Dear M,
if you have two subgroups (something like sex?) within each of the two groups (something like controls, patients?), then it seems it is actually a 2x2x2 design. You could set up a corresponding model with GLM flex. I think there was a bug somewhere if different groups had different numbers of subjects, but see https://groups.google.com/forum/#!topic/fmri_matlab_tools/KC1jh2PB5D4 for details.
Alternatively, you should also be able to go with models within SPM. If you have different numbers of subjects within groups, probably go with two Full factorial models:
1) A1 vs. A2, B1 vs. B2, interaction A x B (between-subject factors and interaction): Build a contrast (C1 + C2)/2 on single-subject level (= averaging across the two conditions), foward the con images into a 2x2 Full factorial with two between-subject factors A, B. Assuming the order is A1B1 A1B2 A2B1 A2B2, then F contrast [0.5 0.5 -0.5 -0.5] corresponds to main effect A (averaging across the two levels of B, thus +/-0.5), [0.5 -0.5 0.5 -0.5] corresponds to main effect B, and [1 -1 -1 1] corresponds to the interaction A x B (here we don't average, so +/-1).
2) C1 vs. C2 and interactions with C (within-subject factor and interactions): Build a contrast C1 - C2 on single-subject level, forward the con images into a 2x2 Full factorial. Assuming the order is A1B1 A1B2 A2B1 A2B2, then [0.25 0.25 0.25 0.25] corresponds to main effect C (averaging across the four groups, thus 1/4), [0.5 0.5 -0.5 -0.5] corresponds to interaction A x C, [0.5 -0.5 0.5 -0.5] corresponds to interaction B x C and [1 -1 -1 1] corresponds to the three-way interaction.
Best,
Helmut
|