I'm running PROC COUNTREG for count data (n~6800) and that is fine. But when I try to run the model for subgroups with the "by" option, one of the two categories shows nulls for the _Alpha parameter, and the log says the Hessian matrix is singular. The variables do not have high collinearity (I checked), nor is there a dumb error like putting a continuous variable in the CLASS line or leaving a categorical variable out of the CLASS line. The "by" splits the data into ~5800 and ~1000 record groups. The problem is with the ~5800 group, the lower record group runs fine.

Any ideas would be greatly appreciated. Thanks!