> If you perform a Bayesian comparison between, say, model 1 (modulation
> by the main effect) and model 2 (modulation by the interaction), what
> kind of answer would this comparison provide?
there are a number of relevant references out there about this, but you
might want to start with:
Penny WD, Stephan KE, Mechelli A, Friston KJ.
Comparing dynamic causal models.
Neuroimage. 2004 Jul;22(3):1157-72.