GEE: LispStat objects for Generalised Estimating Equation models
Description
GEEs are probably the simplest of the regression techniques for clustered
data or repeated measurements. They involve fitting a generalised linear
model with an assumed "working correlation". Inference is valid for any
working assumption but inefficient if the assumption is grossly wrong.
The LispStat gee objects are modelled on the glim and regression objects
but do not inherit from these as their internal workings are different.
Availability
You can download the current version from my Web page. Currently it implements Normal, Binomial and Poisson errors and independence and exchangeable working correlations.
I have checked it against an S-Plus gee() library but there may well still be bugs. It does have some documentation.
Project Investigators
Thomas Lumley (PhD student, Biostatistics, University of Washington)
Contact person
thomas@biostat.washington.edu