Dear Csaba
Sorry for the delay in replying. Firstly, regarding whether post-hoc DCM can give models which are unrealistic. The post-hoc algorithm operates on the simplifying assumption that you assign equal prior probability to all nested models of the full model you start with. If that's not the case - if you have strong hypotheses for certain model structures - you have a couple options. You can run the post_hoc analysis with the families option (see spm_dcm_post_hoc.m). This groups the models during the search according to your hypotheses. Alternatively, you could not use the spm_dcm_post_hoc.m, and just do a regular model comparison instead with a small number of models.
Regarding whether you can have B-matrix parameters acting on A-matrix connections which are fixed at zero. Yes you can. It just means there's no evidence for effective connectivity unless there's a modulatory influence present.
Best
Peter
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Csaba Aranyi
Sent: 30 November 2016 14:29
To: [log in to unmask]
Subject: [SPM] Post hoc DCM optimization
Dear SPM Developers / Users!
I estimated a model of the Attention to Motion dataset (model: 3 nodes, 3 external inputs, deterministic, one-state, bilinear DCM) to try out the post hoc optimization method for DCM. Is it possible to get invalid models out of the optimization results? And although the results seem mathematically correct, how can a model like in my example be interpreted?
I attached my example as an image. Here I estimated a full model of this network (A), and the results are the parameter estimates in (B). I tried to optimize this model with post hoc optimization option in DCM, and got the results in (C). Some connections in the A matrix are zeroed out (or resetted to prior values) like A(3:1), while some inputs still modulate this connection. Can I consider the optimized model invalid, as it seems like that some connections which are modulated should not be present? Or can models like this be interpreted somehow?
Best regards,
Csaba Aranyi
|