Hi Jeff,

I should have figured it could not be an easy question :)

I imagine that you could perhaps also do a conjunction analysis for this. If, for example, you select 'patients', 'controls', 'patients_cognitivechange','controls_cognitivechange' and enter a [1 -1 0 0] contrast (assuming here that both '*_cognitivechange' variables are mean-centered to their respective average values across each group i.e. separately centered), and also select the 'baseline' and 'follow-up' conditions as before entering a [-1 1] contrast there, that will give you those regions where the task functional connectivity change differs between the two groups. Then you use the 'display SPM' option to take these results to SPM, and define there (in addition to the existing [1 -1 0 0] contrast) another [0 0 1 0] F-contrast (to look at those regions where cognitive change scores in patients correlate with task connectivity), and then select both contrast and a 'null conjunction' analysis, that, if I am interpreting correctly, should also give what you are after (those regions that show differences in task-related functional connectivity changes in time, and where also those connectivity changes are associated with cognitive scores in patients).

That said, it is not entirely clear a priori which version of the analysis (whether importing the values as you suggest and then using the calculator to explore associations with cognitive change scores, or the conjunction analysis above) is going to be more sensitive. The main tricky issue in either version of these analyses is, when comparing the functional connectivity changes between the two groups, how to properly interpret these differences in the presence of potential differences between groups in cognitive score changes, and potential associations between cognitive score changes and functional connectivity changes (and worse yet if there are potential differences in these associations between groups). I would probably suggest to also perform the same analyses as above but now using the same centering across the two '*_cognitivechange' variables (i.e. centering both simultaneously to the same average across all subjects level). Any differences between this analysis results and the previous one probably require a closer look, because they point to either regions where there may be connectivity differences between the groups but those can be explained simply by differences in cognitiveChange scores between the groups, or regions where there may not be connectivity differences between the groups but that lack of difference is inconsistent with the differences in cognitiveChange scores observed between the groups and the expected level of association between cognitiveChange and connectivity changes.  

Hope this helps, and let me know your thoughts
Alfonso

Originally posted by Jeff Browndyke: