Conjunctions versus Averages
A different situation arises when you want to see what is common across two runs. You have two choices:
1. Finding the average voxel beta (or z) score across both conditions
2. Finding a conjunction – eg, that the voxel was statistically significantly active in both conditions
As should be clear on reflection, using an average voxel score can be misleading, because a very significant score in one condition can ‘make up’ for an insignificant score in a second. For example, if voxel X in run 1 has a z score of 4.5 (significant) and 1.8 in run 2 (insignificant) its average z score is 2.3 and it will show up as significant in the group image.
Average analyses are what FSL automatically does when you set the group level regressor to be identical for all subjects, for example in the figure below, there are 7 subjects being compared along a since EV, and the experimenter wants simply to see what voxels, on AVERAGE, activate across all the subjects. By entering 1 in each EV box per each input (individual subject), each voxel will be averaged and, if the z-score that results exceeds the arbitrary threshold set, that voxel will be marked as significant on the final image.
Alternately, in a conjunction analysis, a significant cluster in run 1 is used as a mask for the same image from run 2. Anything active in both runs emerges as significant in the group image. This image thus ensures that the voxel met statistical criteria in both images. the way that you do such cluster analyses is at the group level.
By pressing the contrast masking tab, you get this GUI And the truth is you don’t understand it yet. I would have thought you wanted to mask each contrast by itself across all subjects to see what is common across subjects within a given contrast. Look into this.