 |
 |
Dear Roshini, On Mon, Oct 9, 2017 at 12:42 AM, Roshini Randeniya <[log in to unmask]> wrote: > Dear Vladamir, > > I am a Masters student in Dr. Marta Garrido’s lab, using DCM for a EEG > mismatch negativity paradigm using SPM12. > > > > We have some questions about doing post hoc statistical tests on BMA > parameters from the B-matrix/A-matrix. > > > > 1. Is the more accurate way to obtain the exponential (exp) of the > parameter estimates before doing any t-tests/ANOVAs? > I think this was discussed on the list before but quite a long time ago. As far as I remember, the conclusion was that it is better to do stats on log-parameters (i.e. as they are stored in the DCM struct) because the logged values can be both positive an negative and therefore better conform to the normality assumptions of the classical tests. If your two questions are related then you should definitely only include in any stats the values that are being optimised during DCM fitting and not values like -4 or -32 that are only there to indicate that the connection is turned off. 2. Is there a reason for the absence of a connection in the A-matrix > being signified with -4 value (instead of -32)? ….so should we make this > value zero after exponentiating the estimates for t-tests? > > There was a reason for that change that had to do with Bayesian Model Reduction that couldn't handle exp(-32) as the prior. As I said, I don't think it makes sense to compare these values which are only there for disabled connections with connections thate were fitted to data. Best, Vladimir > > > > > I know you must be very busy so I’m sorry to trouble you with this. > > I didn’t find anything on the mailing list, and didn’t get a reply when I > posted this on the list. > > > > Thank you for your time! > > > > Regards, > > Roshini > > > > --- > > *Roshini Randeniya* > > MPhil. Student > > Computational Cognitive Neuroscience Laboratory > <https://qbi.uq.edu.au/garridogroup> > > Queensland Brain Institute > > The University of Queensland > > Brisbane, QLD, 4072, Australia > > >