Ernest,
I think you were unclear and maybe wanting what can only be provided through
customization.
If my guess about your "interactions" is correct (you expanded X2,...,Xk to be
X2m,...,Xkm,X2f,...Xkf where f and m denote male and female and then ran the
regression on the pooled data), then you're not having any trouble estimating
what you want. The problem is you only care about the one constrained
parameter on X1. But you get two estimated coefficients (male and female)
for all the unconstrained independent variables which number in the hundreds.
Yea, that's some ugly output. Stata can't guess at which parameter you really
care about so it gives them all to you.
I'm not too clever with Stata programming, so I just write little ado files
specific to the problem. Since you can access most of the output of the
previous regression through the saved results in e(.), just write a little ado
to display those that you care about. After running the regression and
letting the output scroll by, run the ado and have it display the parameters
and test statistics you are interested in.
As far as using reg3 goes, I think what you really wanted was sureg (you'll
need to restructure your data as well I suspect). Then you could impose the
constraint. But you will still get the masses of output as long as you are
estimating that many parameters.
Tracy
>>> Ernest Berkhout 12/06/02 03:53AM >>>
Hello Statalisters,
two days ago I posted my problem to the Statalist but up to now there was
no response. Was it a stupid question, or was I unclear maybe? If the first
is true than please tell me why because i'm apparently quite ignorant,
otherwise i'll try to be more clear. What i want to do is
- a simple OLS regression of say Y on X1, X2...Xk for two different groups
(men & women). But I want to
- impose the restrictions that for say X1 the coefficients for both groups
are the same.
Of course one can estimate a single regression with interaction terms for
all X2..Xn but that way the output will be immense in my case where there
are >100 dummyvariables to be estimated (because there will be 3rd-order
interactions then as well). I wondered if there would be a more efficient
solution within Stata. I tried the <reg3> with <constraint>, but that
doesn't work because there are no overlapping records between the first
group (men) and the second group (women). The errorcode I get is 'not
enough observations'.
So the real problem is that when I use a single equation with nearly full
interaction I get a list with over a 100 interactionterms which is not
readable anymore. Will the above solution, estimating simultanously, be a
valid alternative or is maybe someone aware of an ado-program that will
restructure the output in a more manageable form?
Thanks in advance,
Ernest Berkhout
Stichting voor Economisch Onderzoek
Universiteit van Amsterdam
De SEO is verhuisd! Tegenwoordig ben ik bereikbaar op:
Kamer 3.08
Roetersstraat 29
1018 WB Amsterdam
tel.:+ 31 20 525 1630
fax:+ 31 20 525 1686
http://www.seo.nl
=================================
"To an economist, real life is a special case"
=================================
*
* 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/
*
* 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/
