Yes. You need to demean CNI_1 and CNI_2 (you can find very clear explanation on the website:
http://mumford.fmripower.org/mean_centering/ by Dr. Jeanette Mumford).
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]]
On Behalf Of R Duke
Sent: Wednesday, May 16, 2012 3:45 PM
To: [log in to unmask]
Subject: Re: [FSL] Randomise Contrast
Cheers Yingying
There are ~25 subjects in each group. I just used 2 as an example to save on space. Given that, it looks like I should use the first design you suggested. Is that correct?
From: "Wang, Yingying" <[log in to unmask]>
To: [log in to unmask]
Sent: Wednesday, May 16, 2012 3:40 PM
Subject: Re: [FSL] Randomise Contrast
A question: are there two subjects in each group (G1, G2, G3)? Or one subject in each group with two scans?
For two subjects case:
G1 G2 G3 CNI_1 CNI_2
1 0 0 -0.667 -9.333
1 0 0 0.333 12.667
0 1 0 0.333 1.667
0 1 0 -0.667 -6.337
0 0 1 0.333 10.667
0 0 1 0.333 -9.333
For one subject case:
You’d better get the average FA map from the two scans. Then use the following design matrix:
G1 G2 G3 CNI_1 CNI_2
1 0 0 -0.167 1.5
0 1 0 -0.167 -3
0 0 1 0.333 1.5
Contrast1: 1 -1 0 0 0
Contrast2: 1 0 -1 0 0
Contrast3: 0 1 -1 0 0
Yingying
From: FSL - FMRIB's Software Library
[mailto:[log in to unmask]] On Behalf Of
R Duke
Sent: Wednesday, May 16, 2012 3:13 PM
To: [log in to unmask]
Subject: [FSL] Randomise Contrast
Dear FSL Experts,
I have a question regarding my Randomise design matrix. I have three groups (G1, G2, G3) and two covariates of no interest (CNI). I am interested in between group FA differences and feel that I should control for my two covariates of no interest. Given this information I was wondering if the design matrix I have listed below is correct?
G1
G2
G3
CNI_1_G1
CNI_1_G2
CNI_1_G3
CNI_2_G1
CNI_2_G2
CNI_2_G3
1
0
0
1
0
0
22
0
0
1
0
0
2
0
0
44
0
0
0
1
0
0
2
0
0
33
0
0
1
0
0
1
0
0
25
0
0
0
1
0
0
2
0
0
42
0
0
1
0
0
2
0
0
22
Contrast1: 1 -1 0 0 0 0 0 0 (give me voxels where FA for G1 > FA for G2)
Contrast2: 1 0 -1 0 0 0 0 0 0 (give me voxels where FA for G1 > FA for G3)
Contrast3: 0 1 -1 0 0 0 0 0 0 (give me voxels where FA for G2 > FA for G3)
I was also wondering if this design controls for within or between group variance?
In other words, what happens if I put my two covariates of no interest in only 2 columns (just listing the covariate of no interest across each group in two columns - one column for CNI_1 across all groups and one column for CNI_2 across all groups)? Would this be controlling for between group differences?
Cheers!