Hi Sally,
Such a difference makes sense. In your larger voxel result the cluster volume goes by the multiples of 70.3125 (=3.75*3.75*5) mm^3 while in the smaller voxel case it is much easier to form clusters since it could increase by 1 (1*1*1) mm^3.
Now the question is: How big is the difference? With both alpha values, the difference (281-233=48 mm^3, and 352-291=61 mm^3) < 1 EPI voxel. Yes, it makes sense to run simulations on EPI resolution since that is the original voxel size, and it would be a little more lenient if done with finer grids. Considering the arbitrary connectivity radius and the strenuous process of creating a group mask in EPI resolution, I feel running simulations on tlrc space is a compromise. How do you usually create such a group mask in EPI space, or a group mask of EPI grids in tlrc space?
Gang