timelockgrandaverage config

Zachary Daniel Cohen zachary.d.cohen at VANDERBILT.EDU
Tue May 4 00:01:00 CEST 2010

Hello FieldTrip Community,
        Let me start off by saying thanks for an amazing analysis
package; I've been using it since november with our scalp EEG data
analyses and am a huge fan. I recently began looking at grandaverage
cluster analyses, and wanted to check in to make sure I'm configuring
everything correctly because a few things don't look right.

My goal is to compare the pattern of voltage activity across the scalp
for taskA vs. taskB

I preprocess my scalp eeg voltage data using my own functions, I
believe the important details are as follows: 60 subjects, each
subject has approx 800 events where they performed taskA and 800
events where they performed taskB, although it varies with each
subject. Each event is timelocked from 0ms to 1200ms. I run each
subject's taskA data through timelock analysis, and then run
timelockgrandaverage on the set of all subjects' taskA averages. I
then do the same for taskB. I then run timelockstatistics on the two

My cfg is set as follows:
layout: 'HCGSN128_nof.sfp'
neighbourdist: 0.1300
correctm: 'cluster'
clusteralpha: 0.0500
alphathresh: 0.0500
method: 'montecarlo'
clusterstatistic: 'maxsum'
dimord: 'chan_time'
numrandomization: 1000
minnbchan: 1
tail: 0
clustertail: 0
alpha: 0.0500
statistic: 'depsamplesT'
uvar: 1
ivar: 2
keepindividual: 'yes'
computecritval: 'no'
clustercritval: 0.0500
clusterthreshold: 'parametric'
cfg.design is a 2X120 double, where the top row contains the subject
numbers [1:60 1:60] and the bottom row identifies what condition the
data is from (1 for taskA and 2 for taskB).

When the script runs, the following gets output:

there are on average 9.8 neighbours per channel
using "statistics_montecarlo" for the statistical testing
using "statfun_depsamplesT" for the single-sample statistics
constructing randomized design
total number of measurements     = 120
total number of variables        = 2
number of independent variables  = 1
number of unit variables         = 1
number of within-cell variables = 0
number of control variables      = 0
using a permutation resampling approach
repeated measurement in variable 1 over 60 levels
number of repeated measurements in each level is 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2
using the specified parametric threshold for clustering
computing statistic

The questions I have are the following:
1. Should the total number of independent variables be 2, and the
total number of unit variables be 60?
2. What should the total number of repeated measures in each level be?
3. I'm not sure whether to leave the data in 10ms bins, or to bin it
as I would for publication, in say 150ms or 300ms bins, before running
it through the cluster analysis.
4. How should I make intelligent decisions about neighbourdist and
minnbchan? Right now I have chosen .13 because that leaves a
reasonable amount of electrodes defined as neighbours, but I would
prefer to base my decision on something more sound, perhaps a methods
5. Finally, would using the computercritval option be a better choice
than simply setting the level at .05?

Thanks so much for your advice in advance! Reading these forums has
helped a great deal in the past. My apologies if I omitted any
important information, please let me know if I did.

       - Zach

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