Hi David,

If I am interpreting correctly, the most likely reason for that discrepancy would be that in CONN you need to explicitly center your covariates at the desired level (e.g. mean-center your age covariate if you wish to estimate the connectivity strength at the average age-level of your sample), since individual effects are always estimated at the zero-level of your covariates. 

To do this, simply go to Setup.CovariatesSecondLevel, create a new set of centered covariates there. For example create a covariate named 'ageCentered' and enter in the values field "age - mean(age)" (without the quotes, and changing "age" to the name of your original age covariate). This new covariate will contain the age values for each subject after subtracting the average age across all of the subjects in your sample. After doing the same for any other covariates, you can then go back to the second-level analysis tab and select there "Group1", "Group2", "ageCentered", "sexCentered", and "variableCentered", and enter a between-subject contrast:

 a) [1 0 0 0 0] to look at the simple main effects within the Group1 subjects
 b) [0 1 0 0 0] to look at the simple main effects within the Group2 subjects
 c) [1 0 0 0 0; 0 1 0 0 0] to look at simple main effects in any of the two groups
 d) [1 -1 0 0 0] to look at between-group differences

A couple of additional notes:

 1) it is only for examples (a)-(c) above that one would need to center the covariates at the desired level at which one wishes to interrogate the main/average connectivity effects. In example (d) above (between-group differences) it does not make a difference whether you center the covariates or not since the individual within-group effects will be compared at the same level of those covariates (no matter what those levels are) 

 2) when considering additional covariate effects in a two-groups design one typically wants to additionally consider whether a group by covariate interaction is meaningful or not. For example, in the case of gender, a group by gender interaction would ask the question of whether the difference in connectivity between group1 and group2 is different for males vs. females. All of the contrasts above assume no interactions. If you want to consider possible interaction effects in CONN you would do that also explicitly by defining new group-specific covariates (e.g. create new "age_Group1" covariate and enter in the values field "age.*Group1" to have age values for Group1 subjects and 0's for Group2 subjects) and then entering those group-specific covariates in your second-level models. 

Hope this helps and let me know if you would like me to further clarify any of the above
Alfonso

Originally posted by David Coynel: