[R] GEEPACK vs GEE: What are the differences in the estimators calculated by geeglm() (GEEPACK) and gee() (GEE)?

[Mohamad Burjak]

Hello,

I am interested in running generalized estimating equation models in R.
Currently there are two main packages for doing so in R, geepack and gee. I
understand that even though one can obtain similar to almost identical
results using either of the two, that there are differences between the
packages.

The paper that introduces the geepack package (
https://www.jstatsoft.org/article/view/v015i02) states that the three
features that distinguishes this package from others in R and other stats
software programs include:

   1. There is an interface function geeglm which is designed to be as
   similar to glm as possible
   2. A jackknife variance estimator is available as an alternative to the
   sandwich estimator
   3. Covariates can be incorporated into the scale and correlation
   parameters in a similar fashion to the mean modeling

However from other pages, it also appears that the underlying algorithm
these packages use to obtain the results are slightly different. According
to this post (https://stat.ethz.ch/pipermail/r-help/2003-October/040995.html),
The two packages are using different estimators for the correlation
parameter, and therefore different weights for the observations. From my
understanding, while you can choose between multiple estimators using the
geepack package, that both involve the use of the sandwich estimators as
the default.

I tried reading the documentation for both packages, but could not find
anything going into detail about the estimators used by these packages and
the specific differences between them.

>From my use of both these packages on the same data and regression, results
can be near identical or at least very close. Is there anything out there
that goes into specifics about the differences between these packages?


Thanks,

Mohamad

	[[alternative HTML version deleted]]