Dear Glad,
Looking at contrasts like [1 0] within an ANOVA is statistically misleading, this contrast should be based on the DOF reflecting the number of subjects within that group, so either
1) two separate one-sample t-tests to present activations of patients and controls, and a two-sample t-test to test for differences [1 -1] or [-1 1]
2) a two-sample t-test with contrasts [1 1] = activations averaged across groups and contrasts [1 -1] or [-1 1] for differences.
Based on the figures there seems to be another problem though. You seem to have covered only a part of the brain, something like several slices oriented with a ~45° angle. Comparing old_one_sample_ttest.png and old_two_sampled_ttest.png, the lower and more anterior significant voxels are missing in the two-sample t-test. For the young subjects, all the upper and more postior activations are lost in the two-sample t-test. The number of voxels within the two-sample t-test reflects voxels covered within patients and controls (or better, covered in all the different subjects), and in your case it seems you covered different regions within different subjects/groups, with a small overlap across all the subjects. In that case you should definitely use the mask from the two-sample t-test (voxels covered in all the subjects) for the one-sample t-tests. Otherwise activation maps for the different groups would be misleading. The results from the one-sample t-tests look as if young people show activations in more posterior regions, whereas the elderly do not, but this is actually due to the fact that this part of the brain is not part of their analysis.
Just in case, if you haven't planned to cover only a subset of voxels (but I guess this is as expected) , you should definitely check your preprocessing pipeline (errors during coreg, normalisation).
Best,
Helmut
|