 |
 |
Yes, if you have epoched data and use the ERP option, the data will automatically be averaged. The convergence times depend on the data and the model and can be quite long for complicated cases. Seven steps is a small number and might indicate a problem (not necessarily though). You should look at the predicted vs. observed data and see that the predicted data captures the features you are interested in. Typically there would be at least 20 steps when everything goes well. Best, Vladimir On Tue, Oct 8, 2013 at 11:15 PM, Yingying Wang <[log in to unmask]>wrote: > Thanks a lot, Valdimir. > > I didn't average my trials. For ERP, do I have to manually averaged > trials first? I thought the option "ERP" would take care of that. If not, > by default, does DCM only use the first trial of each conditon or DCM > averages trials according to the condition label? > > Thanks again. > Ying-Ying > > ------------------------------ > Date: Tue, 8 Oct 2013 19:58:48 +0100 > > From: [log in to unmask] > Subject: Re: [SPM] post-hoc type DCM > To: [log in to unmask] > > Yes, you could do that, but in most cases it would not make sense to use > average waveforms as input. You should use an epoched dataset before > averaging. > > Best, > > Vladimir > > > On Tue, Oct 8, 2013 at 5:01 PM, Yingying Wang <[log in to unmask]>wrote: > > Dear Vladimir, > > Thanks for your clarification. One more question, will it be meaningful > to also look into the induced response using IND option for > frequency-specific informaiton with "LFP" option in the 'electromagnetic > model' part? > Thanks for you time and help. > Best, > Ying-Ying > > ------------------------------ > Date: Tue, 8 Oct 2013 10:30:48 +0100 > From: [log in to unmask] > Subject: Re: [SPM] post-hoc type DCM > To: [log in to unmask] > > > Dear Ying-Ying, > > It is NOT possible to model difference waveforms with any of the DCM > variants because DCM is a biophysical model for how the data are generated > whereas a difference waveform was not generated by any biophysical process. > You should apply DCM to the original conditions (as done for example in the > MMN example available in the manual and explained in our video tutorial ( > http://www.fil.ion.ucl.ac.uk/spm/course/video/#MEEG_Demo_DCM). > > Which DCM to use depends on your data and question, but since you mention > difference waveforms, it's likely that you have evoked responses and then > you should use DCM ERP (ERP ERP in the GUI). There is no problem with using > extracted time series but you should set the channel type of those 5 > channels to 'LFP' and select 'LFP' option when you get to the > 'electromagnetic model' part. > > Finally not sure why you mentioned post-hoc in your subject, but note that > there is such a thing as post-hoc inference for DCMs but it has not been > validated for MEEG DCMs and at this point is unlikely to yield valid > results. We are working on making it available for MEEG in the future. For > now just use the 'invert' button. > > Best, > > Vladimir > > > > On Mon, Oct 7, 2013 at 10:39 PM, Ying-Ying Wang <[log in to unmask]>wrote: > > MEG data time series extracted from VOIs (5 nodes) > This time series is a subtraction of two task conditions. (differential > responses) > 5 nodes 2401 data points and 15 tials > > Interested in modulation of effective connectivity in three models. > > Q: > (1) do I use same proirs for all three models? > (2) do I choose SSR ERP or ERP LFP or ERP ERP or IND? > > Thanks. > > > >