Ian Dohoo wrote in response to my wishes and grumbles for enhanced capabilities for
mixed models in Stata:
--------------------------------------------------------------------------------
I am in complete agreement with the need for an equivalent
to SAS Proc Mixed (for my work - currently the only serious
limitation in Stata)
However, I believe that -gllamm- (written by Sophia Rabe-
Hesketh) is not just an alternative to NLMIXED, but is in fact
more flexible (ie gllamm can handle n-level data while
NLMIXED is limited to 2-level data).
--------------------------------------------------------------------------------
Couple of things:
First, I feel that I was gratuitously harsh in my earlier posting. When the comment was
raised in the wishes and grumbles session, Bill Gould was in complete sympathy. In
addition, at the NASUG meeting, an undocumented capability in Stata 8 was disclosed
that blazes the path for Stata for mixed models, for which at least some subroutines
need to be in compiled code (C, Fortran), and not as interpreted ado-file code, in order
to execute in a reasonable time and in reasonable precision. (This is the tactic that S-
Plus uses in its lme and nlme.) In addition, last December 23rd, Bobby Gutierrez (from
StataCorp) wrote in response to a query, "Hierachical models are at a high priority for
us at Stata Corp., and we will continue to work towards making them part of official
Stata." So we know that this project is well underway. (I recall that there was a more
recent posting from someone at StataCorp to the effect that they were targeting this
spring for a milestone on the project, but I can't locate the posting in the archives now.)
Second, I agree with Ian about -gllamm-'s superiority over PROC NLMIXED as to
flexibility in the number of levels. -gllamm- is a gem of a command. But one feature
that -gllamm- lacks (at least to those like me who don't know how to hack its code to get
it to do it) is the ability to fit arbitrary nonlinear functions so long as you can specify the
function � la -nl-. With PROC NLMIXED, you specify the nonlinear function and its
fixed and (normally distributed) random effects, and it will try to fit the model. This
enables, for example, multi-exponential pharmacokinetic models to be fit. In my
posting, I was going to mention something to the effect that, if -gllamm- could be
opened up so that users could specify an arbitrary nonlinear model analogously to
PROC MIXED, then it would be leagues ahead of what SAS or S-Plus has to offer.
Joseph Coveney
*
* 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/
