Hi Anderson,
Thank you for the exceptionally clear explanations of both the FDR and FWE methods.
In looking closer at FWE, I've iteratively examined how a range of thresholds effect my corrp images. I think that too many of them survive fairly stringent thresholding (much of the maps remain at threshold 0.0002, so much so that I am suspicious of my original design setup. (Just to be clear hear I am directly thresholding the corrp files by say alpha = 0.01/(25 components x 2 directions)
Would the summarized below GLM design make sense with my original question:
- 2 groups, sorted into low IQ and high IQ
Number of main EVs: 2
Number of additional, voxel dep EVs: 0
Input Group High IQ Low IQ
1 2 0 1
2 1 1 0
3 2 0 1
4 ... and so on for a total of 64 subject, with high IQ being group 1, and low IQ being group 2.
Does this look right?
Alternatively, maybe I need to add a higher scale to the number of permutations in randomize, say 20,000 instead of 5000?
Thank you for your kind consideration,
Varina
|