[FieldTrip] single subject coherence statistics

[Raghavan Gopalakrishnan]

Dear all,

There has been a lot of discussion on the topic of finding statistical
significance between two conditions within the same subject, but there still
seems to be some lack of clarity and issues.

 

As I understand, The first step is to compute frequencies using
ft_freqanalysis, with keep trials as ‘yes’ and output=’fourier’

 

cfg            = [];

cfg.output     = 'fourier';

cfg.method     = 'mtmfft';

cfg.foilim     = [5 100];

cfg.tapsmofrq  = 5;

cfg.keeptrials = 'yes';

cfg.trials = find(data_redef.trialinfo(:,1)==trialopt);

freq           = ft_freqanalysis(cfg, data_redef);

 

freq = 

 

  struct with fields:

 

            label: {36×1 cell}

           dimord: 'rpttap_chan_freq'

             freq: [1×191 double]

    fourierspctrm: [1444×36×191 double]

        cumsumcnt: [76×1 double]

        cumtapcnt: [76×1 double]

        trialinfo: [76×5 double]

              cfg: [1×1 struct]

 

The second step is to compute statistics using cfg.statistic =
‘indepsamplesZcoh’ Below is the example code

 

cfg1 = [];

cfg1.method = 'montecarlo';       

cfg1.statistic = 'indepsamplesZcoh';                                 

cfg1.correctm = 'cluster';

cfg1.clusteralpha = 0.05;

cfg1.minnbchan = 2;               

                        

cfg1.neighbours = neighbours;   % neighbours computed separately before

cfg1.tail = 0;                    % -1, 1 or 0 (default = 0); one-sided or
two-sided test

cfg1.clustertail = 0;

cfg1.alpha = 0.025;               % alpha level of the permutation test

cfg1.numrandomization = 500;      % number of draws from the permutation
distribution

 

cfg1.parameter = 'fourierspctrm';

 

design = zeros(1,size(freq1.fourierspctrm,1) + size(freq2.fourierspctrm,1));

design(1,1:size(freq1.fourierspctrm,1)) = 1;

design(1,(size(freq1.fourierspctrm,1)+1):(size(freq1.fourierspctrm,1) +
size(freq2.fourierspctrm,1)))= 2;

                

cfg1.design = design';

cfg1.ivar  = 1;

cfg1.channelcmb       = {'FC1' 'REMG'}; % coherence between two channels

cfg1.computecritval = 'yes';

stat=ft_freqstatistics(cfg1, freq1, freq2); % freq1 and freq2 are two
conditions from same subject

 

 

But the second step gives an error because of dimension issues with the data

 

Assignment has more non-singleton rhs dimensions than non-singleton
subscripts

 

Error in ft_statistics_montecarlo (line 319)

      statrand(:,i) = dum.stat;

 

Error in ft_freqstatistics (line 193)

    [stat, cfg] = statmethod(cfg, dat, design);

 

Error in Griptask_RT_coherence_analysis (line 140)

stat=ft_freqstatistics(cfg1, freq1, freq2);

 

There are couple of previous posts (below) that reported the same kind of
error, but there is no solution yet

 

https://mailman.science.ru.nl/pipermail/fieldtrip/2013-July/006821.html

https://mailman.science.ru.nl/pipermail/fieldtrip/2010-June/002900.html

 

In the below post, Jan-Mathijs says not to use ‘cluster’ for correctm
option. However, if we don’t use then how to reproduce the effects reported
in “Nonparametric statistical testing of coherence differences” paper?

 

https://mailman.science.ru.nl/pipermail/fieldtrip/2010-April/002830.html

 

 

Is this error due to a bug? Or am I doing any mistake? I appreciate
developers addressing this issue.

 

Thanks,

Raghavan

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20181019/df0506c3/attachment.html>