Can I use GEE for repeated measures -AND- somehow adjust for clustered sampling?

[Marc Gameroff]

I have a sample of ~120 subjects assessed 3 times (longitudinally) for
depression severity. I'm analyzing change in depression scores via GEE with
the GENLIN procedure. However, subjects come from 5 distinct villages, and I
want to adjust for any clustering effect of village. My understanding is
that to do the latter, I need to use one of the procedures in the Complex
Samples add-on, such as the CSGLM procedure. In other words, GENLIN seems
designed to take care of correlation among measures WITHIN a subject, while
the correlation corrected for in CSGLM or the other Complex Samples
procedures seems to be correlation BETWEEN subjects belonging to different
clusters (e.g., villages, school, families).

I'm unaware of a way to accomplish both my goals using ONE procedure.
Perhaps I could do the GEE analysis and then do a separate analysis to
estimate the design effect from sampling distinct villages, and then somehow
apply the design effect to the GEE results (perhaps manually) to "correct"
the standard errors from the GEE output?

Does anybody have any advice? Thanks in advance!

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Bruce Weaver]

Hi Marc.  Have you considered a multilevel model (via MIXED or the newer procedure in v19) with Level 1 = Occasion, Level 2 = Subject, and Level 3 = village?

HTH.

Marc Gameroff wrote

I have a sample of ~120 subjects assessed 3 times (longitudinally) for
depression severity. I'm analyzing change in depression scores via GEE with
the GENLIN procedure. However, subjects come from 5 distinct villages, and I
want to adjust for any clustering effect of village. My understanding is
that to do the latter, I need to use one of the procedures in the Complex
Samples add-on, such as the CSGLM procedure. In other words, GENLIN seems
designed to take care of correlation among measures WITHIN a subject, while
the correlation corrected for in CSGLM or the other Complex Samples
procedures seems to be correlation BETWEEN subjects belonging to different
clusters (e.g., villages, school, families).

I'm unaware of a way to accomplish both my goals using ONE procedure.
Perhaps I could do the GEE analysis and then do a separate analysis to
estimate the design effect from sampling distinct villages, and then somehow
apply the design effect to the GEE results (perhaps manually) to "correct"
the standard errors from the GEE output?

Does anybody have any advice? Thanks in advance!

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD