Dear Alejandro, The multinomial logit model in Zelig is implemented via VGAM, a package by Thomas Yee at the University of Auckland. The algorithm is described in Yee, T. W. and Hastie, T. J. (2003), ``Reduced-rank vector generalized linear models,'' Statistical Modelling, 3, 15-41. and a general reference for the multinomial logit model is McCullagh, P. and Nelder, J. A. (1989), Generalized Linear Models, no. 37 in Monograph on Statistics and Applied Probability, Chapman & Hall, 2nd ed. Best, Olivia ----- Original Message ----- From: Alejandro Buren To: zelig(a)latte.harvard.edu Sent: Thursday, December 01, 2005 9:41 AM Subject: [zelig] estimation method details Dear Zelig developers, I'm really glad this package is available for use, it's been really helpful. I'm using the "mlogit" model to fit multinomial logit models. I'd like to ask you which algorithm is Zelig implementing to maximize the loglikelihood, and if it's making any restriction (besides setting all the parameters of the baseline category to zero) to explore the parameter space. I'm asking because I wrote a piece of Fortran code which fits a multinomial logit model and I'm running it with two different simulated annealing algorithms, one just goes crazy and declares it's not able to do the job for me and the second finds a maximum but it has two flaws: 1. it's not computationally efficient (i.e. one run takes forever) and 2. it's not consistent, i.e. it ends up in a different spot of the parameter space depending on where I start it. I've seen that the algorithm Zelig implements doesn! 't have these flaws and besides it yields almost identical (even better!) results than SAS. My big concern is that I must get my Fortran code to work because it is just a subroutine of an ad-hoc simulation I'm trying to run. Thanks a lot, Alejandro ------------------------------------------------------------------------------ 1GB gratis, Antivirus y Antispam Correo Yahoo!, el mejor correo web del mundo Abrí tu cuenta aquí