Quick background: I've obtained some significant results after running randomise on GroupICA data for resting state data, using FSLnets, and just want to confirm my understanding of their implications.

I've modeled in my GLM two groups of subjects (controls and AD patients) with various confound regressors and a behavioral covariate of interest that shows a significant score difference between the groups.

I now have two significant results:

1) Significant result for a "negative behavioral effect" in my control group (similar to the negative age effect modeled in the GLM wiki) between network x & y

 - interpretation: increasing scores for the behavioral covariate in controls is associated with decreased connectivity between network x & y

2) Significant interaction effect for slope of controls < slope of AD patients in network x & y

 - interpretation in consideration of finding #1: behavioral score differential between controls and AD patients is associated with increased connectivity between networks x & y in AD patients, relative to controls.

Also, my understanding is that, for cases where significant interaction results that have no corresponding significant corrp results in other other contrasts, the necessary step is to examine the sign of the t-stats in those corresponding contrasts, in order to interpret that significant interaction. Is that correct?

I would very much appreciate it if someone could let me know if I'm on the right track.