Hi Zakia,

Warning messages during second-level analysis definitions typically indicate an incorrect second-level model/analysis. In this case I imagine that the issue is perhaps that the variables Age and Scores may be defined for all subjects, so the "Age + group1 + Scores" model is missing a constant term. Assuming that the intention of this model was to analyze the association between functional connectivity strength and "Scores" within group1 subjects while correcting for age, the correct form for that model would be:

"group1, Age*group1, Scores*group1" (with a [0 0 1] contrast)

The "Age*group1" variable would be the interaction between the Age and group1 variables (i.e. a variable that contains age values for subjects within group1 and 0 values for everybody else). You can create that automatically in CONN in the Setup.Covariates (2nd-level) tab, by first selecting "Age" and "group1" and then clicking on 'Covariate tools -> Create interaction of selected covariates' (and similarly to create the Scores*group1 variable)

Similarly, if you want to analyze the effect of Scores in group2, that would use a model of the form:

"group2, Age*group2, Scores*group2" (with a [0 0 1] contrast)

To double-check make sure that the N=??? value reported in the lower-right corner when defining these two analyses matches their expected values (i.e. the number of subjects in group1 and group2, respectively)

Hope this helps

Alfonso

Originally posted by Zakia Ben Youss:

Hi,

I have questions regarding the 2nd level analysis in SBC. 

1. Why does the Age group1 Scores [0 0 1] gives different results from AllSubj Age group1 scores [0 0 0 1]. I undestand that the warning is due to the age being overdefined since my subject list contains more groups than just group1 and group2. I have the results for both, and [0 0 0 1] gives me more significant clusters with p-unc 0.005. Which contrast should I use?

2. To check the effect of Scores for each group, I am using Allsubj Age group1 Scores [0 0 0 1], then the same for group2. 
 I noticed that for the same seed, I was getting the same cluster for both analysis (same MNI coordinates) 
However, when camparing the REX values for the group1 Scores effect and group2 Scores effect, I noticed that it is done for all the subjects and the values are different between the REX files when comparing a each subject. Why is that?