Hi Alfonso, 

Thank you for your response. I am using 4 clinical scores of behaviour (labelled C1, C2, C3 and C4) as multiple predictor variables and FC is the outcome variable. However, I am a bit confused because I am looking at how each of these clinical scores is associated with FC in multiple source ROIs. So because I have more than one outcome variable (i.e., multiple ROIs) I thought it was a mutlivariate regression?

The way that I set this up is based on the fact that I initially had two groups (ASD and controls), and this was a within group analysis on ASD subjects only: I chose 'multivariate regression' in the 1st level analysis tab, and then when viewing in second level results I selected the following variables : ASD, C1, C2, C3, C4 and set up the contrast for example as [0, 1, 0, 0, 0] - to look at the association between FC and C1, (while controlling for the association between FC and all of the other clinical variables?). As I move through the source ROIs on the right, I can see the association between the selected predictor variable and connectivity between that source ROI and all target ROIS (brodmann areas). 

So I guess my question was if there was any way to extract the residuals from this model in order to examine the validity of the assumptions. Is there reason to think that the homoscedasticity assumption does not apply here? Would I want the residual errors of observed - predicted Fisher-Z values to be homogenous across all clinical variables?

Thank you so much for your help, and please let me know if I have done something incorrectly. 

Jenna