Clustering in GEE (PROC GENMOD)

[rebeccachau]

I am analysing a cross-sectional GEE model clustered by "facility" (nursing home). The dv is depression score, and I have a series of predictors (sleep, pain, cognition, activities of daily living). I have also completed multivariate normal multiple imputations, so am also using MIANALYZE. Here is my code:

PROC GENMOD DATA=WORK.IMPORT2;

CLASS FAC;

MODEL GDS= PEFIT ME PMS MSPSP / DIST=POISSON LINK=LOG COVB;

BY _Imputation_;

REPEATED SUBJECT=FAC;

ODS OUTPUT

ParameterEstimates=a_mvn;

SCALE=1;

RUN;

PROC MIANALYZE parms=a_mvn;

modeleffects intercept PEFIT ME PMS MSPSP;

RUN;

To explore the effect of the clustering, I have also constructed a model that omits the clustering. Here is the code:

PROC GENMOD DATA=WORK.IMPORT1;

MODEL GDS= PEFIT PMS ME MSPSP / DIST=POISSON LINK=LOG COVB;

BY _Imputation_;

ODS OUTPUT

ParameterEstimates=a_mvn;

SCALE=1;

RUN;

PROC MIANALYZE parms=a_mvn;

modeleffects intercept PEFIT PMS ME MSPSP;

RUN;

I have noticed the results are identical. This makes me wonder whether (a) I have coded the clustering correctly or (b) whether the "clustering" is negligible because the clusters are variable in size and some have very few cases. There are only 15 clusters for N=147. Still, I would have expected at least some very minor differences. Any ideas?

[SAS_Rob]

The issue I think is that when you use the REPEATED statement, you should be providing the GEEEmpPest table to MIANALYZE and not the ParameterEstimates table, which is always the "non-Gee" estimates.