[Eeglablist] Lachaux' synchrony method

[arno]

Dear Jim,

Jim Kroger wrote:
> Hello, I'm trying to implement Jean-Phillipe Lachaux' method for measuring 
> synchrony across two electrodes in an EEGLAB compatible script. There are a 
> few items he does not explain, and I wonder if anybody knows about this and 
> can fill in the gaps.
>
> First, in his form of bootstrapping, he states:
>
> we generate 200 new series of variables, which have the same 
> characteristics as the original signal coming from electrode 2, except that 
> we built them to be independent of the signals coming from electrode 1. 
> These series are created by shuffling the trials within the measures of 
> electrode 2 to make new series y'(n) = y(shuffle(n)), where (i) is the 
> signal recorded at electrode 2 during trial"
>
> I don't understand how he has shuffled the data, or created his surrogate 
> data. I can imagine many ways to do it but would like to know how he did 
> it. He also mentions he mimics the characteristics as the original signal. 
> It's not clear what "characteristics" he chose, or if they are necessary 
> and sufficient for his bootstrapping method to work.
>   
He simply shuffled the trials of electrode 2 and recomputed the measure. 
This way, the signal from electrode 2 is independent of the signal of 
electrode 1. He then repeated this procedure 200 times. Note that this 
procedure might be over-conservative. The problem is that you loose the 
significance of all the effect that are consistently time-locked to the 
stimulus (since they are present in all the trials, they will be present 
in your 200 repetition even if you shuffle trials). You loose all 
"evoked" effect and only get "induced" effect. In theory, these might be 
originating from the same processes though, so this separation might be 
artificial. So I would rather use trials from electrode 2 that are not 
stimulus locked. That is just an opinion and Jean Phillipe might not 
agree with that.
> By the way, another oddity, he is using a Gabor wavelet, but in his figure, 
> states he uses a morlet wavelet. I suppose this is an error, and wouldn't 
> ultimately matter.
>   
I think there might be different types of Morlet wavelet, one of them 
using Gabor tapering.

Best,

Arno