Hi Thomas,

> Although I don't personally have a problem with Gang's notion of testing for differences between each group's
> condition contrasts to get the interaction (i.e., Group1(Cond1-Cond2) - Group2(Cond1-Cond2)), some
> collaborators do not prefer this approach, and I anticipate that reviewers may balk at this approach.

I would like to know what is the basis for the concern regarding my suggestion in testing such an interaction.

> I would like to know the best way to test for the interaction in this case, using an omnibus test. In this case, is
> 3dMVM the only way to go?

The best way? For such a simple 2 x 2 ANOVA, both approaches should end up with exactly the SAME result for the interaction effect: 1) conventional ANOVA, or 2) my suggestion of Group1(Cond1-Cond2) - Group2(Cond1-Cond2).

> I had heard that 3dMEMA can take into account the statistic associated with each input contrast map, thus
> possibly allowing for an omnibus test of the interaction as opposed to a test for differences. Is this true? I had
> thought that the idea of using the contrast t statistic is only useful if you have some research question about
> the variability of your conditions, not if you want test for an interaction in a mixed effects analysis.

Using 3dMEMA in this case would basically adopt the same approach as Group1(Cond1-Cond2) - Group2(Cond1-Cond2). That is, you provide (Cond1-Cond2) and its t-statistic as input for 3dMEMA, and get the interaction effect. The MEMA addresses exactly the same question as the conventional testing strategy, but the unique aspect of the MEMA method is that it does not assume that the variability of (Cond1-Cond2) is the same across all subjects.

Gang