[FieldTrip] source level statistics on one data set.

[Narayanan Kutty]

Hi Jan,

Thank you for the quick reply. I thought the normalization (srcA-srcB)/srcA
 was necessary to get rid of the bias towards the center of the head. So if
I do it as you suggest

1. would the bias be still there, or will it be taken care of?

2. if I wanted to "correct" for multiple comparisons do I follow your
suggestion with a few extras to do the montecarlo as given here

http://fieldtrip.fcdonders.nl/example/source_statistics#group_level_statistics_over_subjects

Thank you again!

best,
Naran

On Sat, Jan 19, 2013 at 2:08 AM, jan-mathijs schoffelen <
jan.schoffelen at donders.ru.nl> wrote:

> Hi Naran,
>
> It seems you're almost there!
> What I would do is the following:
>
> Rather than computing the relative differece srcDiff, I'd keep the 2
> conditions separate. Then, srcA and srcB (for all subjects) are ready to
> enter into ft_sourcestatistics. No need to call ft_sourcegrandaverage.
> Note: that if you call ft_sourcegrandaverage with cfg.keepindividual =
> 'no', you'll only get an average across subjects, so that's not useful for
> doing statistics to begin with. Alternatively, you could call
> ft_sourcegrandaverage with cfg.keepindividual = 'no', but this is a
> redundant step. Moreover, it is a bit silly, because in that case no
> 'grandaveraging' is done to begin with.
> We don't have tutorial documentation (yet) for doing group statistics on
> source level data, but you may get some inspiration from the tutorials that
> deal with sensor-level data. Instead of using ft_timelockanalysis you
> should use ft_sourcestatistics.
> The final step would be something like this:
>
> cfg = [];
> cfg ...
> stat = ft_sourcestatistics(cfg, srcA1, srcA2, srcA3, ...., srcB1, srcB2,
> srcB3, ....
>
> where srcA1 etc pertain to the single subject results for condition A and
> likewise for B. It now boils down to informing ft_sourcestatistics with the
> appropriate design matrix, e.g.
>
> cfg.design = [1:Nsubj 1:Nsubj;ones(1,Nsubj) ones(1,Nsubj)*2];
> cfg.ivar = 2;
> cfg.uvar = 1;
> cfg.statistic = 'depsamplesT';
>
> In this case you will perform a paired T-test between the 2 conditions.
>
> Best,
>
> Jan-Mathijs
>
>
> On Jan 19, 2013, at 7:00 AM, Narayanan Kutty wrote:
>
> Dear All,
>
> I had a question regarding across subject source level statistics using
> fieldtrip.
>
> For 15 subjects I have two conditions conA and conB. For each subject I
> then
>
> 1. calculate a DICS  beaformer for conA, conB, based on a common filter.
> (lets call them srcA and srcB resp.)
>
> 2. calculate normalized srcDiff  by doing (srcA-srcB/srcA)
>
> 3. interpolate and normalize srcDiff  using ft_sourceinterpolate
> and ft_volumenormalise (lets call the output srcDiffNorm)
>
> After I have done that for each subject  I use ft_sourcegrandaverage
> with  cfg.keepindividual = 'yes'; to get grandavgAvsB.
>
> What I would like to do is test just grandavgAvsB to see if any voxels are
> significantly different from zero (zero being conA = conB).
>
> How should I go about doing this using ft_sourcestatistics (or another
> program). The example scripts seem want two datasets.
>
> sincerely
> Naran
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>
>
> Jan-Mathijs Schoffelen, MD PhD
>
> Donders Institute for Brain, Cognition and Behaviour,
> Centre for Cognitive Neuroimaging,
> Radboud University Nijmegen, The Netherlands
>
> Max Planck Institute for Psycholinguistics,
> Nijmegen, The Netherlands
>
> J.Schoffelen at donders.ru.nl
> Telephone: +31-24-3614793
>
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20130119/9f3ed83c/attachment.html>