 |
 |
This sounds very conservative to me, like all Bonferroni corrections (see the FIL's slides on random field theory, etc)......imagine you have a cluster which has a 0.0125 uncorrected p-value. In a small volume correction that uses this correction method, your cluster will have to include at least 1/4 of the voxels in your search region to remain below a p-value of 0.05 FWE-corrected.RFT isn't this conservative (again see abovementioned slides for the reasons) Whether it is 'safe' or not depends on your VOI. In fact this doesn't matter whether you use the RFT based correction or a Bonferroni correction. If you use a very small VOI for a small volume correction, you are much more likely to have a voxel or cluster which survives correction, because you are overly constraining that correction and cheating the multiple comparison problem. R 2011/3/4 Victor Vorobyev <[log in to unmask]> > Dear SPM experts, > > Recently I read about a PET study, in which after the data were processed > (SPM2, 2x2x2mm voxel, 8mm smoothing, paired t-test) and a VOI defined, > authors used the following way of correction for multiple comparisons: > > a Bonferroni-corrected cluster-level P value for a cluster in the VOI was > calculated as a function of the cluster volume: > > Pcorr=P*Sv/Cv > > where Sv = number of voxels in the VOI; Cv = cluster volume. > > I wonder if this is a safe way to obtain a corrected P value for clusters > in a VOI? > If it is ok, in which case this method can be preferred over the one used > by SPM for cluster p value correction?. > > Thanks, > > Victor > >