GLM with Linear Dependencies


Newbie to SPSS. Tons of stats and programming knowledge ... but helping a novice with their SPSS work.

If I was doing this as a regression in package where I had to type in my own variables, this wouldn't be an issue.

I have 2 fixed factors, A and B. There 3 values of A. B is a subdivision of A: each A has 5 B's.

If I were to do a regression (in some other program) with 3 A's, I'd create 2 dummy variables to include in the regression with an intercept to avoid a linear dependence.

GLM in SPSS takes care of this just fine when reporting ANOVA results.

Alternatively, if I wanted to put all 15 B's (3 x 5) as into a regression, I'd create 14 dummies, and add them.

GLM takes care of this just fine too when reporting ANOVA results.

But, perhaps I think that both A and B are insightful if included.

If I were doing this in a regression (in some other program), I could also use an intercept, the 2 dummies for A, and 3 sets of 4 dummies for the B's.

However, if I put both A and B into GLM as fixed factors, it delivers back flawed output, because it doesn't catch the linear dependence.

How do I tell GLM that there is a linear dependence?

Alternatively, is the preferred way to do this in SPSS to put only B in as a fixed factor, and then do the appropriate hypothesis tests to see if a group of B's can be restricted to be equal to one of the A's? If so, is there a primer somewhere?