st: Re: random effects binary outcome

["Rodrigo A. Alfaro" <raalfaroa@gmail.com>]

I couldn't find what you claimed that is in the manual: "more quadrature 
points usually generate more stable estimation". Quadrature is usually very 
low number 12 or 20, you want to approximate the normal distribution, not 
drawing it!! The stability concept is to keep mostly the same results 
(coefficients) when you change the number of points of the quadrature. If 
your model does not fit the assumptions of normality neither 20 nor 100 
points will solve the problem. Now, if your concern is about precision (not 
stability)... maybe you could switch to adapted-quadrature coded in GLLAMM 
(type: findit gllamm).

Rodrigo.



----- Original Message ----- 
From: "Xiaodong Chen" <cxdgis@gmail.com>
To: "statalist" <statalist@hsphsun2.harvard.edu>
Sent: Monday, June 11, 2007 4:50 PM
Subject: st: random effects binary outcome



> Hi All,
> For random effects logit and probit models, we can specify the number
> of quadrature points (12 by default). According to the manual, more
> quadrature points usually generate more stable estimation. If
> neglecting the computation expense, are there any shortcomings of
> using many quadrature points (say 100) for xtlogit and xtprobit?
>
> Thanks for your help.
> *
> *   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/