Jim,
Is your model like the following?
Y = a + b*X + e
where Y is the beta that corresponds to average predictor error for each subject. Does the *average* prediction error vary substantially across subjects?
> what do those means and t-stats REPRESENT? Do they represent the main effect of each factor, NOT controlling for the other?
Could you show the output of the following command?
3dinfo -verb outputFrom3dttest++
> Is it possible to define an interaction term, in this case? Is it possible to do this with 3dMVM? Or is it impossible, because it is not really multivariate?
How many explanatory variables do you have in the model? It's possible to include interaction terms in both 3dttest++ and 3dMVM as I mentioned in my previous response.
Gang