[FieldTrip] Fwd: A question about connectivity values

[Schoffelen, J.M. (Jan Mathijs)]

Hi Masoud et al.,

Apologies: I was inaccurate in my previous e-mail in explaining the GC coefficient. It’s the negative natural logarithm of the residual variance ratio, where the model with the larger number of parameters (i.e. the one which predicts a signal based on the past of the other signals as well) is in the numerator.

BW,
JM


Begin forwarded message:

From: Jan Mathijs Schoffelen <janmathijs.schoffelen at donders.ru.nl<mailto:janmathijs.schoffelen at donders.ru.nl>>
Subject: Re: [FieldTrip] A question about connectivity values
Date: 4 January 2023 at 13:25:18 CET
To: FieldTrip discussion list <fieldtrip at science.ru.nl<mailto:fieldtrip at science.ru.nl>>

Hi Masoud,

Theoretically, the Granger causality coefficient is a natural logarithm of a ratio of residual variances, where this ratio is typically >1, because the model implicitly used for the numerator has more parameters, so typically a lower residual variance. In other words, GC > 0 but can become larger than 1.

If the input data into the algorithm are numerically not well behaved, I can imagine that the output GC will be unexpected. In your case, you estimate a series of models consisting of 62x62x6 (i.e. >18.000) parameters (for each time window). If the number of samples in your data is insufficient and/or there’s linear dependence between channels, the estimated parameters, and by consequence the spectral quantities, and the GC will not make sense.

Best wishes and good luck,

Jan-Mathijs







On 4 Jan 2023, at 09:26, masoud via fieldtrip <fieldtrip at science.ru.nl<mailto:fieldtrip at science.ru.nl>> wrote:

Hello dear fieldtrip community
I have been using fieldtrip for quite some time and now I am dealing with dynamic connectivity, the preprocessing of the data which I am using was done in fieldtrip as well as MVAR analysis, this is the code for calculating dynamic connectivity:
   cfg.method = 'biosig';
   cfg.t_ftimwin = 5;
   cfg.toi = toi;
   cfg.order = 6;
   cfg.output = 'parameters';
   mvardata = ft_mvaranalysis(cfg, data);
   cfg = [];
   cfg.method = 'mvar';
   cfg.output = 'fourier';
   freq = ft_freqanalysis(cfg, mvardata);
   cfg = [];
   cfg.method = 'granger';
   stat = ft_connectivityanalysis(cfg, freq);
so the output of connectivity (stat) is a 4-d matrix(channel, channel, frequency, windows) which for my data is (62, 62, 126, 12)
the values of the channel x channel are mostly negative like -13.2650 , -5.7060 or between -1 and 1.
as my understanding the output should be between 0 and 1. So where am I doing it wrong? And if I change the method from 'granger' to 'pdc' or 'dtf' what should be the range of output?
thank you so much in advance
best regards
Masoud
_______________________________________________
fieldtrip mailing list
https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
https://urldefense.com/v3/__https://doi.org/10.1371/journal.pcbi.1002202__;!!HJOPV4FYYWzcc1jazlU!_A7iaW2bTmQVqTwzWp2CiqVYS7OUDPsHcVqbD8h4gfg9PVZjhzf7b4hCe3x8u5eqEe-Y7lb48cayVioAe8K0M9KMxEvk9Oz5pNDoEA$


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20230104/d917a1fb/attachment.htm>