I am performing vertex based cortical analyses, but am also interested in effects in specific subcortical regions (e.g. hippocampus and amygdala) for which I am performing ROI analyses. I am wondering how to control my vertex based analyses for my ROI analyses and vice versa. Do I simply perform an FDR or Bonferroni correction of the (clusterwise) p-values or is there a different way to approach this?

Sorry, I meant how to correct the p-value of my vertex based analyses for additional ROI analyses, and vice versa: how do I correct the p-value of my ROI analyses for the vertex based analysis I also performed. Does that make more sense?

I am interested in the effect of an exposure on brain functioning. Effects on the cortex are assessed using vertex based whole brain analyses. For my subcortical ROI analyses, I have extracted summary statistics and am performing regression analyses in a general statistical software.

In all, I am performing a whole brain surface based analysis, plus 4 regression analyses. All analyses are answering the same research question (effect of exposure on brain functioning), so it is my understanding that it would be good practice to correct my p-values (or alpha’s) for all the analyses I am performing.

My question is, how do I correct the p-value (or alpha) of my regression analyses for the other analyses? I am performing four regression analyses + a cluster corrected whole-brain analysis. Would I perform a Bonferroni correction? If yes, does 4 ROI analyses + a cluster corrected whole brain analysis equal 5 analyses (so my alpha would be .05/5=.01)? Or does the whole brain analysis get a different weight? Would I apply the same threshold to my vertex based analysis (so cluster corrected threshold of .01 instead of .05)?
Or is Bonferroni too strict?

I think it hasn’t been common practice to correct for multiple testing over and above the correction for multiple comparisons done within a single (whole-brain) voxel/vertex based analysis, but I am trying to figure out what could be done to more properly control for the number of tests/comparisons that I perform to answer a single research question. Any thoughts would be helpful!

As you’ve already realized, there is no effective way to correct for such parallel analyses (node-wise plus subcortical ROIs) under the conventional testing framework. I would just perform the correction separately: one on the surface, and one for those four ROIs. Ideally it would be better and more efficient to analyze all the data (cortical and subcortical) in one model.

The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.