Vector-valued ClusterStats

Eric Maris e.maris at DONDERS.RU.NL
Fri Jul 2 07:48:55 CEST 2010


Hi Allen,





Thank you very much for the fast response.  So if I understand correctly,
for a cluster map with X clusters the largest cluster is compared to a
permutation distribution created from the largest ClusterStat of each
permutation, the 2nd largest cluster to a distribution constructed from the
2nd ClusterStat of each, etc.  If this is correct and the largest cluster is
just being tested using Max(ClusterStat), why do you say that this
vector-based approach has less power for the largest cluster?



You need a vector-valued critical value that controls the false alarm rate
for all cluster sizes jointly. The first element of this vector-valued
critical value is always larger than the critical value for a scalar test
statistic.



 Also, the permutation distribution for the smaller clusters may not have
all 1000 values if some permutations have fewer clusters, right?  I
appreciate the advice.



Right, but that does not create any problems.



I'm curious to find out if the cluster-based permutation tests reveal an
interesting pattern in your data.





Best,



Eric









Best,

Allen



____________________________________________________________________________
______

Allen Ardestani    Email: aardesta at ucla.edu

Phone: (310) 825-5528

Medical Scientist Training Program

David Geffen School of Medicine at UCLA

Semel Institute for Neuroscience and Human Behavior

760 Westwood Plaza

Los Angeles, CA 90095-1759

USA



From: FieldTrip discussion list [mailto:FIELDTRIP at NIC.SURFNET.NL] On Behalf
Of Eric Maris
Sent: Thursday, July 01, 2010 2:55 PM
To: FIELDTRIP at NIC.SURFNET.NL
Subject: Re: [FIELDTRIP] Vector-valued ClusterStats



Hi Allen,





I've already looked at the 2007 JMN paper by Maris and Oostenveld, but still
have some remaining questions about using a vector-valued test statistic to
assess differences between 2 conditions (correct vs incorrect memory task
performance).  I've attached a cluster map showing the many different sizes
and shapes that make it necessary to use a vector-valued test statistic.



The different sizes and shapes of the clusters does not force one to use a
vector-valued test statistic.



1)      Do the existing FT functions handle vector-valued stats?



Vector-valued stats are not supported by the ft_timelockstatistics and
ft_freqstatistics functions (which call statistics_montecarlo). They are
supported by the older function clusterrandanalysis, but I'm not sure how
this function will behave give that Fieldtrip has gone through several
changes.



Don't expect too much of a vector-valued test statistic. For detecting the
largest cluster, the vector-valued test statistic has LESS power than the
scalar test statistic (Max(clusterstat)). I have played around with the
vector-valued test statistic but there were few cases where it made a
difference.



2)      Are all ClusterStats from each permutation (bigger than a minimum
size) included when constructing a multivariate permutation distribution?



Yes.



3)      Are clusters of all sizes compared to the same permutation
distribution?  I.e.,  would the 3rd largest cluster be compared to a
permutation distribution obtained from just the 3rd largest stat from each
permutation?



The answer to your specific question is "yes".



4)      Our trials each have a baseline period - in order to assess the
differences between the two conditions, should we limit the permutation
distribution to stats selected just from the active periods of each
permutation?



You can compare to a baseline that has the same number of samples as a data
segment obtained in some experimental condition.





Best,



Eric Maris





dr. Eric Maris
Donders Institute for Brain, Cognition and Behavior

Center for Cognition and F.C. Donders Center for Cognitive Neuroimaging

Radboud University
P.O. Box 9104
6500 HE Nijmegen
The Netherlands
T:+31 24 3612651
Mobile: 06 39584581

F:+31 24 3616066
E: e. <mailto:e.maris at donders.ru.nl> maris at donders.ru.nl







Thank you very much in advance!



____________________________________________________________________________
______

Allen Ardestani    Email: aardesta at ucla.edu

Phone: (310) 825-5528

Medical Scientist Training Program

David Geffen School of Medicine at UCLA

Semel Institute for Neuroscience and Human Behavior

760 Westwood Plaza

Los Angeles, CA 90095-1759

USA



----------------------------------

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.

http://listserv.surfnet.nl/archives/fieldtrip.html

http://www.ru.nl/fcdonders/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.

http://listserv.surfnet.nl/archives/fieldtrip.html

http://www.ru.nl/fcdonders/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.

http://listserv.surfnet.nl/archives/fieldtrip.html

http://www.ru.nl/fcdonders/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.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20100702/631bcfa1/attachment-0002.html>


More information about the fieldtrip mailing list