PROC GLM constrains categorical data parameterization such that the coefficient for the last level is automatically zeroed, which can often be handy. But I'm trying to create estimates constrained such that the sum of all the level coefficients is zero, like that produced in PROC CATMOD. Any ideas? I'd use PROC CATMOD instead, but I don't have any fancy response function in mind, just a straightforward least squares regression.
Many thanks, - Jonathan