Update: Freqstatistics Now Yields (Som e) Significant Clusters

Charles Cook charles.cook at ULETH.CA
Thu Jul 2 02:54:59 CEST 2009

Hi Michael,

We are interested in determining just that, the presence of spatially and
temporally contiguous effects that cross a certain threshold between my two
groups (males vs. females) and within them (task 1 vs. task 2). After fully
completing the analysis, I was able to find significant clusters in my
within groups analysis (e.g. TF differences in male participants, task 1 vs.
task 2, using dependent t-test) within a few discrete frequency bands. What
we're now interested in doing is beamforming those discrete windows and
frequencies that we have found with the cluster analysis in FieldTrip. Is
this a feasible approach with the BESA exported TFC files? We had considered
exporting VMP files from BESA and beamforming using Brain Voyager but
thought that it might be simpler to do this all within FieldTrip.



On Thu, 25 Jun 2009 12:00:31 +0200, Michael Wibral
<wibral at BIC.UNI-FRANKFURT.DE> wrote:

>Hi Charles,
>thanks for the update. Cluster based statistics is exactly what the name
says: A statistics telling you whether you have spatially and temporally
contiguous effects that cross a certain threshold - in sum over the cluster.
It is sometimes worth considering, whether this is what you want to test
after all. e.g. extended effects of small effect size per electrode but
large time/frequency extent and effects of large effect size but small
time/frequency extent may have similar cluster statistics. The even compete
in the sense that randomizations of the larger of the two (in total cluster
sum) may still have larger cluster statistics than the smaller of the two,
thus effectively rendering in non-significant. Bear in mind that the only
thing really tested is the exchangeability of the data (which is the null
hypothesis). That may sometimes make your results more difficult to
interpret. You could also try cfg.correctm = 'fdr', to get classical FDR
correction, but you may loose sensitivity in some cases.
>One last thing: Check carefully that there is no factor that has been
balanced over subjects (e.g. response hand) that may be resorted in the
randomizations. For example: half of the subjects report match with the
right hand and non-match with the left hand, the other half responding with
an inverted assignment. This analysis setup:
>1. violates the exchangeability hypothesis from the start (and you know!),
but not in the way you wanted to test it - this is a serious error in
applying randomization testing...
>2. Consequently, it renders all other effects insignificant because the
sorted response hand effects in the randomizations most likely exceed any
other effect in the unrandomized data.

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.

More information about the fieldtrip mailing list