I'd say in this case you have a continuous variable of interest (GDP
per capita and its growth), so to me there is no point in categorizing
them that way: you impose some rather artificial criteria of the
subgroup selection, then lose some of the information, and further you
are trying to recover that lost information by introducing covariates.
That's weird. The fact that everybody is talking about ATEs and
excited about them does not mean that it should be applied to each and
every problem. If what you are analyzing is the effect of the IMF
macro stabilization package on the country growth, then yes, you can
envision a program evaluation estimator of some kind (treatreg,
propensity score matching, DID, whatever). You will likely have
problems with the sample size, as the matching estimators are
semiparametric in their nature: they may not have parametric (square
root of n) convergenece rates, and thus require huge sample sizes to
be of any substantial practical value. Also, I've seen papers showing
that the matching estimators with fixed number of matches are biased,
in pretty much the same way the non-parametric/kernel regression
estimators with fixed bandwidth are biased, and to correct for that
bias, one would have to go through something like bandwidth selection.
Stas
On Tue, 07 Dec 2004 16:25:40 +0000, Katarina Lynch
<[email protected]> wrote:
> Mark,
>
> Thanks! I think -treatreg- does something different, isn't it like
> "heckman" command.
>
> I want to first create a treated group and a comparable control group
> (countries that are in the list of poor countries v.s. non-poor countries),
> then select them to low/high GDP growth rate countries, and finally, see how
> such a selection into poor/low growth (poor/high growth) affects the
> relationship between growth and explanatory variables. I was instructed to
> read Heckman et al (1997) "Matching as an Evaluation Estimator" for the case
> of participation in job training and apply the same method to my case.
> That's why I thought -match- could be a command to start with. Isn't clogit
> something similar to match? Or would you still suggest -treatreg-?
>
> Thank you!
>
> Katharina
>
--
Stas Kolenikov
http://stas.kolenikov.name
*
* 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/
