<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
<title></title>
</head>
<body bgcolor="#ffffff" text="#000000">
<font face="Calibri">Hi Shogo,<br>
<br>
After computing your cluster-level statistics you indeed take the
largest one, and this will be compared to the permutation
distribution. However, this distribution is constructed by doing
the following e.g. 500 times:<br>
1) randomly swapping the units of observation (UO) between
conditions (in case of a between UO design)<br>
2) computing your cluster-level-statistic for this random
combination of UOs<br>
3) taking the largest of these (largest of <i>all clusters</i> in
this random permutation, not a <i>'selected' cluster</i>)<br>
This will result in 500 'largest cluster-level-statistics'. If
your original statistic is bigger/smaller than a certain
percentage, say 95% when testing single-sided to an alpha of 0.05,
then you reject the null-hypothesis of interchangeability of your
conditions. Whether it should be bigger or smaller depends on your
direction of testing.<br>
<br>
</font>Hope this helps,<br>
<br>
Best,<br>
Roemer<br>
<br>
<br>
On 9-1-2011 3:03, 平野 昭吾 wrote:
<blockquote
cite="mid:CAD5AB44-1CB2-4046-8931-2EE66E5D569D@npsych.med.kyushu-u.ac.jp"
type="cite">Dear Dr Eric Maris and members of Fieldtrip mailing
list <br>
<br>
At first, I'm sorry for my poor English. <br>
I'm a post-doc research fellow of Department of Neuropsychiatry,
Kyushu University, Fukuoka, Japan. <br>
I'm studying about auditory abnormalities in mental disorders, for
example, schizophrenia, bipolar disorder, using MEG. <br>
I'm very impressed and interested in Dr Eric Maris and Dr Robert
Oostenveld paper, "Nonparametrical statistical testing of EEG and
MEG data, on Journal of Neuroscience Methods, 2007". <br>
I'd like to apply their stastical methods to our data, but there
are some points that I can't understand in their paper. <br>
Those are about cluster-based statistics. <br>
In the right column in page 180 in their paper, they wrote 5 steps
of statistics methods. At 5th step, they take the largest of the
cluster-level statistics, then how do they test this largest of
the cluster-level statistics? <br>
I thought that, after 5th step, I should make the permutation
distribution of cluster-level statistics of selected cluster, and
I should test the given cluster-level statistics with this
permutation distribution. Is this right? <br>
Any answer will help me. <br>
<br>
Best regards, <br>
<br>
Shogo Hirano, M.D., Ph.D. <br>
<pre wrap=""><fieldset class="mimeAttachmentHeader"></fieldset>
_______________________________________________
fieldtrip mailing list
<a class="moz-txt-link-abbreviated" href="mailto:fieldtrip@donders.ru.nl">fieldtrip@donders.ru.nl</a>
<a class="moz-txt-link-freetext" href="http://mailman.science.ru.nl/mailman/listinfo/fieldtrip">http://mailman.science.ru.nl/mailman/listinfo/fieldtrip</a></pre>
</blockquote>
<br>
<div class="moz-signature">-- <br>
<font size="3"><font color="darkblue"><font face="calibri">Roemer
van der Meij M.Sc.<br>
PhD student<br>
Donders Institute for Brain, Cognition and Behaviour<br>
Centre for Cognition<br>
P.O. Box 9104<br>
6500 HE Nijmegen<br>
The Netherlands<br>
Tel: +31(0)24 3655932<br>
E-mail: <a class="moz-txt-link-abbreviated"
href="mailto:r.vandermeij@donders.ru.nl">r.vandermeij@donders.ru.nl</a><br>
</font></font></font></div>
</body>
</html>