OK, having implemented that design matrix, the problem I'm getting is that it's rank deficient, on the order of 1E-17 (i.e. well past the recommended tolerance of 1E-4). Here's the matrix again, just for reference, with two subjects (one in each group):
Grp TRPR TRPO SHPR SHPO S1 S2
1 1 0 0 0 1 0
1 0 1 0 0 1 0
2 0 0 1 0 0 1
2 0 0 0 1 0 1
With contrast:
Con TPR TPO SPR SPO
Ixn -1 1 1 -1
... or (TPO-TPR)-(SPO-SPR).
In the actual design matrix, there are 22 subject EVs and multiple subjects per group. A TIFF of the full design matrix is attached, in case it's of interest. I believe the issue is in the first four EVs (call them the "experimental" EVs, since they represent conditions and not subjects). Going by the FEAT examples at http://www.fmrib.ox.ac.uk/fsl/feat5/detail.html, it looks like one easy way to deal with this is have the experimental EVs code factors rather than levels, and change the contrasts a bit:
Grp TT ST S1 S2
1 -1 0 1 0
1 1 0 1 0
2 0 -1 0 1
2 0 1 0 1
(where "TT" means "truly stimulated, time factor" and "ST" likewise for sham.)
With contrast:
CON TT ST
Ixn 1 -1
... and then I should be able to get the overall effect of time with:
CON TT ST
Time 1 1
... and the simple effects of time with:
CON TT ST
True 1 0
Stim 0 1
(I mention these latter things less because I'm interested in them and more as opportunities to betray any flaws in my reasoning... as if I needed more opportunities.)
I'm semi-confident about this, but if anyone has the time and inclination to check my logic, I'd be grateful. Thanks! (And thanks again to Donald for getting me this far.)
|