Hi Josh,

> I would like to have enough power for my 3dANOVA to pick up on significant differences across three conditions.

Just to clarify: with 3 conditions, did you really use 3dANOVA or 3dANOVA2 -type 3?

> 1) Is there a way to easily run an FDR correction at a specified cluster level rather than the voxel level?

What do you mean specifically by "cluster level"? Performing FDR with a mask?

> It seems to me that if I were able to do a FDR correction for clusters >= 9 voxels

Unless you have a lot of voxels with p-value below 0.05 in the brain, I doubt that a cluster of 9 voxels with FDR correction would make it. However, even if you can't manage to make the clusters survive the stringent correction, it does not necessarily mean that you can't report the results.

> I am currently considering running my 3dANOVA with an OR_MASK comprised of my three conditions
> (vs. rest) at an alphasim corrected threshold of alpha <.05 (this would reduce my voxels from 34549
> above to 6752 voxels more active in any or all conditions versus rest). This is intuitively appealing to me,
> however I know of no precedence for doing this, does anything seem off about using an approach such as this?

The rigorous approach would be presenting the corrected main effect F-stat of the three conditions. The OR_MASK methodology is invalid as pointed out in the following paper:

TE Nichols,MBrett, J Andersson, TWager and J-B Poline. Valid conjunction inference with the minimum statistic. NeuroImage, 25:653–660, 2005.

Gang