A bug (misfeature) has been found in the S gee library (and thus in the R
gee library). The problem, which is shared by nearly all gee
implementations, involves the calculation of working correlations when
some clusters have only one observation. For compatibility reasons nearly
everyone uses the computing formula from the first SAS macro by Karim,
rather than the formula from the original GEE paper. This results in
biased estimates of the working correlation.
This does not affect the (large sample) validity of inferences from GEE,
which doesn't depend on the working correlation matrix. It may mean
lower efficiency in some cases.
A more complete description of the problem is given by the author of the S
library at
http://biosun1.harvard.edu/~carey/gee.html
As soon as the S code stabilises I will redo the port to R.
Thomas Lumley
------------------------------------------------------+------
Biostatistics : "Never attribute to malice what :
Uni of Washington : can be adequately explained by :
Box 357232 : incompetence" - Hanlon's Razor :
Seattle WA 98195-7232 : :
------------------------------------------------------------
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-r-help
mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !) To: r-help-request at
stat.math.ethz.ch
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
