Incorrect results may be produced by PROC GENMOD if you specify full-rank parameterization (other than PARAM=GLM) for some predictors in the CLASS statement and also use the AGGREGATE option (without specifying any variables). For example:

   proc genmod;
     class a b / param=ref;
     model y = a b / disb=binomial aggregate;
     run;

To avoid the problem, specify the variables defining the aggregation or use GLM parameterization (PARAM=GLM). For example:

   proc genmod;
     class a b / param=ref;
     model y = a b / disb=binomial aggregate=(a b);
     run;

or

   proc genmod;
     class a b / param=glm;
     model y = a b / disb=binomial aggregate;
     run;

Operating System and Release Information

A fix for this issue for SAS/STAT 9.21_M1 is available at: