AFNI version info (afni -ver): Latest
Hi experts,
How can I do voxelwise FWE using Bonferroni correction in AFNI? I ran a 3dLMEr and would like to find voxels in contrasts that survive Bonferroni correction.
Bonferroni is usually considered far too strict a correction term, esp. in voxelwise brain analysis.
If you really wanted to do that, you would take your p-value of choice and then divide it by N voxels in your brain mask, and use the result as a threshold value. So, if your p=0.01 and you have 10,000 voxels in your mask, you would threshold at p=0.000001.
It would be worth considering the lessons here, that showing more results helps reproducibility, interpretation and understanding:
- Taylor PA, Reynolds RC, Calhoun V, Gonzalez-Castillo J, Handwerker DA, Bandettini PA, Mejia AF, Chen G (2023). Highlight Results, Don’t Hide Them: Enhance interpretation, reduce biases and improve reproducibility. Neuroimage 274:120138. doi: 10.1016/j.neuroimage.2023.120138
Highlight results, don't hide them: Enhance interpretation, reduce biases and improve reproducibility - PubMed
--pt
Thank you. Yes, for whole-brain analysis Bonferroni might be too strict and unwarranted. However, suppose one is doing something similar to small-volume correction (SVC) like that in SPM or even FSL. In that case, I think an option to find significant voxels within the small volume using Bonferroni (rather than FDR or clustering) might be helpful.
The implementation/adoption of small volume correction highlights a fundamental issue with conventional multiple comparison adjustment methods (e.g., FWE and FDR): their sensitivity to the data domain’s size. Beyond this arbitrariness issue, there exist several other problems associated with these adjustments in common practice. To address these issues, prioritizing information integrity becomes crucial. One effective approach, as @ptaylor pointed out, is to promote transparency and reproducibility by emphasizing relevant findings without obscuring them through artificial dichotomization—following the principle of "highlight, but don’t hide".
Gang Chen