[picchionid@cn0012 ~]$ afni -ver
Precompiled binary linux_rocky_8: May 6 2025 (Version AFNI_25.1.08 'Maximinus')
[picchionid@cn0012 ~]$ R --version
R version 4.4.3 (2025-02-28) -- "Trophy Case"
Dear Colleagues,
Hi. How are you? I understand that Bayesian approaches should be considered because they render the following discussion moot, and I promise to integrate them in the future. However, if possible, I would like to discuss the traditional approaches to multiple testing correction because I am optimizing some old analyses. Perhaps I only need help with my statistical musings, but should the minimum number of voxels in a cluster increase for a 1sided versus 2sided question when using statistics that can only be positive?
I understand the arguments a little when using statistics that can be positive or negative. For example, suppose I am using a t-test. The minimum number of voxels for a nonspurious cluster from 3dclustsim for the 1sided simulations should be higher than for the 2sided simulations. I understand that a little. The test is appropriately more liberal because I am only examining one tail, so the number of voxels for a nonspurious cluster should increase because more voxels will be significant at the level of the individual voxel thresholding. However, should this apply when the statistic can only be positive? For example, suppose I am using Chi Square. Should I use the 1sided or 2sided 3dclustsim simulations results? I do not understand why I should be punished, for lack of a better word, and use the 1sided simulation results when I am getting no reward, for lack of a better word, at the level of the individual voxel thresholding because the distribution only has one tail.
Sincerely,
Dante