Hi Tom,

Design and contrasts are fine. For the first 3 contrasts, you can use "-e design.grp" so that permutations will happen only within block. For the remaining contrasts, use also "--permuteBlocks".

The groups can be unbalanced, but the number of timepoints need to be the same for all subjects.

All the best,

Anderson

On 9 December 2014 at 15:51, Thomas Nickson <[log in to unmask]> wrote:

Hi All,

I was looking to do a 3 factor, 2 level repeated measure analysis in randomise for some VBM that I am doing and thought that I would run the design matrix past the list to make sure that it's okay.

I looked through the list and found this

https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1005&L=FSL&D=0&1=FSL&9=A&J=on&K=2&X=190C12580BEC088223&Y=thomas.nickson%40gmail.com&d=No+Match%3BMatch%3BMatches&z=4&P=343446

and thought that it would be possible to extend it in the following way from 2 to 3 factors.

For 6 subjects I get the following design matrix:

Grp  G1   G2   G3   s1   s2   s3   s4   s5   s6
1      -1     0      0     1     0     0     0    0     0
2      -1     0      0     0     1     0     0    0     0
3       0    -1      0     0     0     1     0    0     0
4       0    -1      0     0     0     0     1    0     0
5       0     0     -1     0     0     0     0    1     0
6       0     0     -1     0     0     0     0    0     1
1       1     0      0     1     0     0     0    0     0
2       1     0      0     0     1     0     0    0     0
3       0     1      0     0     0     1     0    0     0
4       0     1      0     0     0     0     1    0     0
5       0     0      1     0     0     0     0    1     0
6       0     0      1     0     0     0     0    0     1

Where, Grp is the group variable, G's are the groups and s is for the subject variables. The negative numbers in the group column give T1 and non-negative give T2.

Contrasts:

G1   G2   G3   s1   s2   s3   s4   s5   s6

 1     0      0     0     0     0     0    0     0    Time2 > Time1 G1

 0     1      0     0     0     0     0    0     0    Time2 > Time1 G2
 0     0      1     0     0     0     0    0     0    Time2 > Time1 G3

-1     1      0     0     0     0     0    0     0    G2>G1

-1     0      1     0     0     0     0    0     0    G3>G1

 1    -1      0     0     0     0     0    0     0    G1>G2

 0    -1      1     0     0     0     0    0     0    G3>G2

 1     0     -1     0     0     0     0    0     0    G1>G3

 0     1     -1     0     0     0     0    0     0    G2>G3

Is this correct?

Also, is this only possible in a balanced design where there are equal number of people at both time points?

Thanks,

Tom