Hi Y'all, I have just found a bug in our implementation of the permutation test. It does not affect all permutations, so read on to figure out if it affects you.
The bug is that the histogram that is created by the permutation simulation is always based on the last contrast in the analysis. The permutation p-value is computed from the histogram, so this means that the wrong histogram is being applied to the contrasts (except for the last contrast).
This affects you IF:
1. You are doing permutation tests with mri_glmfit-sim or mri_glmfit to correct for multiple comparisons 2. You have more than one contrast 3. You are reporting p-values from a contrast that is not the last contrast on the mri_glmfit command-line 4. Note: the monte carlo simulations are not affected.
Until I can find a better solution, the work-around is to run a separate instance of mri_glmfit for each contrast, then run mri_glmfit-sim separately for each mri_glmfit run.
doug