Dear Philipp
The function spm_dcm_reduce will give you the posterior parameter estimates that you would have got, if rE had been the prior expectations and rC had been the prior covariance of the parameters (under Gaussian assumptions). So in your case, you'd supply this with the posterior expectations of the group level estimates: BMA.Ep and BMA.Cp respectively. However, this isn't necessary - you could just plot the PEB.Ep parameters. Any parameters pruned by Bayesian model reduction will probably be close to zero in PEB.Ep - so it won't make much difference.
Best
Peter
On 09/04/2020, 10:59, "Philipp Kuhnke" <[log in to unmask]> wrote:
Dear Peter,
thanks a lot for your rapid response! What would I use for arguments 'rE' and 'rC' in spm_dcm_reduce(DCM,rE,rC)?
All the best,
Philipp
----- Ursprüngliche Mail -----
Von: "Zeidman, Peter" <[log in to unmask]>
An: "Philipp Kuhnke" <[log in to unmask]>, "SPM" <[log in to unmask]>
Gesendet: Donnerstag, 9. April 2020 11:16:00
Betreff: Re: DCM subject-wise parameters
Dear Philipp
Sorry no that's not possible with the current tools. Subject-wise parameters are returned as the second output from spm_dcm_peb:
[PEB, rGCM] = spm_dcm_peb(GCM);
The function spm_dcm_peb_bmc is then used to compare the full PEB model against reduced ones. In principle, it would be possible to take the parameters from the BMA and use them as empirical priors to update each subject's DCM (you would use the function spm_dcm_reduce for this). I've not tried that :-)
Best
Peter
On 09/04/2020, 09:59, "Philipp Kuhnke" <[log in to unmask]> wrote:
Dear Dr. Zeidman and the SPM list,
I have run a DCM group-analysis using PEB and BMR, following your tutorial papers in NeuroImage. That is, after estimating the full models in each subject, I executed:
[PEB, RCM] = spm_dcm_peb(GCM,M,{'B'}); % build PEB model (GCM: all subject-wise full-model DCMs)
[BMA, BMR] = spm_dcm_peb_bmc(PEB); % Automatic search
I know that the BMA now contains the main group results: the parameter estimates in 'Ep' and the covariance in 'Cp'. Is it possible to also get "corresponding" subject-wise values, and if so where do I find them? For instance, I'd like to correlate a certain modulatory parameter, e.g. B(3,1,2) in each subject, with a certain behavioral variable.
Thanks a lot in advance,
Philipp
-------------
Philipp Kuhnke, M.Sc.
Ph.D. Student
Lise Meitner Research Group 'Cognition and Plasticity'
Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany
|