Hi Daniel, yes, it is true for an unbalanced design. The values that are being compared are averages. If you compared the average of a group of 15 to the average of a group of 5, you would not scale the averages by the number of members in the group.

doug

daniel geisler wrote:

Hi Freesurfers,

I am currently dealing with some examples of group analysis from freesurfer-wikipages (see http://surfer.nmr.mgh.harvard.edu/fswiki/Fsgdf4G0V). Could you please give some clarifications in the following scenario:

Assume we have four groups MaleRight, MaleLeft, FemaleRight, and FemaleLeft (from factors gender and handedness). Now we want to set a contrast that is equivalent to the question: "Is there a difference between Males and Females regressing out the effects of handedness?"

An appropriate contrast would be [0.5 0.5 -0.5 -0.5].

(see http://surfer.nmr.mgh.harvard.edu/fswiki/Fsgdf4G0V#Contrast5male-female.mtx)

Is that also true for an imbalanced design, where MaleRight and MaleLeft consist of 15 and 5 subjects, respectively? I suspect that in this case the above contrast would weigh MaleLeft more heavily, i.e. each subject in this rather small subgroup would have a greater effect on the final result.

If that is true - I am wondering whether the contrast [0.75 0.25 ...] would fix this issue, i.e. adjust for the unbalanced design. Do you have any ideas?

Best,

Daniel

---

---

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer