Does the usual conversion of the logit regression coefficient to marginal effects (in percentage points) the same for GEE with clustered data?
For a logit coefficient of .68 I get roughly a 13 pp increase based on a outcome mean (proportion) of .26.
This is an interaction coefficient (2 by 2 dichotomous vars).
The EM means don't match the converted coeff. but that is another issue.
------------------------------
tille johns
------------------------------