[FieldTrip] cluster based timelock statistics minimum time parameter

[Aaron Schurger]

That is a good point. And the choice of min number of time samples
depends heavily on how you have low-pass filtered or smoothed your
data. If you low-pass filter without downsampling, then you've
"artificially" given yourself a much bigger N for a comparison at some
peak in your signal. In general the lower the cutoff frequency of your
low-pass filter (without downsampling), the more time samples (or
lower samplewise p value) you need to make for a significant cluster.
A good rule of thumb is simply to always down-sample to ~ 3 x the
cutoff frequency when you apply a low-pass filter (or smoothing). So
if you low-pass filter at 20 Hz, then downsample to 60 Hz, before
running any permutation test. The permutation test will run a lot
faster too!
Aaron

On Wed, Sep 11, 2013 at 8:53 PM, Max Cantor <mcantor at umich.edu> wrote:
> For a group analysis of ERP data we are interested in limiting cluster size
> by temporal extent (E.g. Only consider clusters 20 ms or longer). Is there a
> clustering parameter equivalent to minnbchan for time samples? Searching
> through the tutorial and help files didn't turn up anything.
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip



-- 
Aaron Schurger, PhD
Post-doctoral researcher
INSERM U992 / NeuroSpin
CEA - Saclay, France
+33-1-69-08-66-47
aaron.schurger at gmail.com
http://www.unicog.org