Hi Paul,
Thank you for the detailed response. After investigating this a little more, I see that I have conflated two separate issues. With the example I gave above, the two actually are pretty much identical (though I take your point on the wisdom of using bisided vs. 2sided). Embarrassingly, I think the discrepancy I saw was because I checked the "pos" button and thought I was missing some clusters. Oops!
However, the double p value does come up when I use 3dClusterize with a 1-sided test. So here is what I would request, in this case from an F-statistic.
3dClusterize -inset filename.nii \
-mask mask.nii -ithr 5 \
-idat 5 -NN 1 -1sided RIGHT_TAIL p=0.001 \
-clust_nvox 11 -pref_map test3.nii
Here is the GUI report for what I think should be the equivalent (it isn't):
3dClusterize -nosum -1Dformat -inset filename.nii \
-idat 5 -ithr 5 \
-NN 1 -clust_nvox K \
-2sided -7.3247 7.3247 \
-pref_map Clust_mask
* I can confirm 7.3247 is the threshold for p < 0.001 in the GUI in this example
To actually match the clusters generated through my 3dClusterize command, I need to change the p value to 0.002 in the GUI. I guess this narrows it down to my misunderstanding how the p-value is set for a 1-sided versus 2-sided test. Two follow up questions would be:
1. Should I be using a bisided test even for this F statistic? (I thought 1-sided was appropriate here).
2. Since I requested a 1 sided test, is this why the p-value is double in the 3dClusterize output? Do I need to adjust the desired p value to half in the command?
Thanks again for your help!