Re: st: RE: regression coefficient by group

[Maarten Buis <maartenlbuis@gmail.com>]

On Thu, Dec 8, 2011 at 6:18 AM, Rui Zeng wrote:
> Nick's suggestion of statsby works really well. However, I still have an issue: the speed.  I have to run 4000 regressions/month over a period of 300 months, with statsby I think that will takes me about 10+days to get the result. That is something that should never happen if I use STATA, so dear statalisters, any good suggestions?

That has nothing to do with Stata but just with the number of
regression you want to run: 1.2 milion. Stata is quick, but even
-regress- takes a little time. 1.2 milion times a little is still a
lot. Say your regression takes 1 second than just doing the
regression, and ignoring the overhead due to saving the results, will
take 1.2*10^6/(60*60*24) = 13.9 days. I suspect that for a lot of
models -regress- will actually be quicker than 1 second, but saving
results will also takes a little time, so I think that your estimate
of approximately 10 days is about right and as good as can be
expected.

One option would be to split this task over multiple computers. Say
there is a computerlab in your institution, that is closed during the
weekend. Than, 30 computers doing each 2 months should be finished in
one weekend.

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany


http://www.maartenbuis.nl
--------------------------

*
*   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/