SPM2 example dataset for more complex 2nd-level design with nonsphericity
=========================================================================

36 contrast images for 3 basis functions from 12 subjects are in
the directory ./con-images where rcon*img number is:

         3-14   = canonical HRF, sub 1-12
        15-26   = Temp derivative, sub 1-12
        27-38   = Disp derivative, sub 1-12

The contrast was the main effect of faces versus baseline (see ../rfxdata and
../advefmri).

These can be entered into a "one-way ANOVA" (no constant terms),
with "yes" to nonsphericity (inhomogeniety of variance) and
"yes" to repeated measures (repeated over "replications (12)").

The estimates are for the basic event-related BOLD response to 
faces versus (chequerboard) baseline (the faces also required 
a motor response - hence the motor activation).

The original contrast images (con*.img) were created in SPM99, 
which in SPM2 would appear LEFT-RIGHT flipped. We therefore
created flipped and resliced con images (called rcon*.img) where 
left is left in SPM2. The motor cortex activation should therefore
be on the left, since subjects responded with their right hand.

These images were from the same experiment that is already
used on the web to illustrate a single-subject "advanced
event-related fMRI" analysis, and also used on the web to 
illustrate a "random effects" analysis with a SINGLE basis 
function (the canonical HRF) in both SPM2 and SNPM.

This is a good thing, though the first-level design matrices
are not EXACTLY the same, so the con*imgs are SLIGHTLY
different (this is because I actually estimated the above 
con*imgs from a single huge fixed effects analysis including
all 12 subjects, and so I collapsed the various event-types 
to reduce the the size of X and so reduce computation time).

Nonetheless, the existing single-subject fixed-effects example
dataset could be used to illustrate the "kind of" first-level
design matrix that would be used to generate these contrast images.


As well as the basic effects-of-interest F-contrast in the 2nd-level
model, it is quite interesting to test F-contrasts on individual
basis functions. You will see that each basis function explains
additional variability in some areas at p<.05 corrected. A similar
point was made in an HBM abstract from years ago (this
also used the same dataset) - see hbm-fir.pdf.

Rik Henson & Will Penny, May 2004.