Hi Paul,
>One last confusion. How would I model multiple subjects? Do I include them
all
>in a the design matrix and contrast all conditions and all subjects? This
>would seem to allow large activations in one individual to scew the data.
>Analyses I have done in the past used either random effects or conjunctions
>for the group data.
Ok, that's a tough one. If you use a fixed effects model, you may find
activations
due to a single subject. To do this type of analysis using a random
effects model may be
trickier, though. Let me start by saying I've never actually done this but
I think there might be
a method. Essentially you'd want to create a random effects analysis of
the main contrast and
then mask it with the individual simple contrasts to avoid a main effect
due to a subset
of conditions (eg, just from condition A).
So you could start by generating three con_* images per subject
corresponding to [1 0 0],
[0 1 0], and [0 0 1] and create a random effect SPM{t} for each of these
contrasts.
Then, because you won't be able to mask your main efffect with these
contrast images
in the random effects analysis, you could create binary masks from these
images using ImCalc.
If you chose each of the group contrast spmT_* images in turn
(corresponding to the simple
contrasts), you could threshold the map at something like p<0.05
uncorrected by using
ImCalc: "i1 > X" where X is the actual t value that corresponds to the
threshold you desire. Then
again using ImCalc you could combine these masks into one: "i1 .* i2 .*
i3". This should yield a binary
mask with 1s in areas which were active, at the threshold chosen earlier,
in all three conditions
and zeros elsewhere. This can now be used to mask your main effect.
Now, for each subject, you can used a fixed effet analysis to generate the
con_* images
for the main contrast [1 1 1] and the enter them into a one sample t-test
as usual. When
asked if you want to explicitly mask images, choose 'yes' and enter the
combined
binary mask from above.
I think this will effectively give you an analysis of the common areas of
activation across
your three conditions and in your group of subjects where the results are
1) not due to any
individual subjects (because of the RFX) and 2) not due to individual
conditions (because of
the masking).
Perhaps someone else has a more clever method for doing this?
I guess it would be much easier to have separate baselines in future
studies and simply use
conjunctions...oh well. I'm in exactly the same boat because I hadn't
thought of this before
running a study either!
Best of luck,
Joe
- Joe
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|