I have data from a study where institutions were randomized to an intervention or control group. The outcomes are in principal binary at the person-level but as the data came to me there is no person level ID or covariates -- all I have are the institution ID, group assignment, and the number of individuals positive for a condition and the total number of individuals -- basically the raw material for the events/trials syntax. One additional feature is that following randomization data was collected quaterly for eight quarters.
Is it possible to use Genmod with GEE to account for two sources of dependency -- i.e. dependency induced by individuals being in the same instution in the same time period, and the association within an institution accross time periods?
It sounds like you just need to use events/trials syntax and then specify institutions as the SUBJECT= effect indicating that the sets of trials are correlated within institutions. The data would contain one observation for each quarter. For example, if you assume an AR correlation structure across quarters:
class instID quarter group;
model NumPos / NumSubj = group;
repeated subject=instID / type=AR within=quarter;