Hi Alberto and other FSL users,
Thanks a lot for putting together this information on covariate issue's in FSL. Some questions pop up reading your post.
In scenario 3 you state: "you must demean over all subjects and not in the two groups separately, otherwise you will move some age-related variance to group differences one". Since groups are not matched, means are different. By separating the two groups you will control for age-related variance to group differences. This depends on what you would like to model. Are you checking for variance within each group or are you checking for variance across groups?
Scenario 4 goes into the age*group interactions. The average Age1 is unequal to the average Age2, since both are demeaned around a global mean. (example: x=-1, y=-2, z=+1 and w=+2). The contrast is a bit strange to me, since the second row has 1 for Age1 and -1 for Age2. This compares the ages in groups and shows significant voxels? And the third row, why do you set 0.5 and 0.5 for the weights?
Another more general thing I noticed is the additional Grp column. Does it have any function in FSL processing or is it just for a checking/visualisation purpose?
Kind regards,
Stijn Michielse
Research Assistant
Dept. Psychiatry and Neuropsychology
Maastricht University
E-mail: [log in to unmask]
|