[Eeglablist] ask again about bootstrap and group analysis. please help

[Arnaud Delorme]

Dear all,

I agree with Chris's description below. It is important to note that 
this is not bootstrap but permutation statistics. Bootstraping would be 
a way to find better confidence interval for the t-value BEFORE 
performing a parametric test on it (comparing it to a table). 
Permutation statistics involve recomputing the distribution of t-values 
using permutation of conditions (by the way, this does not have 
necessarily to be a t-value and can be any distance measure between the 
two distributions). This may also be generalized to more than two 
conditions using ANOVA and reconstruvtion of the ANOVA distribution 
using permutation of conditions. This is described with example page 19 
in the following document:

Delorme, A. (2005) Statistical methods. /Encyclopedia of Medical Device 
and Instrumentation/, in press. PDF link 
<http://www.sccn.ucsd.edu/%7Earno/mypapers/statistics.pdf>.

A custom way to compensate for multiple comparisons is also described below:

Tanji, K., Suzuki, F., Delorme, A., Shamoto, H., and Nakasato N. (2005) 
High-Frequency Gamma-Band Activity in the Basal Temporal Cortex during 
Picture-Naming and Lexical-Decision Tasks. /J Neuroscience/, 25, 
3287-3293. Journal's link 
<http://www.jneurosci.org/cgi/content/full/25/13/3287?maxtoshow=&HITS=10&hits=10&RESULTFORMAT=1&author1=delorme&andorexacttitle=and&andorexacttitleabs=and&andorexactfulltext=and&searchid=1112980421492_6553&stored_search=&FIRSTINDEX=0&sortspec=relevance&journalcode=jneuro>.

Hope this helps,

Arno

Christopher Summerfield wrote:

>dear wu xiang
>
>I think the procedure is outlined quite clearly in the Burgess & Gruzelier
>article Phil Grieve just sent you, but to explain step by step...
>
>1) you have coherence estimates at p electrode pairs for s subjects in 2
>conditions.  This means that you can perform a t-test at each electrode.
>If you have several estimates per pair, for example several time bins, or
>perhaps even an entire time-frequency map, you may have several
>t-estimates per pair. I will call these 'point estimate' statistics.
>
>2) at each pair, take your subject-condition assignments and swap them
>1000 (or more times), recomputing your t-values at every pair or at every
>timepoint or whatever.  This will give you a null distribution
>corresponding to every point estimate stat.
>
>3) you can then determine whether each point estimate falls within alpha
>of its null distribution.
>
>4) to correct for multiple comparisons (for example, imagine that you have
>20 electrodes=190 pairings), rather than checking each point estimate
>against its distribution, enter (for each permutation) the maximum
>statistic across your entire search space. So, in our example, if the
>maximum permuted value from permutation 1 comes from electrode 188, you
>enter that value into the null distribtuion. on the 2nd permutation, the
>max value might come from another electrode.  You then check all your
>electrodes against this distribution. This will provide an 'familywise
>error' correction for multiple comparisons.
>
>good luck
>Chris
>
>_______________________________________________
>eeglablist mailing list eeglablist at sccn.ucsd.edu
>Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
>To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu
>
>
>  
>


-- 

*Arnaud Delorme, Ph.D.*
Swartz Center for Computational Neuroscience, INC, University of San 
Diego California
La Jolla, CA92093-0961, USA

*Tel* :/(+1)-858-458-1927 ext 15/
*Fax* :/(+1)-858-458-1847/
*Web page*: sccn.ucsd.edu/~arno <http://www.sccn.ucsd.edu/%7Earno>
*To think upon*:

    We have met the enemy, and he is us.

        /Walt Kelly/  

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20060304/ea52b3cc/attachment.html>