Hello Jennifer:

Sorry about taking so long to respond. My excuse is that the entropy of
the universe is increasing, primarily inside my office.

The 3dANOVA2 and 3dANOVA3 programs require equal sample sizes. Since, in your
case, the unequal sample sizes (n=14 and 16) are nearly equal, there would
perhaps not be a severe loss of statistical power if you randomly chose two
of the subjects to discard from the later group.

You are correct that "3dANOVA3 -type 5" is appropriate, since you have a "mixed
design with tasks and groups as fixed factors, subjects as a random factor and
with subjects being nested in groups." Here, group is the fixed factor A,
task is the fixed factor B, and random factor C, subject, is nested within A.

You are also correct that, when using AlphaSim to determine the cluster size
threshold, the input to AlphaSim should be specified in terms of the Orig
dataset dimensions, not Tlrc dimensions. Transforming the data from Orig
to Tlrc does NOT increase the spatial resolution of the data; you are merely
duplicating the data.

Important: The output cluster size threshold from AlphaSim is specified in
units of Orig voxels. Now, in order to actually perform the clustering, it is
necessary to specify the cluster volume in mm^3. So, to enter the 3dclust
"vmul" input, one must multiply the cluster threshold from AlphaSim (in # of
voxels) by the volume of an Orig voxel (in mm^3). Since the cluster size
threshold is now specified as a volume, it is independent of the size of an
individual voxel (in Tlrc space).

Using the smaller "rmm" in Tlrc space does not directly enter into this;
however, since the smaller rmm makes it more difficult to form clusters,
it is a conservative choice.

To answer your last question, I would prefer to put it this way:
Program AlphaSim corrects for simultaneous inferences (i.e., the multitude of
voxels) by calculating the tradeoffs between the individual voxel probability
threshold (Pthr) and the cluster size threshold (Cthr) which yield the desired
probability of a false detection anywhere in the volume (Alpha). Note that the
Bonferroni method also corrects for simultaneous inferences; however, it is
limited to Cthr = 1.

Both AlphaSim and Bonferroni assume that the externally calculated p-values
for individual voxels are correct.

Doug Ward