why choose ParameterEstimates not GEEEmpPEst in PROC MIANALYZE ?

[xiangpang]

Hello, 

I am trying to run a proc mianalyze on my genmod ouput. I see many example use ParameterEstimates to do analysis. But I thought ParameterEstimates (a_mvn) used initial parameter estimates for analysis from GEE. Why not use GEEEmpPEst result  (b_mvn)to do analysis in proc mianalyze step? please correct me if my understanding is wrong.

Thanks for help

proc genmod data = mi_mvn ;
model read = write female math progcat1 progcat2/ dist=bin link=logit;
		repeated subject=id / type=ind PRINTMLE;
ods output ParameterEstimates=a_mvn;
ODS output GEEEmpPEst =b_mvn;
run;

proc mianalyze parms=a_mvn;
modeleffects intercept write female math progcat1 progcat2;
ods output ParameterEstimates=POOL;
run;

[StatDave · Accepted solution]

Yes, if you are using the REPEATED statement to fit a GEE model, then the parameter estimates from the final GEE model are in the GEEEmpPest object. As you note, the ParameterEstimates object only contains the estimates from the initial model that begins the GEE estimation process, so would not generally be of interest when using a GEE model. However, the ParameterEstimates object is what you would use for non-GEE models which are fit when the REPEATED statement is not specified.

[xiangpang]

Thanks for your confirmation.