> In our lab we've done PSC calculations using similar approaches (if for shorter
> lengths of time) and just averaged the 3-4 betas (say between time 8 seconds
> and 12 seconds) where it looks like the response is greatest (after examining the
> iresps). We've never done an AUC - I just threw it out there to consider.

Focusing only on a few effect estimates and summing them may lose the statistical power in detecting some shape subtleties. It would be more preferable to use all the betas without any dimensional reduction.

> Yes, I plan to do a Group X Condition ANOVA at the group Level. Two groups,
> two conditions. Maybe a single voxelwise covariate (ASL) as well.

I'm still trying to patch up the following modeling methodology. If you don't mind, you can try this.

0) Make sure you have the most recent version of AFNI.
1) For the Group X Condition interaction, obtain the beta difference between the two conditions for each tent separately. You have one between-subjects factor Group, and one within-subject factor that you may call it time, which represents the those tents. Then run 3dMVM with option -mvE4 on the condition differences.
2) If you also want to see the group effect at each condition separately, run two more analyses (one for each condition) with 3dMVM -mvE4. You still have two factors: one between-subjects factor Group, and one within-subject factor tent.

Let me know how it pans out.

Gang