
> > specification because my endogenous variable does not vary within xt
> > units.
> > I have done something like
> >
> > ivreg y x2 x3 (x1= z1 z2)
> > estimates store ivreg
> > xtset w
> > xtivreg y x2  x3 (x1= z1 z2), re
> > estimates store ivreg_re
> > hausman ivreg xtivreg_re, sigmamore
>
> I have not tried this out, but it is surely improper for the
> hausman command as written. The first estimator named in the hausman
>  command is name-consistent, which is supposed to be
> consistent in all states of the world. The IV estimator
> applied to panel data is not consistent if there is
> unobserved heterogeneity;  xtivreg, fe would be consistent in
> that circumstance. Nor would the IV estimator be efficient in
> a panel context unless the weight applied in the
> quasi-differencing to the unit was precisely zero. So the
> standard IV estimator cannot be either of the estimators
> listed in the hausman command.
>
> Why don't you want to contrast xtivreg,fe (which is always
> consistent under the maintained hypothesis of correct
> specification) with xtivreg, re (which is relatively
> efficient if the random component is orthgonal to the
> regressors)? That is what can be readily done with -hausman-
> in a panel context.
>
> If your 'endogenous variable does not vary within xt units',
> it smells like a fixed effect to me!
>
>
> 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/
>

--
Heriot-Watt University is a Scottish charity
registered under charity number SC000278.

*
*   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/
```