Dear all, 

I'm new to CONN, and not sure what I'm doing with my results. I have 2 groups (Group_A (n=16) ; Group_B (n=14)) defined in "Setup Covariates (2nd-level)" and 2 sessions (Sess_1 ; Sess_2) defined in "Setup Conditions". 
Then I have a clinical score (Score) defined in "Setup Covariates (2nd-level)", and I created an "All Subjects" group, also defined in "Setup Covariates (2nd-level)".

I try to see what's different between the two sessions depending on my clinical score. 

1. I choose "All Subjects" , "Score" [0 1] / "Sess_1" , "Sess_2" [-1 1]

2. I choose "Group_A" , "Group_B" , "Score" [0 0 1] / "Sess_1" , "Sess_2" [-1 1]

Why do I have differents results with these two contrats ? 

Thank you for your help in advance,
Best regards,
Louis

Dear Louis,

That's an interesting question, thanks for the clear/detailed description.

Those results are different due to differences in the covariate (clinical scores) between the two groups. Analysis (1) is looking at the overall association between clinical scores and connectivity changes across all of your subjects (disregarding group membership), while analysis (2) is looking only at the within-group association between clinical scores and connectivity changes (i.e. discounting potential associations that may be explained more easily by between-group differences in clinical scores).

In general, part of that overall association observed across all subjects between clinical scores and connectivity values may be due to potential between-group differences in these measures. That may be just fine and expected (e.g. if Group_A and Group_B are defined based on severity of symptoms; or if Group_A are "controls" and Group_B are "patients" and clinical scores are only defined within patients; etc.) or it may be a confounding effect that you wish to control for (e.g. if the differences in clinical scores and/or connectivity between Group_A and Group_B were due to unintended sampling biases). Analysis (1) is what you would typically run in the former case, while analysis (2) is what you would typically run in the latter case. If you run analysis (1) in the latter case then you would be failing to correct for those unintended biases, while if you run analysis (2) in the former case then you would be unnecessarily discounting an important/meaningful part of the association between clinical scores and connectivity values. 

Hope this helps
Alfonso
Originally posted by louisr:

Dear Alfonso, 

It does help a lot !
Actually, one group is "treatment" and the other group is "placebo". So analysis (1) is what I should run, am I right ?

Another question : how can I know, in that overall association between clinical scores and connectivity changes across all of my subjects what's the effect of my treatment ? 
I want to do "Group_A" , "Group_B" , "Score" [1 -1 1] / "Sess_1" , "Sess_2" [-1 1]. Is it correct ?

Thanks a lot,
Best regards,
Louis