Hello Lukas:
Since you are currently performing a "within subjects analysis", you can
use program 3dDeconvolve to perform the ANOVA. See Section 1.4.5, Multiple
Linear Regression Approach to ANOVA, contained in file 3dDeconvolve.ps.
Try working through the examples in that section. Note how the GLT matrices
can be used to test for main effects and interactions.
(I hope you have an older version of AFNI. Someone, who's name I won't mention,
but his initials are RWC, introduced an error into my matrix arithmetic
library, which may cause the current version of program 3dDeconvolve to crash
when the polort -1 option is used.)
In any case, when working with the actual fMRI time series data, do NOT use
polort -1. Another difference between the examples in Section 1.4.5 and the
intended application is that, if you are using multiple time lags to model
the IRF, you will need to extend the GLT matrices accordingly. That is,
multiple (IRF) parameters correspond to the different levels of the different
factors.
Since this is a within subject analysis, the fixed effects model is
appropriate. That is, the different levels of the 2 factors are of inherent
interest. If, in the future, you combine results across subjects, then
a mixed effects model would be appropriate, where the Subject factor is
treated as a random factor. The reason is that the different subjects
constitute a random sample from the general population, and, as such, are not
of inherent interest. This would probably require using program 3dANOVA3.
Doug Ward
