This is basically a continuation of this thread:
[
afni.nimh.nih.gov]
But my questions are more general. As far as I understand the only reasonable way to perform a group analysis of 3dNLfim statistical outputs is to do a multivariate test with 3dMVM. It wouldn't make sense to put all the parameters of a nonlinear equation (a^b-c or something) in an ANOVA model where they would be incorrectly modeled (as y=ax1+bx2+cx3), right?
Here is a picture of the same group dataset analyzed with 3dMVM:
the left is done with cspline (7 parameters), the middle one with SPMG3 (3 parmeters) the right one with 4-half cosine model (7parameters).
I am not sure if these maps (all unthresholded F-values with maximum of colorbar set to F>40) are really comparable.
Aren't the F-values dependent on the number of model parameters as well as the number of subjects?
Is there some way to normalize F-maps so that they are independent of these degrees of freedom?