Dear Feng

I hope you don’t mind me CC’ing the SPM mailing so others can benefit. Yes, you can do this using the PEB framework – you can treat the design matrix of the PEB model as you would a normal design matrix with a second level GLM analysis.

As an example, I will assume this is a task-based experiment, where the effect of drug is a modulatory input in your DCMs, and that you have 2 DCMs per subject (one per time point). Then, you would specify a PEB model on the ‘B’ matrix parameters of your DCM which relate to drug. You would include a regressor in the PEB design matrix representing time. This will tell you the effect of time on the response to drug for each connection. The PEB will also have a regressor which represents the common effects across both time points (a column of 1s in the design matrix). This will tell you the effect of drug averaged across time points (if the time regressor is coded with 1s and -1s) or the effect of drug at the first time point (if the time regressor is coded with 1s and 0s).

Do let me know if anything is unclear.

Best,

Peter

From: Zhou Feng [mailto:[log in to unmask]]
Sent: 13 January 2017 07:32
To: Zeidman, Peter <[log in to unmask]>
Subject: Questions regarding DCM analyses and PEB framework

Hi Dr. Peter Zeidman,

I have questions regarding DCM analyses.

I have a 2 (Time) by 2 (drug vs. plc) repeated measures ANOVA design and I set up 2 GLMs per each subject instead of 1 GLM including 2 sessions.

I'm wondering if  Parametric Empirical Bayes (PEB) framework can estimate the repeated measures ANOVA model over DCM parameters.

As far as I know the alternative way is doing Bayesian Model Selection in each group and each time separately with Bayesian Model Averaging (BMA) turned on and then take the estimated connection strengths from the BMA and compare using classical repeated measures ANOVA. What if I find different winning models in each group or time?

Any information would be greatly appreciated. Thanks in advance!

Best regards,

Feng