Antwort: st: xtreg postestimation and regression using residuals or dummies<Watchdog: Virus checked>

 From Justina Fischer To statalist@hsphsun2.harvard.edu Subject Antwort: st: xtreg postestimation and regression using residuals or dummies Date Sun, 30 Sep 2007 21:11:09 +0200

HI

I think you should not do this two-step approach by hand as the standard errors have to be corrected in the second stage, among other things.

But there is a user written ado-file by Troeger and Pluemper which seems to do exactly what you want.

it is called 'xtfevd'
(try findit)

Justina

Dr. Justina A.V. Fischer, M.A.
University of St. Gallen
CH-9000 St. Gallen

-----owner-statalist@hsphsun2.harvard.edu schrieb: -----

An: statalist@hsphsun2.harvard.edu
Von: "alessia matano" <alexis.rtd@gmail.com>
Gesendet von: owner-statalist@hsphsun2.harvard.edu
Datum: 30.09.2007 07:53PM
Thema: st: xtreg postestimation and regression using residuals or dummies <Watchdog: Virus checked>

Hi to everybody,

I have some questions for you about two issues:
1. xtreg , fe postestimation: which are the residuals the option
predict, e gives after xtreg, fe?? are the centered or not residuals?
I think they are the residuals in absolute levels..but I am not sure

2. the most important question is the following:

I have the following model:

Yrt=a+bXrt+cQrt+eit     (1)

where both Yrt and Xrt comes from the aggregation at r level (r say
region) of individual varibales (y wages and X average age, percentage
of female, etc.)
Qrs are a set of  aggregate variables by definition (say regional
exports) whose I want to study the effect on Y. I directly estimate
with a fe model this model and it works. Afterwards some people
suggest me to perform in two steps this estimation, in three possible
ways, in the way to account for individual observed and unobserved
heterogeneity (instead of relying on some average of those, such ase
average age...) :

A. regressing Yit=a+bXit+eit with a fe model and then take the
residuals (added to the constant) the perform the regression on Xrt
(like averaging the residuals over rt). In this way it does not work.
Anyone of you does know why?

B. regressing Yi=a+bXi+Dr+ei for any year with (I suppose) a simple
ols model and then keep the estimates for the dummies and put them
into the regression Drt=cXrt+urt.

C. regressing Yit=a+bXit+Drt+eit with a fe model and then keep the
estimates for the dummies and put them into the regression
Drt=cXrt+urt. In the last way (I think it is the better) you get the
estimates cleaned by observed and unobserved individual heterogeneity.
However I know you have to do some manipulations to the estimates in
order to get reliable estimates (the dummy are in comparison with a
base value), but i do not know exactly which. Can anyone help me?

Which of those do you think is more correct, for such an estimation i
want to do? It is a robustness check of the estimation (1).

Many many thanks for your suggestions
alessia
*
*   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/