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

Re: st: GLLAMM: how fast for models with latent variables?

From   Jean-Benoit Hardouin <[email protected]>
To   [email protected]
Subject   Re: st: GLLAMM: how fast for models with latent variables?
Date   Mon, 20 Oct 2003 17:26:46 +0200

Dear Philippe,
For a comparison, I use GLAMM to estimate parameters of an Multidimensional Item Response Model. I use generally simulated data with 2000 individuals. I model their response to 14 dichotomous items. I model 2 latent traits and none covariate.

I use 12 adaptive quadrature points per dimension (I have read that, where the dimension is 1 or 2, we can use a small number of quadrature points, so I could use less of 12 points).

I have a PC candenced to 1GHz and 512 Mo RAM.

Each estimation take until 45 minutes. With my old PC (Pentium 133MHz, 32 Mo RAM), the same process take more of 5 hours.

So your time of computing seem to me logical.

So I actually work to estimate the parameters of my model with GEE (15 seconds) and to estimate a posteriori the marginal likelihood by gaussian quadratures (only one use of the quadrature, no more at each step of the estimation procedure).

Jean-Benoit Hardouin,
Regional Observatory of Health
Hospital of Orléans, France

On Mon, 20 Oct 2003 09:50:52 +0200, Philippe Huber <[email protected]> wrote:

Hello all,
I am just new here. I started using Stata this week because I want to
estimate models with latent variables. I'm trying to use gllamm and I am
a little bit confused.
My model is rather simple: there are 10 manifest variables (5 normal and
5 binomial), 2 latent variables (or factors) and no covariates. I
experienced that the latent variables have free variance and the first
loading for each is set to 1. Is it possible to free these loadings and
set the factors variances to 1?
Secondly, I used 8 adaptive quadrature points and my sample is only 60
observations big. Is it normal that Stata needs about 25 minutes to find
a result (I have a 1900MHz CPU)? Am I doing something wrong?
I thank you for you help.
Philippe Huber
Department of Econometrics
University of Geneva

*   For searches and help try:

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