Hi Jasmin,
The reason that the option '-xdiff' is not available for all model types in 3dANOVA3 except '-type 1' is because it does not make sense to estimate the difference between any two cells in a model with random effects (and nesting as well in type 5). Only with fixed effect is such a cell difference estimation meaningful.
And it seems to me not correct to include AUC contrasts as additional factor levels because all of the levels for each factor in 3dANOVA have to be independent with each other. Therefore if you really want to do group analysis on some contrast results, you would probably have to run 3dttest with a slightly different assumption (i.e., fixed effect).
I don't really understand your 2nd question because I assume that it is usually those beta coefficients (percent signal changes), not the corresponding statistical values, from 3dDeconvolve that should go into 3dANOVA. Could you please elaborate the situation a little more?
Gang