Buzz Burhans <[email protected]> asks:
> I'm curious about the capacity for hierarchical models in the new version;
> xtreg for instance allows random intercepts, but not random slopes; when I
> need mixed hierarchical models with both random intercepts and slopes I've
> used and appreciate GLLAMM in Stata, but it runs quite slowly in Stata 7,
> even under adaptive quadrature. I have had to turn to alternative programs
> ( S-Plus primarily) for running such models for primarily two reasons; they
> run faster, enough faster that it makes a difference between working with an
> extensive project feasible versus infeasible given the time constraints
> imposed by trying to run them in Stata, and secondly I also appreciate the
> ease of using predefined alternative variance and correlation structures in
> S-Plus, as well as the graphical approach facilitated by S-Plus. Stata 8's
> new graphics look really exciting, and will undoubtedly give Stata an edge
> in graphical evaluation and application to such models, but my other two
> concerns seem not to be addressed in the new Stata.
> I am a great fan of Dr. Rabe-Heskeths work ( the author of GLLAMM, who also
> has a great book "A handbook of Statistical Analysis using Stata"), and was
> hoping that perhaps GLLAMM would be hard wired in to the next Stata upgrade,
> but I don't see it on the list of new statistical elements in Stata 8. I
> seem to remember that her web site at one point suggested that this would
> occur (hard wiring GLLAMM into Stata which would make it faster?)
Some hard-wiring of the GLLAMM code has already been done under Stata 7,
resulting in a modest gain in speed. As far as Stata 8 goes, we plan to
internalize yet another GLLAMM subroutine into the Stata 8 executable, which
hopefully will result in a more-than-modest-gain in speed. How much, of
course, depends on the problem at hand, and on factoring in the implicit
speeding-up of ado code under Stata 8. Although the Stata 8 executable which
ships will not contain this new internalized routine, we plan on making it
available via an update to take place quite soon after shipping.
> I really like working in Stata, and will be very interested to see how much
> faster GLLAMM runs in Stata 8. Are there plans to incorporate GLLAMM into
> Stata rather than run it as an ado? Will this improve the speed? Or are
> there other plans to enhance the capacity of Stata to run hierarchical
> models? Requests for or inquiries about such capacities seem to come up
> regularly on the list.
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. Whether this
involves a complete adoption of Dr. Rabe-Hesketh's work, independently-written
code, or some combination of the two, I cannot say at this time.
--Bobby
[email protected]
*
* 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/
