I am using Randomise with the TFCE function (using version Randomise version
2.1) to do a paired t-test, where my two groups are the same set of subjects
processed in two different ways. We want to see the voxel wise differences
between these two processing methods. I am running this on whole brain
anisotropy maps which have been scaled to values between 0 and 1000. I
used the -T option for TFCE, not the --T2, as I am interested in whole
brain, and did not run TBSS on my data. I can provide my design matrix if
that might help explain things.
The results are somewhat concerning, in that it appears that Randomise is
picking up on negative differences, but not positive differences. In other
words, there are significant differences in the tfce_corrp_tstat2 file, but
none in the tfce_corrp_tstat1 file. While qualitatively in our data we see
some positive and some negative differences. We are, therefore, concerned
that the results are biased in some way.
Has anyone else experienced anything like this?