Dear Mark,
I'm not sure which kind of group model you ran, but it's likely invalid. You should go with two separate one-sample t-tests for each of the 2 contrasts instead. The thresholding/binarizing approach is correct.
Concerning your other question: (2) can be expected to result in more overlapping voxels overall than (1), as it would also show the overlap between clusters of contrast 1 and contrast 2 which are smaller than these 50 voxels. However, the overlap already present in (1) should equal the overlap in (2). No extent threshold (k = 0) means more clusters go into the mask for contrasts 1 and 2, and possibly more overlapping clusters, but not larger overlapping clusters. In other words, if the overlap is already present in (1) and becomes larger in (2), then this has to be related to a different initial voxel threshold or other modifications, but cannot be explained by the change of the extent threshold.
Best,
Helmut
|