However, if you use setup A, doesn't that assume that all the measures came from the same person or be the same as if the observations came from the same person. It doesn't take advantage of the random subjects. In the case of setup A, you can't calculate the subject effect, but subject differences do account for some variance. In setup B, you lose that information. In the AFNI manual, after taking a closer look, Example 2 of 3dANOVA3 uses subject as a random factor and has only one datapoint per cell. Is this a bad example? I also, was looking at "Design Analysis, A researcher's Handbook" by Keppel under the chapters on within-subject designs. Of which I was talking about above was. Essentially, the two factors a and b are within-subject factors and the factor c is between subjects. Does that change the interpretation or decision on which to use? It seems like I would want to use his (AxBxS) matrix and only interpret the A and B factors.
Donald