<div class="gmail_quote"><div>Hi Caroline,<br><br>That all looks good to me, as far as I can tell from a quick glance (but I've only ever tried cluster statistics on ERPs, not on time-frequency data, so there might be differences I don't know about).<br>
<br>But 2^6 (64) seems like a very small number of randomizations. The rule of thumb I follow (based on the recommendation in the tutorial) is to use 500 the first time, and if I get some marginal or borderline results then re-do it with something like 1000 or even 5000 randomizations just to be sure. 64 randomizations might not be enough to converge on a reliable result (i.e., if you run the same test several times with only 64 randomizations, you might get different results each time, whereas if you run it several times with 1000 or more randomizations your results probably will be stable)<br>
<br>Best,<br>Steve<br><br> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Message: 2<br>
Date: Tue, 13 Nov 2012 13:40:51 +0000<br>
From: Lustenberger Caroline <<a href="mailto:Caroline.Lustenberger@kispi.uzh.ch">Caroline.Lustenberger@kispi.uzh.ch</a>><br>
To: FieldTrip discussion list <<a href="mailto:fieldtrip@science.ru.nl">fieldtrip@science.ru.nl</a>><br>
Subject: Re: [FieldTrip] fieldtrip Digest, Vol 24, Issue 16<br>
Message-ID:<br>
<<a href="mailto:7C66D90E0C18014E85B44B560D4D5BC81488F538@EXZH1VM.kispi.int">7C66D90E0C18014E85B44B560D4D5BC81488F538@EXZH1VM.kispi.int</a>><br>
Content-Type: text/plain; charset="us-ascii"<br>
<br>
Dear Steve<br>
<br>
Thank you so much for your advice. My script looks now as follows:<br>
<br>
<br>
cfg = [];<br>
<br>
cfg.elec = elec;<br>
<br>
cfg.computeprob = 'yes'<br>
<br>
cfg.neighbours = neighbours;<br>
<br>
%cfg.latency = 'all';<br>
<br>
cfg.frequency = 'all';<br>
<br>
cfg.channel = 'all'%'all';%eleselection % see CHANNELSELECTION<br>
<br>
%cfg.avgovertime = 'no';<br>
<br>
%cfg.avgoverfreq = 'no';<br>
<br>
cfg.avgoverchan = 'no';<br>
<br>
cfg.statistic = 'depsamplesT';<br>
<br>
cfg.numrandomization = 2^6;% 2^number subjects<br>
<br>
cfg.correctm = 'cluster';<br>
<br>
cfg.method = 'montecarlo';<br>
<br>
cfg.clusteralpha = 0.05;<br>
<br>
cfg.alpha = 0.05;<br>
<br>
cfg.tail = 0;<br>
<br>
cfg.clustertail = 0;<br>
<br>
cfg.correcttail = 'alpha'<br>
<br>
cfg.feedback = 'gui';<br>
<br>
cfg.parameter = 'powspctrm';<br>
<br>
% cfg.design = [<br>
<br>
% 1 2 3 4 1 2 3 4 % subject number<br>
<br>
% 1 1 1 1 2 2 2 2 ]; % condition number<br>
<br>
cfg.design(1,:)=[1:6,1:6]<br>
<br>
cfg.design(2,:)=[ones(1,6),ones(1,6)+1]<br>
<br>
cfg.uvar = 1; % "subject" is unit of observation<br>
<br>
cfg.ivar = 2;<br>
<br>
% "condition" is the independent variable<br>
<br>
stat = ft_freqstatistics(cfg,sd, ld);<br>
<br>
<br>
Do you think this is now correct? We perform statistics on high-density EEG Power values and want to see whether differences that we see doing simple t-test also remain when we correct for multiple comparison.<br>
<br>
Again many thanks and all the best<br>
Caroline<br>
<br>
</blockquote></div><br>