The choice of voxel-wise threshold is always a little tricky. For standard gaussian random fields, you don't want it to be too liberal because the theory breaks down. This not a problem with the simulations. The voxel-wise threshold does not affect the false positive rate (the p-value). The affect on the true positive rate is a power question and depends on your data. There is not (or should not) be an interaction between the number of iterations and the p value, though the more iterations the more accurate the p-value should be.
doug
Dankner, Nathan (NIH/NIMH) [F] wrote:
Hi all,
I have a rather general set of questions regarding the dynamics of the cluster simulation. I recently ran a simulation with 10,000 iterations with the threshold at 2 (.01), and received an output that seemed far too liberal, while when I ran it at 5,000 iterations, it appeared to be reasonable. Is there an interaction between threshold level and p value, such that increasing iterations allow areas of dubious significance to become spuriously significant given low threshold levels? I also am having a hard time determining what is kosher in terms of the threshold. Is .01 too liberal? Is there a general consensus on which level to use?
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer