> I have noticed that SPM does things one-sidedly
The issue of one- vs. two-sided testing: If there is prior information about the directionality of a contrast, one-sided testing is justified; Otherwise, two-sided testing is more reasonable, and two separate one-sided tests would be inflated. In fact, a two-sided thresholding is basically a Bonferroni correction for the two one-sided tests.
> if you run AlphaSim or 3dClustSim with an uncorrected p of .001, and it says that you need,
> say, 118 voxels for an alpha of .05, should the 3dmerge command typically use the value
> given on the AFNI slider for that p as the -1thresh option (in this case, 4.07--the 2-sided
> value for t(15)) or 3.73 (the one-sided value). As in, what is typically expected of the user?
> I feel as though the two-sided value is appropriate since in AFNI the contrasts are typically
> output as two-sided (i.e., if I contrast two conditions I get both positive and negative clusters)
> whereas in SPM the contrasts have to be run in both directions.
Yes everything in AFNI is associated with two-sided thresholding by default except for 3dClustSim/AlphaSim. However, the user can still adopt one-sided thresholding in AFNI by finding the threshold on the GUI that corresponds to doubling the desired one-sided p-value, and then use that threshold for -1thresh in 3dmerge.
Gang
Edited 2 time(s). Last edit at 12/18/2014 07:24PM by Gang.
