[R-sig-ME] different fits for geese and geeglm in geepack?

[Paul Johnson]

On Wed, Nov 25, 2009 at 5:35 AM, Achaz von Hardenberg
<achaz.hardenberg at gmail.com> wrote:
>>>
> Dear all,
>
> I am trying to fit a GEE model on eagle productivity (number of hatched
> offspring per nest) using the geeglm function in the library geepack and I
> found an odd result.
> My understanding is that the function geese and geeglm should give the same
> fits, as actually geeglm uses geese to fit the model, providing a glm style
> output.

If you gave us the exact commands that were run & the data, we could
probably figure this out.

I was interested to observe the problem you found, but can't reproduce
it.  I've installed geepack and run the examples for geese.fit, and
then too exact same options and put them inside geeglm. I get the
exact same results.

As a result, I think you have either not run exactly the same model in
the 2 cases, or geeglm is putting some settings in an unexpected way.

I think you will find your answer if you study the defaults as they
are set in geeglm.  Type

> geeglm

It will show the actual formula, you can see options that the function
is setting on your behalf. It appears to me geeglm assumes your link
function is identity, but the other might offer you the usual poisson
with a log link.

Then again, with a working example of geese.fit and geeglm that give
different estimates, I could probably get further.

-- 
Paul E. Johnson
Professor, Political Science
1541 Lilac Lane, Room 504
University of Kansas