[FieldTrip] ft_timelockstatistics cluster correction using ECoG data

[Eric Maris]

Hi Kristen,



If you supply ft_timelockstatistics with data containing multiple channels 
it will always “correct” (this is not the best word to denote what the 
method does, but let’s use it anyhow) for the number of channels. By 
requesting not to do clustering in space you just achieve that a clusters 
are computed in a different way (ie, assuming that the channels measure 
functionally independent signals).



If you do not want correction for multiple channels, then you must analyze 
the data per channel, as you did for Elec11. If you have a low number of 
channels, then you could combine the permutation-based p-value with regular 
Bonferroni correction (dividing the critical alpha-level by the number of 
channels).



Good luck,

Eric Maris



From: Kristen Berry [mailto:kkb268 at nyu.edu]
Sent: vrijdag 2 mei 2014 22:41
To: fieldtrip at science.ru.nl
Subject: [FieldTrip] ft_timelockstatistics cluster correction using ECoG 
data



Dear all,

I am using an intracranial EEG dataset to compare average high gamma power 
waveforms between conditions. I want to correct for multiple comparisons 
across time, but not across number of electrodes. Here is my input code:

cfg=[];

cfg.latency = [0.05 .5];

cfg.parameter = 'trial';

cfg.method = 'montecarlo';

cfg.correctm = 'cluster';

cfg.neighbours = [];

cfg.numrandomization = 1000;

cfg.statistic = 'indepsamplesT';

cfg.channel = ‘all’;

cfg.alpha = 0.05;

cfg.clusteralpha = 0.05;

cfg.tail = 0;

stats = ft_timelockstatistics(cfg, data,control);

Fieldtrip’s montecarlo reference page ( 
<http://fieldtrip.fcdonders.nl/reference/ft_statistics_montecarlo> 
http://fieldtrip.fcdonders.nl/reference/ft_statistics_montecarlo) says, “If 
you specify an empty neighbourhood structure, clustering will only be done 
in frequency and time (if available) and not over neighbouring channels.”

However, my input code corrects for the number of channels as well. As an 
example, when cfg.channel = ‘all’ for ft_timelockstatistics, then Elec11 is 
not significant. However, if the analysis is limited to Elec11 (cfg.channel 
= 11), then Elec11 has a significant cluster.

Is there a mistake in my code that is causing this? Or is there an older 
version of fieldtrip that doesn’t correct for # of channels, when 
cfg.neighbours is empty? Thank you for your input!

Best regards,

Kristen

Kristen Berry

Research Assistant

NYU School of Medicine

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20140504/1f696bbd/attachment-0002.html>