Re: Re: st: Regression with about 5000 (dummy) variables

[Christopher Baum <kit.baum@bc.edu>]

<>
On Apr 20, 2012, at 2:33 AM, John wrote:

> The estimator still seems good. Notice, though, that the F-test 
> numerator DFs are only 3.  So that's what I meant when I said we save on 
> DF (as compared to the OLS fixed-effects estimator).

You're cheating, John, by overlooking that theorem about the absence of a free lunch. 
The number of DF is not 3. When you gave the commands

bys idcode : egen double cl_age_id = mean(age)
bys south : egen double cl_age_south = mean(age)

you computed a large number of sample means, so the cluster-mean regressors in the Mundlak 
model must be considered as not single DF, but as many DF as it took to compute them,
following the logic of the FE model. 

I could make the same mistake by applying the within transformation to Y and X and 
running OLS, which would have one DF for the one slope estimated. But if I used the
FE estimator it would properly account for the fact that the within transformation is not
free, as this is just the LSDV (dummy-variable) model, and putting in all those dummies is not free. 
Neither is this method. Your DF should be adjusted to reflect the creation of those
cl_* variables.

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/