[FieldTrip] granger causality
tobias.staudigl at uni-konstanz.de
Mon Jun 13 15:02:55 CEST 2011
Thanks a lot for your answer, Jan-Mathijs ,
what "other confounding quantity" were you thinking of concerning the
shuffling across conditions?
There is another question i have:
Is there a way to estimate the order of the mvar-model in fieldtrip, or
would one do it as implemented in the bsmart toolbox?
The bsmart toolbox uses the Akaike Information Criterion to find the
right order parameter. However, this criterion sometimes seems to have
trouble finding a reasonable model-order.
Does anybody know a toolbox that uses other criterions, e.g., Bayesian
And, what is a "reasonable range of the model-order"? Is there any
convention or standard?
E.g., I m thinking of something like: "This order is too small/big to be
meaningful in this particular frequency band / for this particular
Any advice, ideas or literature suggestions welcome!
Thank you very much,
Am 10.06.2011 13:00, schrieb jan-mathijs schoffelen:
> Hi Tobias,
>> Q 1:
>> If I understood correctly, the grangerspectrum
>> (granger.grangerspctrm) gives me values for the granger-causality in
>> a matrix [nChannel, nChannel, Freqs].
>> Does [1,2,:] give me the granger-causality of Channel1 predicting
>> Channel2, or the other way round?
> The FieldTrip convention is indeed (1,2,:) relates to 1->2
>> The plotting function gives me an error:
>> ft_connectivityplot(cfg, granger)
>> ??? Error using ==> seloverdim at 36
>> cannot select over multiple dimensions at the same time
>> Error in ==> ft_selectdata at 528
>> if selectchan, data = seloverdim(data, 'chan', selchan, fb); end
>> Error in ==> ft_connectivityplot at 79
>> data = ft_selectdata(data, 'channel', cfg.channel);
> Please update your fieldtrip version. There is a fix (dated 22-04)
> that should take care of this
>> How do i statistically validate the measures i get fromthe
>> Is there some recommended statistical test?
>> Is it correct to do bootstrapping / permutation testing?
>> What I could do on my data is shuffle the trials across conditions,
>> compute granger causality on the shuffeld data, and compair this
>> distribution with the value i get from the data.
> This is a very relevant issue. From what you describe it seems as if
> you want to compare across conditions. The shuffling across conditions
> is what we do as well, although it remains open to debate how
> meaningful it is to compare granger causality indices across
> conditions. And, also, if you are able to reject your null-hypothesis,
> whether your decision to reject it is actually due to the actual
> granger causality being different, or by some other confounding quantity.
> Best wishes,
>> Maybe, somebody is experienced in using granger and could help me out!
>> Any advice appreciated!
>> Thanks a lot!
>> I used the following ft_functions according to the online tutorial:
>> %% MVAR
>> cfg = ;
>> cfg.order = 2;
>> cfg.toolbox = 'bsmart';
>> mdata = ft_mvaranalysis(cfg, data);
>> %% transfer function
>> cfg = ;
>> cfg.method = 'mvar';
>> mfreq = ft_freqanalysis(cfg, mdata);
>> %% granger
>> cfg = ;
>> cfg.method = 'granger';
>> granger = ft_connectivityanalysis(cfg, mfreq);
>> cfg = ;
>> cfg.zparam = 'grangerspctrm';
>> %cfg.channel = 'all';
>> ft_connectivityplot(cfg, granger);
>> Tobias Staudigl
>> Fachbereich Psychologie - ZPR
>> Postfach ZPR
>> 78457 Konstanz
>> ZPR, Haus 12
>> Tel.: +49 (0)7531 / 88 - 5703
>> fieldtrip mailing list
>> fieldtrip at donders.ru.nl
> Dr. J.M. (Jan-Mathijs) Schoffelen
> Donders Institute for Brain, Cognition and Behaviour,
> Centre for Cognitive Neuroimaging,
> Radboud University Nijmegen, The Netherlands
> J.Schoffelen at donders.ru.nl
> Telephone: 0031-24-3614793
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
Fachbereich Psychologie - ZPR
ZPR, Haus 12
Tel.: +49 (0)7531 / 88 - 5703
More information about the fieldtrip