Dear DCM Experts, I have a problem regarding the constant confound that is supposed to estimate the mean. The result seems to depend on the amplitude of the constant part. The mean of the signal extracted using the VOI function of SPM has a mean of somewhere around 100 to 700. Now if I put in only ones for the confound DCM does not seem to be able to deal with the mean properly since it the estimated signal slowly ascends towards the real signal (takes about half the task) and the rest does not look anything alike the real signal. If I multiply the confound by 100 this problem is gone. I haven't tried values in between but I guess it might come up with some other solutions. I understand that DCM probably somehow gets caught in some local minimum which can happen but is still strange considering the high prior variance that is put in for confounds (1E8 if I'm not mistaken). Anyway, I'm not quite sure how to deal with it. I thought maybe I'll just multiply the constant prior with the corresponding mean of each region (not possible without modifications). Is this a valid way to go about it? Furthermore, would it be sensible to also normalize the cosine (and motion) confounds using the variance of the region? Thank you for your help Christian