[FieldTrip] one-dimensional cluster stats

[Roy Cox]

hi all,

I'm trying to trick Fieldtrip into doing cluster stats on "one-dimensional"
EEG data. What I mean by that is that for every subject, and every
electrode, I have only one value (average density of sleep spindles). So no
time dimension, just a spatial one.

I assumed (but I'm not sure) that Fieldtrip always requires a time
dimension, so I'm essentially copying the data over to a second sample, and
set 'avgovertime' to yes.

I have a simple between-subjects design (at least for a first pass) with 16
and 20 subjects respectively.

my code:

cfg=[];

*%I get my chanlocs from eeglab*
cfg=eeglab2fieldtrip(myeeg,'chanloc');
cfg.rotate=90; %because eeglab and fieldtrip have different orientations
cfg.layout=ft_prepare_layout(cfg);
cfg.label=cfg.elec.label;

cfg_neighb.method    = 'distance';
cfg.neighbours       = ft_prepare_neighbours(cfg_neighb,cfg.elec);

cfg.channel          = {'EEG'};
cfg.latency          = 'all';
cfg.method           = 'montecarlo';
cfg.statistic        = 'ft_statfun_indepsamplesT';
cfg.correctm         = 'cluster';
cfg.clusteralpha     = 0.05;
cfg.clusterstatistic = 'maxsum';
cfg.minnbchan        = 2;
cfg.alpha              = 0.05;
cfg.tail = 2;
cfg.numrandomization = 1000;
cfg.avgovertime      = 'yes';

cfg.design: [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2]
cfg.ivar: 1

*%data to compare*
spindledata.dimord='subj_chan_time'; %
spindledata.time=[0 1]; %pretend I have two timepoints
spindledata.label=cfg.label;

%*paste data (and use repmat to make a copy for the second timepoint)*
spindledata1=spindledata; spindledata2=spindledata;
spindledata1.avg=repmat(allData(allLabels==1,:),[1 1 2]);
spindledata2.avg=repmat(allData(allLabels==2,:),[1 1 2]);


[statT] = ft_timelockstatistics(cfg,spindledata1,spindledata2);

Matlab's output:

total number of measurements     = 36
total number of variables        = 1
number of independent variables  = 1
number of unit variables         = 0
number of within-cell variables  = 0
number of control variables      = 0
using a permutation resampling approach
computing a parametric threshold for clustering
Reference to non-existent field 'critval'.
Error using ft_statistics_montecarlo (line 250)
could not determine the parametric critical value for clustering

Error in ft_timelockstatistics (line 181)
    [stat, cfg] = statmethod(cfg, dat, design);

Error in spindlestats (line 149)
[statT] = ft_timelockstatistics(cfg,spindledata1,spindledata2);


When I step into the ft_timelockstatistics to figure out what's happening,
I notice that the cfg.channel field holds only 32 channels (while I have
58). Similarly, right before it crashes the "dat" variable is also a 32x36
matrix while I'd expect a 58x36.

All my data are free of Nans, so that's not it.

It's a great mystery to me so any suggestions are welcome.

Roy
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20151103/31d49649/attachment-0001.html>