Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: question on GLLAMM

From   "Stanislav Kolenikov" <[email protected]>
To   [email protected]
Subject   Re: st: question on GLLAMM
Date   Thu, 19 Aug 2004 14:53:21 -0000

--- In [email protected], lm0542 wrote:
> Dear Stata Experts,
> I would like to estimate a model of occupational
> choice of university graduates, taking as endogenous
> each of the following 3 sequential decision nodes:
> 1) choice of university course (polychotomous variable
> with 1 to K groups, e.g. economics, law, humanities
> etc..)
> 2) choice on labour market participation (binary
> variable: 1 = employed, 0 = non employed)
> 3) choice of occupation (polychotomous variable with 1
> to J groups, e.g. craft, menial, blue-collar, white
> collar workers etc.)
> I was thinking to use GLLAMM for this purpose but
> since I'm new to the program I'm not sure whether it
> might be suitable. 
> Thank you very much in advance for any suggestion.

I am not much of an expert, either, but one thing I know is that
unless you have the latest Cray at your disposal, the model should be
kept to a moderate size. The model you are trying to put forward will
have something like (K-1)*1*(J-1) correlated random effects, and the
time to converge is exponential in this number, as it involves
integration over the real space of that dimension. If you have a
dataset of say 10000 individuals, and you also want to take some
sample design clustering into account... you are doomed to wait for a
few weeks for the model to converge. (It took a few days with
-oprobit- link and the panel structure with just one random effect on
my computer.) Another computational difficulty is an artefact of the
multidimensional integration. You would want to have a good resolution
where the joint density is supported, but as far as you would have to
have a grid of points, it means that you should have a lot of
integration points in each of the dimensions, and the number of the
integration points is in fact the base of the exponent I mentioned before.

Chapter 9 of the manual does have the example of the -gllamm-
equivalent / extension of -mlogit-, but you also would need to look at
the multiple equations models with (outdated) -eq- command in Chapters
4 and 5.


*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index