Hello Rutger:
Yes, Sally and Andy are correct. By entering Subject as a Random Factor,
the ANOVA DOES take into account the fact that the same measurements were
repeated in a single subject. For example, a two-factor ANOVA, with 2 levels
for the fixed factor A, and Subject as the Random Factor B, is equivalent to
a paired t-test.
The second study that you described sounds like a 3 factor crossed-nested
ANOVA, with random factor C (Subject) nested with fixed factor A (Group).
See the documentation in file 3dANOVA.ps for program 3dANOVA3.
However, let me suggest an alternative approach to the analysis. The major
difficulty with comparing fMRI measurement results across subjects is that
people's brains are different. Now, in order to compensate for the variation
in neural anatomy, we "normalize" the brains by TLRC transformation followed
by Gaussian blurring. However, this is, at best, an imperfect solution.
Because of the difficulties inherent in the across-subject analysis, I prefer
to delay the across-subject comparisons as late as possible in the data
analysis stream. That is, I think it is better to perform as much of the
analysis within-subject (using the ORIG data) as possible.
I don't know the details of your experiments. However, I believe that it is
possible to extract the Drug vs. Baseline, and Placebo vs. Baseline, response
for each individual subject before it is necessary to compare results across
subjects. That is, you can carefully concatenate runs (within-subject only!),
and then use the "-glt" option of program 3dDeconvolve to calculate the
relative differences in response for each subject. Then, only this final
result need be TLRC'd and blurred, before using, say, a t-test to compare
the groups. See the documentation in file 3dDeconvolve.ps for more details.
The same reasoning applies to your first question about comparing medication
regimes for different subjects. First, compare the medication regimes within
the individual subjects, then combine results across subjects.
Opinions expressed herein are only those of the author.
Doug Ward