[FieldTrip] WPLI statistic, permutation like test??
Matteo Demuru
suforraxi at gmail.com
Wed Jun 29 10:01:15 CEST 2011
Hi Micheal,
It seems to me that if you shuffle the trials and then compute the WPLI, the
result is the same as if you do not shuffle.
For example I have tried compute the WPLI on my trials and then switch
trial{1} with trial{2} and the obtained WPLI is the same.
Matteo
On Tue, Jun 28, 2011 at 8:34 PM, Michael Wibral <michael.wibral at web.de>wrote:
> Hi Matteo,
>
> I am not an expert on the WPLI measures, but to me it seems that in doing
>
>
> "2) Randomly permute the ch2 time series"
>
> you're destroying a lot of ch2's properties (ie.g. it's spectrum will get a
> lot of high ferquencies this way) and this will typically lead to false
> positives. This is why permutation tests for connectivity measures typically
> shuffle trials (i.e. permute data in a very controlled way, keeping the
> intrinsic structure of the data).
>
> Michael
>
>
> ------------------------------
> *Von:* "Matteo Demuru" <suforraxi at gmail.com>
> *Gesendet:* Jun 28, 2011 2:46:28 PM
> *An:* fieldtrip at donders.ru.nl
> *Betreff:* [FieldTrip] WPLI statistic, permutation like test??
>
>
> Hi,
>
> I have a couple of questions about using the WPLI index to assess the phase
> on my MEG data.
>
> The experiment consists of recordings during a mental calculation task: I
> have 30 sec in which each subject performed continuously an arithmetic
> operation.
>
> It seems to me that WPLI index required more than one trial in order to be
> computed. Am I right? (Is this necessary in order to reduce volume
> conduction problems?)
> I could divide my 30 sec in 5 sec-trials to create my trials, but I was
> wandering if this could be a misuse of the WPLI, i.e. WPLI is not
> appropriate for my experiment.
>
> I am also interested in assessing the significance of WPLI index, I would
> like to gauge the significance per se of my WPLI values.
> The idea is to calculate the WPLI distribution under the null hypothesis
> (not phase coupling) for each pair of channels in this way:
>
> Example to assess the significance of WPLI value for ch1 vs ch2
>
> 1) Calculate the WPLI for ch1 and ch2, this would be the observed WPLI
> (WPLI_observed)
>
> 2) Randomly permute the ch2 time series
>
> 3) Calculate the WPLI for ch1 and ch2 (WPLI_i)
>
> 4) Repeat step 2 and 3 (for instance 100 times) in order to create the WPLI_i
> distribution
>
> 5) Calculate the proportion ( # (WPLI_i > WPLI_observed) / # (WPLI_i ) )
> of WPLI_i which are greater than the WPLI_observed, if this proportion
> is < 0.05 I could say that the WPLI_observed represents a significant
> degree of phase, otherwise not.
>
> Does it make sense or is it not the right approach?
>
> Let suppose this is a correct approach, I have two other questions:
>
> First, usually when I compute the WPLI value between two channels I obtain
> a number of WPLI values according to the cross-spectrum times (one WPLI for
> each sliding window), in the steps above I am assuming to compute the
> average WPLI_observed and the average WPLI_i for each step. Does this
> raise any problems?
>
> Second, is it a problem using the same random permutations employed to
> obtain ch1-ch2 (WPLI_i distribution) to calculate also the ch1-ch3 (WPLI_i distribution).
> This is just an implementatiion question. I would like to know if I could
> shuffle the time series of other channels in one step (i.e. for ch1
> something like data.trial{other_than_ch1,perm}), and finally extract just
> the column relative to ch1 from WPLI matrix.
>
> thanks
>
> Matteo
>
>
>
>
>
>
>
>
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20110629/a5b2dbeb/attachment-0002.html>
More information about the fieldtrip
mailing list