[18620] in s-news-athena
[S] Summary: generalized linear mixed models
daemon@ATHENA.MIT.EDU (Melanie Wigg)
Tue Jul 13 14:10:11 1999
From: Melanie Wigg <mbwigg@navigator.math.uwaterloo.ca>
Message-Id: <199907131755.NAA12746@navigator.math.uwaterloo.ca>
To: s-news@wubios.wustl.edu
Date: Tue, 13 Jul 1999 13:55:30 -0400 (EDT)
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Hi Splus users,
I posted the following question regarding fitting generalized linear
mixed models in Splus. It sounds like this issue has come up before.
There doesn't seem to be software in Splus to do this - using SAS appears
to be the best bet.
My original question and the responses I received are included below.
Melanie Wigg
Original question:
------------------
A few weeks ago Jose' Pinheiro posted the following to the list:
> Version 3.0 of the NLME library, software for fitting and analyzing
> mixed-effects models in S-PLUS, is being released after extensive beta
> testing.
This software covers a broad range of mixed effects models. If I
specifically want to fit generalized linear mixed models, is this the
software I should use? Or is there software available to fit these
types of models specifically (either as part of the main Splus package
or not)?
I use Splus 3.4 or 5.0 on Unix.
Thanks for your help,
Melanie Wigg
University of Waterloo
mbwigg@navigator.math.uwaterloo.ca
-----------------------------------------------------------------------
From: Chuck Cleland
--------------------
You may want to look at the following thread which came up on s-news
back in May.
http://www.biostat.wustl.edu/hyperlists/s-news/199905/msg00274.html
-----------------------------------------------------------------------
From: Viswanath Devanarayan
---------------------------
I think NLME is suitable for fitting only nonlinear mixed models with
continuous response.
I think your best option would be the GLIMMIX macro of SAS to fit
generalized-linear mixed models.
-----------------------------------------------------------------------
From: Ming Ji
-------------
I posted a question before in newgroup collecting references on GLMM.
It seems that there is no standard software for GLMM yet although
different researchers may have their own programs. BTW, GLMM is far more
general and difficult than NLME. From the responses I got, it seems
that there are only SAS macro GLMMIX and Genstat that can fit GLMM. Then
GLMMIX may lead to biased estimates.
GLMM is still an active research topic. Jiming Jiang showed the
consistency for his method of moment estimate. Thomas TenHave used
approximation to the multivariate normal density to develop his algorithm
for logistic regression with (Gaussian)random effects. There are other
algorithms basded on Monte Carlo EM and Iterative Re-weighted REML (in
Genstat). If you have new information about the progress in this topic,
please also forward a copy to me.
I have attached the summary of my question with regard to GLMM for your
reference.
============================Attachment=====================================
From: James MacDougall <jmacdoug@bios.unc.edu>
Subject: Re: References for Generalized Linear Mixed Effects Models
Organization: The University of North Carolina at Chapel Hill
Ming Ji,
There is a SAS Macro called GLIMMIX which sits atop SAS Proc Mixed. It
performs Generalized Linear Mixed Effects Models. It is well documented
code and I think there are some good references in the code documentation.
I found the Macro via an AltaVista search, but SAS's webpages may have it
as well (if you have trouble let me know and I will e-mail it to you).
I have heard that the results of Generalized Linear Mixed Effects Models,
particularly logistic random effects models are biased. I have not
researched this but if you know of any references that discuss this I
would like to see it.
I hope this helps.
Jim
----------------------------------------------------------------------
From: Timothy R. Johnson <tjohnson@s.psych.uiuc.edu>
Subject: Re: References for Generalized Linear Mixed Effects Models
As far as software is concerned, you might be interested in Donald Hedeker's
suite of programs for a variety of mixed-effects models. Look at
http://www.uic.edu/~hedeker/mix.html
Tim
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Timothy R. Johnson
Division of Quantitative Psychology
Department of Psychology
University of Illinois @ Urbana-Champaign
mailto:tjohnson@s.psych.uiuc.edu
http://www.psych.uiuc.edu/~tjohnson/tjohnson.html
-------------------------------------------------------
From: peter.lane <peter.lane@bbsrc.ac.uk>
Dear Ming Ji
Here are some papers:
Breslow & Clayton (1993) JASA 88, 421, 9-25
Schall (1991) Biometrika 78, 719-727
Karim & Zeger (1992) Biometrics 48, 631-644
Lee & Nelder (1996) JRSS B 58, 619-678
I know that John Nelder and Youngjo Lee are currently working on improvements
to their hierarchical GLM approach, which should be published soon.
I am biased about software, being a developer of Genstat. However, you may be
interested to know that there is a GLMM procedure in Genstat's standard
library, written by my colleague Sue Welham, and John Nelder has constructed a
whole subsystem of Genstat, cryptically called HG (freely available, but
requiring the standard system) to fit HGLMs. Take at look at the Genstat
home-page:
http://www.res.bbsrc.ac.uk/stats/work/genstat/index.htm
Peter Lane
Peter W Lane Statistics Dept, IACR Rothamsted, Harpenden AL5 2JQ, England
peter.lane@bbsrc.ac.uk tel: +44 1582 763133 ext 2372 fax: +44 1582 760981
---------------------------------------------------------------------------
From: Jos Jansen <aja@telekabel.nl>
To: mji@wald.ucdavis.edu
Subject: GLMM's
Dear Ming Ji,
You may have a look on the papers:
Engel, B. and Keen, A. (1996): An introduction to generalized linear mixed
models. Invited paper XIII th International Biometric Conference. July 1996
Amsterdam. (and the other papers presented in this meeting on the topic)
Engel, B. and Keen, A. (1994): A simple approach for the analysis of
generalized linear mixed models. Statistica Neerlandica 48, 1-22.
or contact dr Engel (b.engel@id.dlo.nl).
Kind regards,
Jos Jansen
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
