I notice a disparity in the results of cluster analysis from the Monte
Carlo simulation AlphaSim distributed with AFNI, and the CorrClusTh
script kindly provided by Tom Nichols. I admit this is not my area of
expertise, and seek enlightened commentary.
In http://brainimaging.waisman.wisc.edu/~tjohnstone/AFNI_I.htm, an
example AlphSim calculation on a dataset smoothed to 5mm FWHM says that
a cluster of 86 voxels would be significant at corrected Pc<0.05 for an
initial uncorrected threshold of Pu=0.01.
If I run a similar analysis using CorrClusTh on a dataset smoothed to
4mm FWHM, I find I need 248 voxels in a cluster to reach significance
using the same initial uncorrected threshold.
I'm puzzled that there is such a big difference in the two methods.
Many Thanks in Advance,
Jim Lee
|