st: functional improvements in Stata8 and gllam

[Buzz Burhans <wsb2@cornell.edu>]

Let me preference my comments on gllamm by describing myself as a consumer 
and modestly knowledgeable user of statistical software, but by no means a 
statistician, although at various times I serve that function. I should 
also add that I am a fan of both Stata and Sylvia Rabe-Hesketh's work

That said, I can't at the minute give you exact speed differences because I 
am in the middle of another project, and it would require firing up both 
Stata 7 (which I kept operational when I upgraded) and stata8. However, 
running gllamm with continuos outcome variables and two levels is now 
feasible both because of the use of adaptive quadrature as opposed to 
quadrature, and because Stata Corp has reportedly hardwired in some of the 
routines in Stata 8.  The difference for me in my datasets ( 500 to 1000 
observations, tiny by many stata users standards) is in the order of 
dropping the time to run from approximately an hour to a few 
minutes.  Running dichotomous outcomes with gllamm is almost identically as 
fast as xtlogit, at least in the applications where I have used it recently.

My desire for more development in the area of hierarchical models  is due 
to my perceived need for additional capability to incorporate into the 
model the covariance structure, which, as I understand it, can not be done 
in gllamm at this time, other than with a somewhat limited ability to 
specify nocorrel as an option.   I should also add that the gllapred 
package has been improved recently in it's handling of residuals, and that 
has also been useful for me.

Buzz Burhans



At 10:28 AM 2/26/03 -0600, you wrote:

> Buzz - Thank you for your comments. I admit I am very susceptible to "update
> discomfort" syndrome. However your comment about multilevel capability
> sounds like a request for more statistics, not cosmetics. I would gladly
> trade all this new list, GUI and fancy graphics for an effective multilevel
> package that would run under Stata. Or are you suggesting that glamm8 is the
> answer?
> Can you provide any quantitative info on how much faster is glamm in Stata8
> than under Stata 7? This is the sort of thing that would make me want to
> update - not the graphics, GUI, etc.
>
> Buzz wrote:
> "I for instance, would
> like them to keep focusing on developing the capacity for heirarchical
> models...the issues surrounding "list" pale for me compared to the need for
> stronger hierarchical modeling capabilities."
>
> Al Feiveson
> *
> *   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/