Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: xtmelogit vs hlm


From   Stas Kolenikov <skolenik@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: xtmelogit vs hlm
Date   Fri, 12 Mar 2010 10:51:48 -0600

Run a few simulations and see who does a better job in getting the standard
errors right. Skrondal and Rabe-Hesketh discuss the differences in numeric
procedures used by HLM and GLLAMM in their books (-xtm*- may be covered,
too), so check carefully what the algorithms are, and what is known about
their performance by now. Does HLM use PQL to perform maximization? I
personally have more trust in Stata than in HLM, mostly because of the
existing certification and verification procedures that must have included
simulations. Bobby Gutierrez might want to describe what was done to make
sure -xtmelogit- produces reasonable and expected results.

On Fri, Mar 12, 2010 at 3:21 AM, Neal Beck <nealinmadrid@gmail.com> wrote:

> Hola
>
> A colleague here uses HLM, I was curious how it compared to State's
> xtmelogit. The data are surveys in 13 countries, roughly 3000 respondents by
> country (no weighting used, about 39000 respondents). Taking a vanilla set
> of options in both HLM and Stata, and a simple model with random effects and
> one random coefficient (and about 8 fixed coefs), I compared my results to
> her's. Not surprisingly, the estimates of the fixed coefficients (and their
> se's) and the fixed part of the random coefficients (the means) are quite
> similar, within about 5%, close enough for government work. The sd on the
> intercept is quite similar, but the sd on the random slopes differ by a
> factor of 3. I do not own nor play with HLM, but many of my friends who do
> this stuff for a living swear by it. Does anyone have any thoughts on the
> difference.  I find it hard to believe that a difference of this magnitude
> could simply be an artifact of different variations of estimation method
> (there is decent precisi!
>  on on the sd of the random coefficient, the se on the sd is about 1/4th
> the size of the sd itself.
>
> (This is not a part of my research, but a question of whether for practical
> reasons a course here should switch from HLM to Stata, there are huge
> advantages to Stata but the difference in results makes this a hard sell to
> HLM people.)
>
> TIA
> neal
>
>
> Neal Beck
> On leave AY 2009 at the Instituto Juan March, Madrid Spain
>
>
>
>
> *
> *   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/
>



-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.

*
*   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/


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