Dear Christopher, I don't see a problem with only including a subset of the sources in your DCM, the ones which make sense theoretically for your question. A good model should be as simple as possible (but not simpler). Best, Vladimir On Wed, Jun 28, 2017 at 11:03 PM, Christopher Brown <[log in to unmask]> wrote: > Dear DCM experts > > I'm developing an analysis pipeline for ERP data involving DCM. I'm first > identifying sources (using MSP) and using those locations as inputs to DCM. > However, many of the sources that come up are not of interest to my > hypotheses; furthermore in the MSP analysis it's not uncommon for more than > 10 source clusters to evident (e.g. ~15 in extreme cases). In either > scenario, I don't want to construct DCM models with 15 connected sources, > nor I believe can DCM practically cope with this. Rather, I would prefer to > work on maybe 5 or so sources of interest. > > However, if I only model 5 sources in DCM despite knowing that the > previous source model explains the EEG data with a far larger number of > sources, surely this would invalidate the DCM model? If so, how can I model > a small number of connections while maintaining validity? I heard that > "hidden sources" can be included in a DCM but my understanding is that > these are sources that do not directly contribute to the scalp recordings > (only indirectly) and if they are included in the DCM model but not > connected, they do not contribute to the model at all. So this doesn't > appear to be a solution. Does anyone have any thoughts on how best to > manage a situation with a large number of sources? > > With gratitude > Chris >