Hi,
I have a multi-centre longitudinal dataset, and basically I want to
compare group1 and group2 on a binary variable. I would also like to know
whether the relationship differ across sites (4 sites). In one site, there
is no positive outcome for all periods in group1. As a result I can't run
my usual GEE. Either I remove that site altogether, or I do something
else.
I wonder, is it acceptable to add 1 or 2 observations to that site, so
that I can carry on with the analysis? For that group in that site, I have
only 14 subjects (around 3 to 4 longitudinal obs each).
Is there a rough rule which says that if you have n>something, it would be
ok to do something like this?
Thanks.
Tim
|