<div dir="ltr">Dear Cyril,<div><br></div><div>I agree with you. That's why it's always recommended to show confidence intervals or error variances whenever possible.</div><div><br></div><div>Makoto</div></div><div class="gmail_extra">
<br><br><div class="gmail_quote">2014-05-13 12:44 GMT-07:00 Dr Cyril Pernet <span dir="ltr"><<a href="mailto:cyril.pernet@ed.ac.uk" target="_blank">cyril.pernet@ed.ac.uk</a>></span>:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<u></u>
<div>
<p>
Hi Makoto & Michal,<br>
<br>
I agree with Makoto about the ICA subspace which can be quite different - there is however another thing to consider<br>
You said ' Should the effect be stronger (in terms of more statistically significant electrodes (dipoles) and timeperiods) on scalp electrodes or in DIPFIT clusters?'<br>
<br>
the problem here is that statistically significant is an estimate under H0, so beside the hypothesis test, you cannot tell if the effect is stronger or weaker in one case or the other because a p value tells nothing about H1 -- to do that you need to look at the actual effect size (like what is the mean uV difference between conditions) and not base your judgment the (correted) p values. You could also test if the effects are different using a test for apparied measures (eg. a paired t-test between (condition A - condition B) on one compoment vs (A -B) on one channel).<br>
<br>
Cyril<br>
<br>
--------------------------------<br>
Dear Michal,
</p><div class="">
<div>
</div>
<div>
That's a simple but deep question.
</div>
<div>
Theoretically the difference between condition can't be smaller in ICA recults since canceling happens in the mixing process and not the other way around (like the law of entropy?)
</div>
<div>
</div>
<div>
However, I believe a major problem in comparing channels with ICs is component selection. The question is how you guarantee that the ICs you choose is a right representative (projecting source) to the channel? What if some subject don't have such ICs? What if some subjects have multiple of such ICs (subspace)?
</div>
<div>
</div>
<div>
One way to investigate this problem is run pvaf analysis (you have pvaftopo under EEGLAB plugin manager)
</div>
<div>
I have an experience of computing the pvaf analysis across subjects per cluster (unpublished data), and the result showed very large standard deviations... it was like mean 30% and SD=30, range 5-80. This means a cluster can explain a channel activity (in my result, of course) only by 30%, and there are huge inter-subject variance.
</div>
<div>
</div>
<div>
This being said, I think it is still ok to stay optimistic and take the theoretical conclusion. You haven't observed horrendously contradicting results, have you?
</div>
<div>
</div>
<div>
Makoto
</div>
<div>
</div>
<p>
2014-05-12 14:02 GMT-07:00 Michal Vavrecka <span dir="ltr"><<a>vavrecka@fel.cvut.cz</a>></span>:<br>
Hello,<br>
<br>
I do have few simple questions and I am curious about your intuitions and arguments:<br>
<br>
I am finishing the paper where I did group analysis of two cognitive states. I visualized both scalp maps and dipoles and their statistical tests. Both visualization are based on fieldtrip monte carlo permutation with cluster based statistics (correction for multiple comparison). I would like to interpret the difference between results on the scalp and inside the brain (DIPFIT). What are your intuitions:<br>
<br>
Should the effect be stronger (in terms of more statistically significant electrodes (dipoles) and timeperiods) on scalp electrodes or in DIPFIT clusters?<br>
<br>
How to interpret the stronger effect on the scalp?<br>
Does the ICA and DIPFIT calculation somehow weaken the ERSP difference?<br>
My intuition is opposite as the source reconstruction has to clean the noise and strengthen the effect that should result in more statistically significant timeperiods in the spectrograms compared to scalp data?<br>
Is there any paper that compares these two approaches?<br>
<br>
Thanks for your answers.<br>
<br>
Michal<br>
</p>
</div><p>
--<br>
Dr Cyril Pernet,<br>
Academic Fellow<br>
Brain Research Imaging Center<br>
Neuroimaging Sciences<br>
University of Edinburgh<br>
<br>
Western General Hospital<br>
Division of Clinical Neurosciences<br>
Crewe Road<br>
Edinburgh<br>
EH4 2XU<br>
Scotland, UK<br>
<br>
<a href="mailto:cyril.pernet@ed.ac.uk" target="_blank">cyril.pernet@ed.ac.uk</a><br>
tel: <a href="tel:%2B44%280%291315373661" value="+441315373661" target="_blank">+44(0)1315373661</a><br>
<a href="http://www.sinapse.ac.uk/" target="_blank">http://www.sinapse.ac.uk/</a><br>
<a href="http://www.sbirc.ed.ac.uk/cyril" target="_blank">http://www.sbirc.ed.ac.uk/cyril</a>
</p>
<p>
<br>
<br>
--<br>
The University of Edinburgh is a charitable body, registered in<br>
Scotland, with registration number SC005336.<br>
</p>
</div>
<br>_______________________________________________<br>
Eeglablist page: <a href="http://sccn.ucsd.edu/eeglab/eeglabmail.html" target="_blank">http://sccn.ucsd.edu/eeglab/eeglabmail.html</a><br>
To unsubscribe, send an empty email to <a href="mailto:eeglablist-unsubscribe@sccn.ucsd.edu">eeglablist-unsubscribe@sccn.ucsd.edu</a><br>
For digest mode, send an email with the subject "set digest mime" to <a href="mailto:eeglablist-request@sccn.ucsd.edu">eeglablist-request@sccn.ucsd.edu</a><br></blockquote></div><br><br clear="all"><div><br></div>
-- <br><div dir="ltr">Makoto Miyakoshi<br>Swartz Center for Computational Neuroscience<br>Institute for Neural Computation, University of California San Diego<br></div>
</div>