[Eeglablist] Statistics in EEGlab.eml

[cyril pernet]

Hi Stefon,

Arno forgot that the LIMO_EEG toolbox 
(https://gforge.dcn.ed.ac.uk/gf/project/limo_eeg/) includes such design.
Although the toolbox is designed for hierarchical linear modelling, 
nothing prevents you to input your inter-trial coherence matrix into the 
rep ANOVA function.

Make a matrix Y [electrode, time frames, subjects, conditions] and a 
vector gp (e.g. 1 1 1 1 1 1 2 2 2 2 2 2)
Then simply call limo_random_robust(6,Y,gp,[2],0) the [2] indicates you 
have 1 factor with 2 conditions and 0 indicates you don't want to 
bootstrap this. This will compute the Hotelling test for each electrodes 
and time frames. If you want to bootstrap change e.g. to 1000 and then 
you can use LIMO EEG to look at the results using clustering (ie 
accounting for multiple testing). In this case it will require to 
specify the electrode configuration etc ..

Best,
Cyril

> [Eeglablist] Statistics in EEGlab.eml
>
> Subject:
> [Eeglablist] Statistics in EEGlab
> From:
> Stefon van Noordt <sv05lz at brocku.ca>
> Date:
> 21/06/2012 03:29
>
> To:
> eeglablist at sccn.ucsd.edu
>
>
> Dear colleagues,
>
> If the ‘parametric’ option is selected, does the study module treat a 
> 2 (paired) x 2 (unpaired) design as a mixed-ANOVA with repeated measures?
>
> Comparing differences in inter-trial coherence with a 2 (feedback type 
> – paired
> statistics) x 2 (gender – unpaired statistics) study design seems to 
> produce the same results as when both variables are set to be paired 
> statistics (i.e., for the purpose of testing the outcome, setting 
> gender as paired instead of unpaired).
>
> Thanks for your time.
>
> Best regards,
> Stefon van Noordt
>
>
>
> Re: [Eeglablist] Statistics in EEGlab.eml
>
> Subject:
> Re: [Eeglablist] Statistics in EEGlab
> From:
> Arnaud Delorme <arno at ucsd.edu>
> Date:
> 23/06/2012 17:11
>
> To:
> Stefon van Noordt <sv05lz at brocku.ca>
> CC:
> eeglablist <eeglablist at sccn.ucsd.edu>, David Groppe 
> <david.m.groppe at gmail.com>
>
>
> Dear Stefon,
>
> yes you are correct. When you have a mixed design (paired) x (unpaired), EEGLAB will use (unpaired) x (unpaired) so it is better to use surrogate statistical approaches. There is a warning on the command line that indicates that EEGLAB does so if I remember well. When using surrogate methods, EEGLAB will use a balanced ANOVA approach (unpaired x unpaired) but data shuffling will be computed according to the pairing selection. In the surrogate case, ANOVA is simply use as a metric of distance for your design, so it is not (as) critical to use the correct ANOVA method.
>
> Ideally, we would use the correct repeated ANOVA function for mixed design (paired x unpaired) but there is no such function in Matlab to our knowledge. The only function available is in R. David Groppe (copied to this message) would know more about this. We are planning in the long run to automatically bridge some statistics from Matlab to R. If anybody wants to help, let us know.
>
> Best,
>
> Arno
>
>


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20120627/bda4b723/attachment.html>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: not available
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20120627/bda4b723/attachment.ksh>