[R] significance tests with gee?

[Thomas Lumley]

On Mon, 10 Jun 2002, Jonathan Baron wrote:

> Any hints on how to do a significance test using gee?  Is the
> "robust z" the thing to use?

Yes. Typically.

The "robust Z" is a Wald test done with standard error estimates that are
(in large samples) accurate if the systematic part of the regression is
correctly specified and conservative otherwise.

The other Z test uses the working weight matrix as if it were the real
correlation, which is typically a bad idea (and in my opinion missing the
point of GEE).

However, with small numbers of independent groups the robust standard
errors are biased down. I'm happy with 40 groups and get really nervous
below about 20.  In this case you might do better by trying to pick a
working correlation structure that's close to the true correlation (or by
using some other technique entirely).


	-thomas

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
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
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._