Hi Olivier,

If you are referring to Linear Mixed Effects models unfortunately that is not possible in CONN, sorry! (as LME models go beyond the capabilities of GLM). That said, if you are referring to repeated measures models that include both within-subjects and between-subjects factors, that is perfectly fine within GLM. For example in your case, with only two timepoints (Session 1 and 2), and assuming that you do not have missing data, evaluating group-by-time interactions using a repeated measures ANOVA (e.g. as what you are doing in GLM, if I am interpreting correctly, assuming that Session1 and Session2 are defined as two conditions and you are entering a [-1 1] between-conditions contrast, and group is defined as a second-level covariate and you are entering a [-1 1] between-subjects contrast) is perfectly fine and would typically give you all the information you need (it is only when you have more timepoints, and/or would like the ability to model slopes wrt variables that also change across timepoints, and/or you have missing data to worry about, that LME models really excel)

Hope this heps

Alfonso

Originally posted by Olivier Brown:

Hello CONN experts,

 

For our latest study, we have resting-state images from two groups collected at two timepoints: Session1 and Session2. I ran a seed-to-voxel analysis comparing connectivity between groups (AllSubjects [0]; GroupA[1]; GroupB[-1]) at Session2. I also ran a comparison between time points (delta scores) between groups (AllSubjects [0]; GroupA[1]; GroupB[-1]. Session1 [-1] Session2 [1]). So, this compares the delta connectivity values from Session2 > Sesssion1 between GroupA and GroupB.

 

I was wondering whether it was possible to compute a true longitudinal model that would be mixed effects instead of simply comparing delta scores (Session2 connectivity - Session1 connectivity). Is there a way to run mixed effects in CONN, or is it limited to delta scores?

 

Thanks!

-Olivier