[R] Any package to perform HLM (PROC GENMOD) like logistic regression in R?

[Prof Brian Ripley]

On Fri, 21 Oct 2005, William M. Grove wrote:

> I know there are very nice facilities in Pinhiero and Bates for doing
> HLM-type modeling for continuous dependent variables.  But I would
> like to be able to do repeated measures logistic regression, or LR on
> clustered observations (at least with exchangeable correlation
> structures, and preferably with more general choices of
> within-cluster correlation structures).  Have I missed seeing some
> package in R that will do this, more or less the way GENMOD does in
> SAS?  I have not found a package to do thus in my scanning thus far.

Well, you will need to tell us what it is that SAS does (I do have SAS but 
my licence code needs renewing which shows how little I use it).

There are two approaches to repeated measures logistic regression, the 
subject-specific and population-average approach.  Your description sounds 
closer to the latter, for which see the packages gee and geepack.
But GLMMs (the subject-specific approach) are covered in a number of 
packages including MASS, lme4 and glmmML.

My belief is that PROC GENMOD uses the GEE approach, and I would not call 
that an HLM.  To add confusion, there seems to have been a stand-alone 
program called GENMOD which is referenced in HLM-fitting reviews.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595