cluster-based analysis
Emanuel van den Broeke
E.vandenBroeke at ANES.UMCN.NL
Thu Sep 23 15:29:35 CEST 2010
Thanks Michael!
You helped me a lot, now everything is clear for me!
Best Emanuel
-----Oorspronkelijk bericht-----
Van: FieldTrip discussion list [mailto:FIELDTRIP at NIC.SURFNET.NL] Namens Michael Wibral
Verzonden: donderdag 23 september 2010 13:54
Aan: FIELDTRIP at NIC.SURFNET.NL
Onderwerp: Re: [FIELDTRIP] cluster-based analysis
Hi Emanuel,
I think there may be a slight misunderstanding about what the cluster correction is doing.
What is actually done is that for the original data the statistical metric is computed (say depsamples t values), and clusters are identified based on the clusterthreshold you set. Then you sum all t-values in every cluster to obtain one t-value sum per cluster. Note that all of these cluster/t-value sums are kept for later!
Then the data are permuted, the statistical metric is computed on the permuted data, clusters are identified, their t-value sums are computed and the maximum t-value sum is kept. This is done again and again (permutation, statistical metric, cluster identification, keeping the max t-sum).
All the maximum t-sums obtained in the perumtations together form the maximum distribution under the null.
Now you go and sort the t-sums from your clusters obtained from the original data in descending order. For each of these you test how many t-values in distribution of maximum t-sums from the permutations are more extreme. If there are no more than alpha% more extreme t-values in in distribution of maximum t-sums from the permutations you consider the original cluster significant.
Note that more than one cluster can be significant. Often, however, the biggest cluster also shows up in some distorted form in the permutations, again with a high t-value sum. Smaller clusters typically have a hard time competing against the t-sums of the permutations of a big cluster. Hence, you often only find one or two clusters.
Michael
-----Ursprüngliche Nachricht-----
Von: "<Emanuel van den Broeke>" <e.vandenbroeke at ANES.UMCN.NL>
Gesendet: Sep 23, 2010 12:18:04 PM
An: FIELDTRIP at NIC.SURFNET.NL
Betreff: [FIELDTRIP] cluster-based analysis
>Perhaps someone can answer my question?
>
>After performing the cluster based analysis of the ERPs one has to do a
>permutation test. I'm wondering on what data this permutation test is
>performed?
>Once you have defined a cluster of adjacent temporal samples you
>calculate the sum of the t-values within each cluster. For statistical
>analysis, I understood, you take the cluster with the highest absolute t-value.
>But on which data does the analysis perfom the permutation test? Is it
>also possible to calculate the mean ERP activity of the cluster period
>in each individual ERP and test these values between the two groups? or
>is the analysis restricted to another way of analysis?
>
>Please let me know,
>
>Best Emanuel
>
>----------------------------------
>The aim of this list is to facilitate the discussion between users of the FieldTrip toolbox, to share experiences and to discuss new ideas for MEG and EEG analysis. See also http://listserv.surfnet.nl/archives/fieldtrip.html and http://www.ru.nl/neuroimaging/fieldtrip.
----------------------------------
The aim of this list is to facilitate the discussion between users of the FieldTrip toolbox, to share experiences and to discuss new ideas for MEG and EEG analysis. See also http://listserv.surfnet.nl/archives/fieldtrip.html and http://www.ru.nl/neuroimaging/fieldtrip.
Het UMC St Radboud staat geregistreerd bij de Kamer van Koophandel in het handelsregister onder nummer 41055629.
The Radboud University Nijmegen Medical Centre is listed in the Commercial Register of the Chamber of Commerce under file number 41055629.
More information about the fieldtrip
mailing list