Hi and thank you for the responses!<div><br></div><div>The problem in getting a value of "effect" per subject is that i don't have two conditions per subject, I have only one...</div><div>Should i compare it against baseline?</div>
<div>Is it the same procedure?</div><div><br></div><div>Jose<br><br><div class="gmail_quote">On Tue, Dec 6, 2011 at 10:18 AM, Guillaume Rousselet <span dir="ltr"><<a href="mailto:Guillaume.Rousselet@psy.gla.ac.uk">Guillaume.Rousselet@psy.gla.ac.uk</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">Hey Jose,<br>
<br>
following Arnaud's comment, I would strongly encourage you to perform analyses in each subject and report a confidence interval around the difference for each subject. You could then report how many subjects show an effect in each group. See similar strategy here:<br>
<br>
<a href="http://www.frontiersin.org/perception_science/10.3389/fpsyg.2011.00137/full" target="_blank">http://www.frontiersin.org/perception_science/10.3389/fpsyg.2011.00137/full</a><br>
<br>
If you want to make inferences at the group level, other strategies would involve mixed effect models (R might be better suited than Matlab for that) or using a strategy in which you build confidence intervals using data from the normal group, and then classify subjects from the patient group as being above, below, or inside the control confidence interval - a strategy that Cyril Pernet used to investigate dyslexia:<br>
<br>
<a href="http://www.biomedcentral.com/1471-2202/10/67/abstract" target="_blank">http://www.biomedcentral.com/1471-2202/10/67/abstract</a><br>
<br>
This strategy could be extended to multivariate classifiers if you wanted to throw other ERP markers in the mix.<br>
<br>
Finally, you should have a look at Elizabeth Milne paper on autism for ideas of P1 analyses:<br>
<br>
<a href="http://www.frontiersin.org/perception_science/10.3389/fpsyg.2011.00051/full" target="_blank">http://www.frontiersin.org/perception_science/10.3389/fpsyg.2011.00051/full</a><br>
<br>
Best wishes,<br>
<br>
Guillaume<br>
<br>
<br>
<br>
<br>
<br>
************************************************************************************<br>
Guillaume A. Rousselet, Ph.D., senior lecturer & deputy post-graduate convenor<br>
<br>
Centre for Cognitive Neuroimaging (CCNi)<br>
Institute of Neuroscience and Psychology<br>
College of Medical, Veterinary and Life Sciences<br>
University of Glasgow<br>
58 Hillhead Street<br>
G12 8QB<br>
<br>
<a href="http://www.psy.gla.ac.uk/staff/index.php?id=GAR01" target="_blank">http://www.psy.gla.ac.uk/staff/index.php?id=GAR01</a><br>
<br>
Email: <a href="mailto:Guillaume.Rousselet@glasgow.ac.uk">Guillaume.Rousselet@glasgow.ac.uk</a><br>
Fax. <a href="tel:%2B44%20%280%29141%20330%204606" value="+441413304606">+44 (0)141 330 4606</a><br>
Tel. <a href="tel:%2B44%20%280%29141%20330%206652" value="+441413306652">+44 (0)141 330 6652</a><br>
Cell <a href="tel:%2B44%20%280%29791%20779%207833" value="+447917797833">+44 (0)791 779 7833</a><br>
<br>
The University of Glasgow, charity number SC004401<br>
************************************************************************************<br>
<div class="HOEnZb"><div class="h5"><br>
On 6 Dec 2011, at 03:27, Arnaud Delorme wrote:<br>
<br>
> Dear Jose,<br>
><br>
> using 1000 trials, you may perform inferential testing on your subject's trial population. For instance, you H0 null hypothesis may be "There is no difference between the trial activity between condition A and condition B in subject X". If you reject the null hypothesis, then you may conclude that there is a significant difference between A and B for subject X. In other words, you make an inference about the trial population of subject X. If you were to record additional trials of conditions A and B in subject X, you should be able to reproduce your significant effect (subject to the p-value uncertainty obviously).<br>
><br>
> Now, if you have several subjects, you may perform inferential testing on the subjects' population (i.e. the general population assuming your subject sample is appropriate). Your H0 null hypothesis may be "There is no difference between the trial activity between condition A and condition B in the general population". If you record additional subjects (not necessarily the same ones), you should be able to reproduce your effect.<br>
><br>
> To combine the two (subjects and trials) means that you need to change the null hypothesis. Now, your H0 null hypothesis may be "There is no difference between the trial activity between condition A and condition B in subject X, Y, and Z". Your are making an inference about the trial population of subject X, Y, and Z. If you reject this null hypothesis, you should be able to record more data for subject X, Y, and Z and reproduce the effect.<br>
><br>
> The last approach is valid (and possible in EEGLAB) but less interesting than the one using one value per subject - because you usually want to make inference about the general population, not about a few subjects.<br>
><br>
> This is my take on it but I am not a statistician, so there might be different approaches/interpretations.<br>
> Best,<br>
><br>
> Arno<br>
><br>
> On Dec 4, 2011, at 11:02 AM, Jose Rebola wrote:<br>
><br>
>> Hi<br>
>><br>
>> I am running a study to investigate differences between williams syndrome patients and controls in a visual task. I have eight subjects of each population.<br>
>><br>
>> How do I compare the amplitude of the P100 between the populations?<br>
>><br>
>> Should i include only one value (the peak around 100ms on each of the subject's average) per subject ?<br>
>> It seems to me that if i do this i will only have one value per subject and i am "throwing away" the 100 trials per condition that i have<br>
>><br>
>> Isn't there a way that i can compare between populations while retaining within-subject variability?<br>
>> Otherwise, how does it compensate to perform 1000 trials instead of 10?<br>
>><br>
>> I only know how to do two-level analysis when all subjects perform two conditions for example, thus getting one value of "effect" per subject and moving to the next level...<br>
>><br>
>> Isn't there any paralle of this to two populations?<br>
>><br>
>> Hope somebody can help me,<br>
>><br>
>> José Rebola<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>
><br>
><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>
<br>
</div></div></blockquote></div><br></div>