Dear Ping,
My model didn't have any particular meaning, I just took some beta images to have some files for the model (con images are nothing else than a linear combination of beta images), and I went with 10 subjects for illustrative purpose. The order of the factors is up to you, just make sure that the numbers entered when setting up the design matrix in "Main effects & Interactions" makes sense with regard to the order of the factors in "Factors".
Concerning the design, as stated, I would go with a design that includes the three-way interaction as long as you don't have any good arguments not to include the three-way interaction term (no sig. effects on behavioral level does not mean there's no effect on BOLD level), and as it's just a different implementation, with the "simpler" design matrix based on the three-way interaction and factor subject (which also facilitates setting up the contrasts, as it's just A1B1C1 A1B1C2 A1B2C1 A1B2C2 A2B1C1 A2B1C2 A2B2C1 A2B2C2 Sub1 ... Sub n).
Leaving this aside, your contrast vectors seem to be correct for that design. Regarding validity, no idea. Yes with regard to the model? I never learnt to run post-hoc tests based on the df of the full model. But I also never learnt to run models with a pooled error term as it's conducted in SPM.
> I defined the regions using the interaction contrast of A X C, masking above regions (uncorrected p=0.05) I conducted the A2C1>A2C2
> contrast to find the active regions.
.05 uncorrected is a very liberal threshold. And probably you should control for no. of conducted post-hoc tests.
Best
Helmut
|