1) when setting up a two-sample t-test I choose that I want to collect
suprathreshold stats but that I don't want to set a threshold at this time.
When I compute the stats it appears that no suprathreshold stats are
collected. (choosing voxel-cluster results produces an error:
??? Error using ==> snpm_combo_pp at 180
Suprathreshold stats not collected! Cannot do cluster-combining!
Look for a SnPM_ST.mat file; that holds all statistic values in all permutations above a given threshold (see snpm_defaults.m for the exact threshold). If you said 'Yes' to 'Collect Supra-Threshold stats?' and No to 'Define the thresh now?', it should have been created. Do you get the same behavior with the usual (non-combining) cluster inference?
One problem that arises sometimes is that it grows larger than 2GB, creating problems for some filesystems. One solution is to increase the threshold in snpm_defaults, so that fewer voxels are saved each permutation.
2) When examining Results if I choose voxelwise then no matter what number I
enter for an uncorrected threshold every voxel seems to show up in the
brain. I have 10 images and the uncorrected p-value says 0.0083 for all
voxels (which I understand is due to the limited number of relabelings) but
if I choose a threshold of 0.001 or 0.0001 I still see all the voxels.
That is very strange. How many permutations are there? What does the distribution of uncorrected P-values look like (it's one of the screens displayed)? Is it just one spike at 0.0083? Does the t image make sense? (i.e. it's not all one constant value?)
Sorry for the barrage of questions, but I've never run into this before.
-Tom
_________________________________________
Thomas Nichols, PhD