[FieldTrip] Cluster-based permutation tests - indepedent variable with 3 levels

[Eelke Spaak]

Dear Matteo,

You are on the right track I think, but are using 'depsamplesT' as a
statfun, which only supports 2 levels of the independent variable
(since it computes t-scores). Try the same with cfg.statistic =
'depsamplesF_univariate'  and I think it should work (using an F stat
to compare scores on more than 2 levels of the independent variable).

Best,
Eelke


On Mon, 16 Mar 2020 at 11:41, Matteo Maran <maran at cbs.mpg.de> wrote:
>
> Dear Fieldtrip users,
>
> I would like to ask you if it is possible to run the cluster-based permutation tests in a within-subject design where one independent variable has more than two levels (e.g. 3). I have a 3x2 within-subject design, with subjects doing a language task in three different settings (A, B and C, which were administered in three different sessions). Factor 1 refers to the setting (setting A, setting B, setting C) and factor 2 to a contrast within the language task (correct item, incorrect item).
>
> I have created 3 cells (all_setting_A, all_setting_B, all_setting_C) with, for each subject, the average of all the trials that were recorded respectively in the session with setting A, B and C. I then tried to run the cluster-based permutation test with the following code:
>
> Nsub = length( all_setting_A);
> cfg = [];
> cfg.channel     = 'EEG';
> cfg.latency     = [0 1];
> cfg.avgovertime = 'no';
> cfg.parameter   = 'avg';
> cfg.method = 'montecarlo';
> cfg.statistic = 'depsamplesT';
> cfg.correctm = 'cluster';
> cfg.clusteralpha = 0.05;
> cfg.clusterstatistic = 'maxsum';
> cfg.minnbchan = 2;
> cfg.neighbours = neighbours;
> cfg.tail = 0;
> cfg.clustertail = 0;
> cfg.alpha = 0.025;
> cfg.numrandomization = 1000;
> cfg.design(1,1:3*Nsub)  = [ones(1,Nsub) 2*ones(1,Nsub), 3*ones(1, Nsub)];
> cfg.design(2,1:3*Nsub)  = [1:Nsub 1:Nsub 1:Nsub];
> cfg.ivar                = 1; % the 1st row in cfg.design contains the independent variable
> cfg.uvar                = 2; % the 2nd row in cfg.design contains the subject number
> main_setting_stat = ft_timelockstatistics(cfg, all_setting_A{:}, all_setting_B{:}, all_setting_C{:});
>
> I then get the following error:
> - Error using ft_statfun_depsamplesT (line 79)
> Invalid specification of the design array.
> - Error using ft_statistics_montecarlo (line 244)
> could not determine the parametric critical value for clustering
> -Error in ft_timelockstatistics (line 184)
>   [stat, cfg] = statmethod(cfg, dat, design);
>
> If I check the line 79 of ft_statfun_depsamplesT, I see that the first error that I get (which then triggers the other two in ft_statistics_montercarlo and in ft_timelockstatistics) refers to this section:
>
> % perform some checks on the design
> sel1 = find(design(cfg.ivar,:)==1);
> sel2 = find(design(cfg.ivar,:)==2);
> n1  = length(sel1);
> n2  = length(sel2);
> if (n1+n2)<size(design,2) || (n1~=n2)
>   ft_error('Invalid specification of the design array.');
> end
>
> The design check generates the error because, since I have a third level, the statement (n1+n2)<size(design,2) is true (n1+n2 would be equal to 2/3 of size(design,2)). Should I then assume that the maximum number of levels of a variable for cluster-based permutation tests is 2? Or is it sufficient to create a copy of ft_statfun_depsamplesT where I allow a third level:
>
> % perform some checks on the design
> sel1 = find(design(cfg.ivar,:)==1);
> sel2 = find(design(cfg.ivar,:)==2);
> sel3 = find(design(cfg.ivar,:)==3);
> n1  = length(sel1);
> n2  = length(sel2);
> n3  = length(sel3);
> if (n1+n2+n3)<size(design,2) || (n1~=n2~=n3)
>   ft_error('Invalid specification of the design array.');
> end
>
> Since in principle by using difference waves it is possible to analyze any K-by-L factorial design (http://www.fieldtriptoolbox.org/faq/how_can_i_test_an_interaction_effect_using_cluster-based_permutation_tests/), I thought that it could be possible to analyze also a main effect with an independent variable with three levels. For example, with K = 3 and L = 2, a 3x2 interaction would be tested by comparing three difference waves.
>
>
>
> Thank you for your help!
>
> Best wishes,
>
> Matteo Maran
> _______________________________________________
> fieldtrip mailing list
> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
> https://doi.org/10.1371/journal.pcbi.1002202