Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: -gllamm- vs -meglm-
From
"Joseph Coveney" <[email protected]>
To
<[email protected]>
Subject
Re: st: -gllamm- vs -meglm-
Date
Wed, 3 Jul 2013 12:34:37 +0900
Daniel Waxman wrote:
containing such large groups..."
3,000 observations in a group isn't what it used to be! There are
many of us who routinely work with gigabytes of data, and it would be
helpful if Stata's documentation made it clear from the outset that
some estimation routines are not meant for us, if they are not.
Somehow SAS's glimmix routine is able to fit at least single random
intercept logistic models on huge datasets relatively quickly (and it
does so while using very little memory). To the extent that the
Stata development team is willing to comment on a competing product,
I'd be interested on their take on whether Stata will be able to do
the same any time soon.
--------------------------------------------------------------------------------
Just for clarification, is PROC GLIMMIX fast and light on gigabyte-sized
datasets even when it's using seven-abscissa adaptive Gauss-Hermite quadrature
as its estimation method? According to its documentation, "The default
estimation technique in generalized linear mixed models is residual
pseudo-likelihood with a subject-specific expansion (METHOD=RSPL)."*
StataCorp can speak for themselves of course, but I'd expect that based upon
this user experience the problematic intermediate computations will end up
getting recast in quad precision. Or something. Regardless, whether Stata will
be able to do the same anytime soon would depend upon the usual factors
affecting software development and maintenance.
In the meantime, I wouldn't be too harsh: StataCorp responded quickly with an
already built-in fix when encountering a glitch in a complicated novel feature
(actually, a suite of new features) after it ran up against an unanticipated use
case within the first two weeks out of the box. Perhaps they ought to have
anticipated the use case, I don't know. Still, that's not too shabby.
Joseph Coveney
*
http://support.sas.com/documentation/cdl/en/statug/65328/HTML/default/viewer.htm
#statug_glimmix_gettingstarted01.htm
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/