st: sub samples

[Kit Baum <baum@bc.edu>]

Jean said

"Thanks Scott - though using an interaction term as your code (I  
think) implies forces the coefficients on all covariates to be the  
same across samples. I actually would want not to do that.


----Message d'origine----
>De: "Scott Merryman" <smerryman@kc.rr.com>
>A: <statalist@hsphsun2.harvard.edu>
>Sujet: st: RE: Sub samples
>Date: Mon, 27 Feb 2006 09:21:13 -0600
>
>-findit chow- points to several FAQ by William Gould that will be of
>interest.
>
>Here is an example testing if the estimated coefficients for the  
postwar
>years differ.
>
>webuse grunfeld, clear
>
>gen byte postwar_years = year >1945
>
>foreach var of varlist mvalue kstock  {
>  gen post_`var' = postwar_years*`var'
>  }
>
>xtreg invest mvalue kstock post* ,fe
>testparm post*"


Obviously not. The whole point of allowing for interactions between  
all regressors (including the constant) and the regime dummy is to  
allow the coefficients on all variables to differ after 1945, as is  
evident if you run the example Scott has provided. The only  
restriction you are imposing in this pooled context is that you are  
assuming that there is homoskedasticity across the subsamples. You  
can deal with this by modifying the estimation command to

xtreg invest mvalue kstock post* ,fe robust

which should also 'robustify' the testparm F-statistic, if I'm not  
mistaken.


Kit Baum, Boston College Economics
http://ideas.repec.org/e/pba1.html


*
*   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/