Proc MIANALYZE for categorical interaction

[ProfB]

I am having trouble generating pooled effect estimates for an interaction between two categorical variables, using multiple linear regression with robust variance estimation and stabilized inverse probability of treatment weights - essentially a 'multiple informant' model.  The code I'm working with is:

Proc genmod data=data plots=(dfbetacs);  
	class studyid predict_1 (ref='0');
	model z_bw=predict_1*time race*time bmi*time age*time educat*time smoke*time/noint dist=normal link=identity id=studyid;
                repeated subject=studyid/type=ind ecovb;
		weight &weight;
		by _imputation_;
	ods output GEEEmpPEst=emp_est GEERCov=emp_covb 
        parmInfo=pinfo;
run;

Proc mianalyze  parms(classvar=level)=emp_est  covb = emp_covb parminfo=gmpinfo mult;
	class predict_1*time;
	modeleffects  predict_1*time;
run;

Predict_1 is a 3 level categorical variables (i.e., 1st, 2nd, 3rd tertile), time is a 2 level categorical variable. My model works fine but I'm unable to generate the pooled estimate for predict_1*time at each of the 6 potential levels.  However, I am receiving an error message that I've been unable to correct in that SAS doesn't appear to be recognizing the "predict_1*time" as an effect/Parm name:

"ERROR 22-322: Syntax error, expecting one of the following: a name, ;, -."
"ERROR 200-322: The symbol is not recognized and will be ignored."

Any ideas? Please!

[PGStats]

Use

class predict_1 time;

no asterisk.

[ProfB]

Thanks, but no luck, "The model effect predict_1 is not in the PARMS= data set." I don't have the individual effect of predict_1 in the model.

[SAS_Rob · Accepted solution]

Try this:

Proc mianalyze  parms(classvar=level)=emp_est;
	class predict_1;
	modeleffects  predict_1*time;
run;

You won't be able to use the MULT option since you have a CLASS variable.  This also means you do not  need the COVB= option either.

[ProfB]

Success!!! Thank you so much!