Fardad,
1) sounds as if you thought Stata should give you some kind of secret test
that let`s you off the hook if -hausman- tells you not to do something. -h
hausman- certainly mentions -suest- as an alternative, but it is probably
not designed to let you circumvent evidence not to apply RE. -hausman-
checks for the differences between an always consistent estimator (FE in
this case) and an efficient one under the null. If the difference is outside
of bounds established by the underlying theory, it will let you know and you
should heed the advice...
2) I understand from your reply to Nils that you think heteroskedasticity
would invalidate your point estimates, which I do not believe to be true. It
does impact your standard errors, though, which can lead to mistakes in
inference.
4) -xthtaylor- allows correlation between parts of your covariates and the
individual effects ("u_i") while the RE estimator requires that neither u_i
nor the e_it are correlated with the covariates. The advantage is that you
do not have to find external instruments because the command constructs them
from within. Check the entry in [XT] for the details.
5) -xtivreg- is designed to remedy correlation btw covariates and the e_it,
while -xthtaylor- deals with correlation btw regressors and the u_i...
HTH
Martin
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Fardad Zand
Sent: Tuesday, October 14, 2008 10:44 AM
To: [email protected]
Subject: Re: st: RE: 3 Problems in Panel Data Analysis
Hi Martin,
Thank you so much for your through answers. Appreciate for your time.
In order to clarify your answers, I just have a few questions if you
or other people in Statalist can respond:
1- concerning point 1: So, how can I justify using RE (or BE) when the
Hausman test is in favor of FE? Is there any test that takes into
account the nature and structure of the data in addition to difference
between the estimators? How can one check for correlation between
explanatory variables and the individual effects u_i, so I can see how
biased my estimation would be if use RE?
2- concerning point 2: robust options to my knowledge is a correction
for heteroskedasticity problems. But, as you refer in your answer,
does it have anything to do with within and between variances and
their difference?
3- concerning point 2: More importantly, the robust option is
available for only xtreg (to my knowledge). But, how about xttobit? Is
there any robust options available?
4- concerning point 3: you refer to xthtaylor. I didn't know about
this command (and indeed the method). how would you compare xthtaylor
and xtreg RE? What are the conditions for xthtalor to be unbiased?
with respect to the nature of my data and the results of the Hausman
Test, can xthtaylor be a good alternative? What are the limitations?
5- concerning point 3: how would you compare xtivreg and xthtaylor in
order to solve the issue of simultaneity? Are they both overcoming the
problem?
Thank you so much for your time and patience for answering and guiding me.
Good luck everyone,
Fardad
On Tue, Oct 7, 2008 at 2:54 PM, Martin Weiss <[email protected]> wrote:
> Concerning 1): FE, because of the way it is constructed, cannot provide
> estimates for time-invariant covariates, so that might be the reason for
> dropped regressors.
>
> Concerning 2), -xtreg- has a -clustered robust- VCE which seems
appropriate
> given your observation of stark differences between between groups vs.
> within group variance. Note that you can check for these two variance
> components with -xtsum-.
>
> Concerning 3), the lack of instruments can sometimes be overcome with
> -xthtaylor- which can remedy correlation between covariates and the
> individual effect u_i and at the same time provide estimates for
> time-invariant regressors.
>
>
> HTH
> Martin
*
* 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/
