[FieldTrip] non parametric testing

[Robert Oostenveld]

Dear Mehmet,

Since you have 19 channels for which you do the test, you have the multiple comparison problem. That means that you are not controlling for the false alarm rate: with more channels, the chances are increasing that one of the channels will have a false alarm and that yu will therefore make the incorrect decision of rejecting your null hypothesis.

A permutation-based nonparametric test using clustering corrects for multiple comparisons. An independent sample t-test does in general not correct for multiple comparisons, unless you adjust the critical value (i.e. Bonferoni correction). 

If you have 19 channels, you can do the t-test with an alpha level of 5% divided by 19, i.e. 0.2632% 

best regards,
Robert

PS please follow up questions like these on the http://fieldtriptoolbox.org/discussion_list <http://fieldtriptoolbox.org/discussion_list>, to which I have also CCed this answer



> On 6 Aug 2018, at 17:41, Mehmet Akif Özçoban <mehmetakifozcoban at gmail.com> wrote:
> 
> Dear professor, I want to ask you about a problem, please could you help me. 
> 
> I am working obsessive compulsive disorder and i have 30 patient and 30 helathy control data. 
> 
> I compute synchronization value of each channel and compared with using independent sample t-test. 
> 
> Some reviewers advice me that you should not use each channels synchronization and you should use nonparametric testing of EEG. But I want to conclude it according to spatial features. I am listening some lessons and they used it for  comparison of two situation not 19 situtaion (synhronization value for channels)
> 
> How can i adapt my research your permutation testing statistical method.
> 
> Thank youmy manuscript had been attached. Please can you read for me
> 
> -- 
> Sincerely
> Mehmet Akif ÖZÇOBAN, phD
> Istanbul Gedik University
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20180807/56bd6d10/attachment-0001.html>