I ran a factor analysis in VBM to look at effects of an intervention (3 groups) in a pre-post scan design. There is no effect of the intervention or any interactions in this factorial model. If I make a change image (scan 2 subtract scan 1), model my groups, and include my raw covariate of interest, plus some demeaned EVs I am not interested in, there is a relationship between change in grey matter and that score that differs according to group.
Note, the second model I refer to follows this FSL example, except I have not demeaned my clinical score http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Two_Groups_with_continuous_covariate_interaction
My question is about whether I can trust this second analysis. I would expect the factor analysis interactions would give some areas that changed following the intervention, and there might be some replication in the second analysis, but this was not what I have. Can you explain?