values of independent variables for a GLM command

I have a dataset with 2 dependent variables and 3 independent variables:
treatment and site, which are fixed, and therapist, which is nested within
treatment.  The design is balanced, with 2 sets of therapists nested within
4 sites, each site having 2 treatment conditions.  When I run a GLM with
the original therapist values (1-20) which are not sequential, I get
different results than if I run the GLM with sequential numbers (therapist
1 and therapist 2, each nested within treatment group and balanced across
site).  The default method for computing the sum of the squares is Type
III, which assumes that there are no empty cells. Type IV, however, does
not assume that there are no empty cells.

Whether I run the GLM with the original therapist values (1-20) or the
sequentially coded therapist values (1-2), Type III Sum of Squares gives
the same results as the Type IV Sum of Squares.  But, the model with
therapist coded 1-20 does not give the same results as the model with the
therapist coded 1-2.  I have plugged in other values for therapist (such as
101-120) and found that the results are also not the same as the previous
GLM models.

So, the question is:  WHY DOES IT MATTER WHAT VALUES ARE USED FOR THE
NESTED VARIABLE IN A MULTIVARIATE GLM MODEL?


Thanks for any help you might be able to provide.

Charla