Can you model more than one cluster in Generalized Estimating Equations?

[rebeccachau]

Is it possible to model for more than one type of cluster with GEE? I have been attempting to cluster not only by participant (within-person correlation T1 to T2, only two measurement points) as well as by facility in which each participant resided (nursing home). However, I notice with my SAS syntax that when I include modelling by facility (by including my variable "FAC" in the CLASS line), that the estimates are identical to when I do not include this variable. This makes me think that something is awry. Here is my code (which uses multiple imputation):

PROC GENMOD DATA=LONGGEE;

CLASS ID TIME FAC;

MODEL GDS=DAYSOLD GENDER EDUCCOLL ADL COGTOTSAS NHP PEFITTOTSAS SD SOCTOTSAS / DIST=POISSON LINK=LOG COVB;

SCALE=1;

BY _Imputation_;

REPEATED SUBJECT=ID / TYPE=unstr MODELSE PRINTMLE MCOVB ECOVB CORRW;

ODS OUTPUT

ParameterEstimates=norepeat_est

Covb=norepeat_covb

ParmInfo=parminfo

GEEModPEst=mod_est

GEENCov=mod_covb

GEEEmpPEst=emp_est

GEERCov=emp_covb

;

RUN;

PROC MIANALYZE parms = norepeat_est covb = norepeat_covb parminfo=parminfo;

CLASS ID TIME FAC;

MODELEFFECTS DAYSOLD GENDER EDUCCOLL ADL COGTOTSAS NHP PEFITTOTSAS SD SOCTOTSAS;

RUN;

PROC MIANALYZE parms = mod_est covb = mod_covb parminfo=parminfo;

CLASS ID TIME FAC;

MODELEFFECTS DAYSOLD GENDER EDUCCOLL ADL COGTOTSAS NHP PEFITTOTSAS SD SOCTOTSAS;

RUN;

PROC MIANALYZE parms=emp_est covb=emp_covb parminfo=parminfo;

CLASS ID TIME FAC;

MODELEFFECTS DAYSOLD GENDER EDUCCOLL ADL COGTOTSAS NHP PEFITTOTSAS SD SOCTOTSAS;

RUN;

[StatDave · Accepted solution]

See this note, particularly the last section on nested structures.

[rebeccachau]

Perfect. Thank you.