consistency when splitting a dependent variable

[Rob Shaw]

I'm running a regression on hospitalisation data. Each observation is a individual person and the dependent variables is the number of hospital attendances per person. In simplified form:

reg attendtotal x1 x2

(I'm actually using poisson regression)

So this gives me a functional output with coefficient of

attendtotal=A*x1 + B*x2 + Const

However, I also have a disaggregation of attendtotal by different reasons. E.g. to give

attendresons1=a1*x1 + b1*x2 + Const1
attendresons2=a2*x1 + b2*x2 + Const2

The problem is that some of the individual reasons have very small numbers. So my question is, is it possible to run a structured set of regressions with the constraints that:

a1 + a2 = A
b1 + b2 = B
Const1 + Const2 = Const

Thanks
Rob

[Carlo Lazzaro]

Rob:
you may want to separate the reason for hospitalzation from the number of hospital referrrals and include is as a categorical predictor in the right-hand side of your regression equation.