[FieldTrip] nonparametric statistics

[Eric Maris]

Dear Maya,

 

 

I was recently referred to your paper on the subject of nonparametric
statistical testing in Journal of Neuroscience Methods (2007). 
I find the paper very interesting and it has contributed a lot to the
statistical analyses I'm currently using. 
 
My experiment includes 3 conditions, of which I was a priori interested in
two. Thus, I ran the analysis proposed by you for two of the conditions only
that were of relevant theoretical interest, and found significant
differences at a specific time window and cluster of electrodes. Now I'm
interested in examining the third condition at this specific time and space
frame. I was wondering if you could advise what would the appropriate
statistical threshold/correction be, as I'm unsure after consulting several
of my colleagues.

 

I see the following options:

 

1.   Compare the third condition with each of the other two conditions and
do Bonferroni-correction to account for the fact that you perform three
tests.

2.   Compare all three conditions with a cluster-based permutation test
using an F-statistic statfun (depsamplesF for a within-UO design and
indepsamplesF for a between-UO design). The power of this test will be
somewhat lower than the power of the cluster-based permutation test using
the t-statistic that you have already performed.

 

Examining the difference between the third condition and the first two at
the data-selected time and space frame does not in general control the false
alarm rate (because the selection of the time and space frame is based on
the data).

 

 

Best,

 

Eric

 

 

 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20101206/3286f28f/attachment-0001.html>