Dear DCM gurus,
I have run a PEB model for a group of healthy controls (HC), a group of schizophrenia patients (SZ), and then a PEB of these two PEB’s to get group average (HC + SZ) and group difference (SZ-HC).
This is our analysis workflow:
1. (first level) estimate DCM for each subject
2. (second level) PEB optimizing A field for HC and SZ separately
3. (second level) PEB optimizing B field for HC and SZ separately
4. (third level) PEB optimizing A field of the HC PEB and SZ PEB from Step 2
5. (third level) PEB optimizing B field of the HC PEB and SZ PEB from Step 3
I was expecting the group average (HC + SZ) estimates from Steps 4/5 to be somewhere in between the HC and SZ values from Steps 2/3. However, almost all of them are not (i.e., either smaller or larger than BOTH of the HC and SZ estimates). For example, one parameter estimate is 0.42 for HC and 0.63 for SZ (from Step 2), but the group average (Step 4) is 0.27.
This does not make any intuitive sense to me. Has anyone experienced this phenomenon, and, if so, could you provide any explanation or references?
Many thanks!
Ivy F. Tso, Ph.D., L.P.
Assistant Professor of Psychiatry
Adjunct Assistant Professor of Psychology
Clinical and Training Director,
Program for Risk Evaluation and Prevention (PREP)
University of Michigan
2553 Rachel Upjohn Building,
4250 Plymouth Road, Ann Arbor, MI 48109
Tel: (734)232-0373
Fax: (734)936-7868
Email: [log in to unmask]
Website: http://sites.lsa.umich.edu/ivytso-lab/
|