Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: gllamm using stata 10


From   "Anders Alexandersson" <andersalex@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: gllamm using stata 10
Date   Wed, 28 Nov 2007 09:47:45 -0500

Last week, Christian Deindl <deindl@soziologie.uzh.ch> wrote:

> I have recently updated Stata 9 to Stata 10 and since then I'm
> getting quite different results using GLLAMM. Most of these
> differences are negligible. Unfortunately there is a huge
> difference in the following model. The significance of my
> macro-indicator change from 0.000 to 0.839 [in P>|z|, Anders' edit].
> Does anybody have experienced anything similar?
>
> syntax: gllamm   y x1 x2 .....   , i(id houseid country) f(binom)
> link(logit) eform

I have not been able to reproduce the problem.

Christian sent me his dataset and -gllamm- syntax privately. Christian's
original do-file was convoluted, and the model to be compared had over
20 predictors. I used a 64-bit PC with Windows XP and a 3.4 GHz CPU and
Stata 10/IC under version control for Stata 9.2 and Stata 10. After
4 days, I had to abandon the run of the modified do-file, because I
needed Stata for something else. In the meanwhile, Christian wrote he
could reproduce the problem using the main predictor only. It took about
24 hours to run this -gllamm- model. My v9.2 and 10 results were exactly
the same, and reproduced Christian's v9.2 exp(b) with 5 digits precision
(.89630). Christian's v10 exp(b) differs already on the first digit
(1.0187 instead of .89630), which I cannot explain. Christian is
investigating why his v10 -gllamm- results are so different.

Anders Alexandersson
andersalex@gmail.com
*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index