st: Re: xtmixed does not converge after 12 hours

["Joseph Coveney" <jcoveney@bigplanet.com>]

Raquel Rangel de Meireles Guimarães wrote:

I wonder what to do when my hierarchical model does not converge. Below 
are two examples of unsuccessful models: the first did not converge 
after 12 hours, and the second could not estimate the standard error of 
the random effects. Is there an algorithm to maximize that is robust and 
can solve this problem?

--------------------------------------------------------------------------------

How about an approach that is something like that below?

Joseph Coveney

* Lighten the iterative burden
keep if ano == 2007

* Explore a compact, representative subsample
set seed `=date("2011-07-17", "YMD")'
generate double randu = runiform()
keep if randu <= 0.1

* Fit pilot model by noniterative GLS
xtreg matematica i.masculino i.branco i.pardo ///
    i.atrasado i.nse, i(escola) re
// Do you see a need for rescaling? If so, do it now.

* If pilot model looks good, then add
* school-by-fixed-effects interaction terms (random slopes)
xtmixed matematica i.masculino i.branco i.pardo ///
    i.atrasado i.nse || escola: nse /*_escola */ ///
    atraso /* _escola */



*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/