[FieldTrip] Time lag for granger with input freq data

Eelke Spaak eelke.spaak at donders.ru.nl
Thu Jun 5 09:47:08 CEST 2014


Dear Clara,

The Granger causality in ft_connectivityanalysis is always
frequency-domain Granger (time-domain Granger is not implemented in
FieldTrip). Frequency-domain Granger is lag-free, i.e., the Granger
causality is estimated as a function of frequency; no explicit
specification of lag is required. In some sense the frequencies
correspond to different time lags at which causality is estimated.

That said, there are two ways of estimating the cross-spectral density
(CSD) required for frequency-domain Granger computation: the
parametric and the non-parametric way. The non-parametric way is I
guess most typically used; it involves ft_freqanalysis with cfg.method
= 'mtmfft' or 'mtmconvol'. The parametric way involves first
estimating a multivariate autoregressive model (ft_mvaranalysis) and
subsequently computing the CSD from the estimated model coefficients
(ft_freqanalysis with cfg.method = 'mvar'). In the parametric way of
estimating the CSD, a time lag specification *is* required, namely
when invoking ft_mvaranalysis. If I'm not mistaken, the choice of time
lag at that step determines the effective frequency resolution of the
CSD.

Best,
Eelke


On 4 June 2014 18:03, Clara A. Scholl <clara.scholl at gmail.com> wrote:
> Dear FieldTrip Community,
>
> I have a question about ft_connectivityanalysis using method granger with
> input freq data:  how do I specify the time lag? If no time lag is
> specified, is a time lag automatically selected?  I get plausible results
> with the following inputs (only specifying channelcmb and method, data are
> freq structure) but I haven't specified a time lag…
>
>>>>>snip<<<<<
>
> cfg=[];
>
> cfg.channelcmb={'E59' 'all'};
>
> cfg.method='granger';
>
>
>
> w=ft_connectivityanalysis(cfg, TFR)
>
>
>>>>>end snip<<<<
>
> Sincerely,
> Clara
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip




More information about the fieldtrip mailing list