Hi everybody,
I've got a question regarding correction for multiple comparisons in the overlap of clusters derived from LME and separately from robust regressions.
Briefly, I've performed some voxelwise LMEs (in R) on task related GLTs produced by 3dDeconvolve. The model was looking at the effect of group, so the R fixed effects model formula was fmri ~ group. There are 2 groups in the study, one control and the other clinical. The f-stat from the LME model was then thresholded voxelwise and corrected for multiple comparisons using 3dClustSim (p-uncorrected =0.05 and p-corrected=0.05).
Separately, using the same GLTs and only in the clinical group from my study, I've performed voxelwise robust linear regressions (also in R) to investigate whether any clusters can be identified from neuropsych measures such as the Beck Depression Inventory. These were also corrected for multiple comparisons in the same manner as above (p-uncorrected =0.05 and p-corrected=0.05).
Now I want to look at the overlap between the 2 sets of clusters, this is a simple 3dcalc command. Now I get to the sticky part, I need figure out how big a cluster in the overlap map should be for it to be significant. Should I simply multiply the corrected p-values from both sets of clusters (0.0025) and feed that into 3dClustSim (with p-uncorrected=0.05)? Should both p-values be multiplied and fed into 3dClustSim? Is this method even valid?
(If you need more info please ask.)
If you've got any thoughts on this I'd be grateful if you could share them.
Thanks,
Colm.