Dear HanGue,

DCM is deterministic so if you run the same model on the same data you should get the same results. I recently encountered a bug which appeared when using the BEM model for EEG in combination with optimising dipole locations. In this case the dipoles could get too close to the inner skull boundary and then the lead-field became invalid and that led to wrong and non-deterministic results. I suggest that you try to use the IMG option and/or the spherical head model where this cannot happen and if that still doesn't help, you can also try updating to the latest in-house version of SPM12 and trying there (write to [log in to unmask] to get a link).



On Tue, Feb 28, 2017 at 10:05 PM, Jo, Han-Gue <[log in to unmask]> wrote:
Thank you for your email!
My interest is rather the DCM in itself.
Is DCM supposed to come up with the same estimated parameters, if the data and model space are the same?
I ran DCM several times with the same data and each time I got different result.

If someone in the SPM list has ran DCM more than once, please share your experience. Any comments would be very helpful.

On 19 Feb 2017, at 23:58, Marta Garrido <[log in to unmask]> wrote:

Dear HanGue,
You can get slightly different results at the group level if you compute the statistics with RFX, as this is based on a sampling procedure, and hence it’s not deterministic. Having said that, this most likely occurs when there’s only very week evidence in favour of one model over the remaining and thus the winning model may fluctuate with the sampling run. As for different results with DCM itself, this can happen if you used different computers.
I CC the SPM list in case others have encountered the same issue and would like to comment.
Hope this helps,
Dr. Marta Garrido, PhD
Queensland Brain Institute
School of Mathematics & Physics
The University of Queensland, Australia
From: "Jo, Han-Gue" <[log in to unmask]>
Date: Saturday, 18 February 2017 at 8:41 PM
To: Marta Garrido <[log in to unmask]>
Subject: EEG DCM replication
Dear Marta I. Garrido,
I read some of your papers published in 2007 and later. I am very impressed with EEG DCM and its applicability for studying the basis of neuronal process. I read SPM manual and other papers about EEG DCM and finally applied it to my own EEG data.
However, I encountered a replication problem. I ran EEG DCM twice with the same setup and data. But the outcome parameters were completely different, which eventually influences the log-evidence result. I wonder if this is only my problem. Do you have an idea about this issue? Any comment or advice would be very helpful for us.
Many thanks for reading this email and I look forward to hearing from you.
Kind regards from Aachen, Germany,