[Eeglablist] Over-inflated degrees of freedom

[Stephen Politzer-Ahles]

Hi Marco,

This doesn't sound quite correct to me. In my experience, one could average
across electrodes rather than including all 32 electrodes in the model
without coding them as a factor--i.e., each cell of your design (each
condition for each participant, if you're testing an ANOVA with only an
effect of Condition; or each condition for each electrode for each
participant, if you're testing Condition*Electrode) should have only one
value. It sounds like this is what you did for your last analysis.

Furthermore, since your first Condition*Electrode ANOVA did not reveal any
effects, I don't think there's a motivation for averaging across electrodes
and re-running it. If there were a broad effect like that (which your last
analysis suggests there is not), it would have appeared as a main effect of
Condition in the first analysis.

If you have an effect that is limited to one part of the scalp and simply
don't have enough power to find the necessary Condition*Electrode (or
Condition*Region) interaction, there may be other ways--such as using
spatio-temporal clustering, or only testing a few electrodes of interest
(if you had *a priori *predictions about where effects would emerge). Also,
for a whole-head analysis, rather than putting in all 32 electrodes as a
single 32-level factor, many people will group them based on regions and/or
represent the electrodes with a composite of several factors (such as
Laterality and Anteriority); you should be able to find examples of such
analyses in most contemporary ERP articles, and you can also check out this
book chapter:


   - Dien, J. and Santuzzi, A. (2004). Application of repeated measures
   ANOVA to high-density ERP datasets: A review and tutorial. In: Handy, T.
   (Ed.), Event-Related Potentials: A Methods Handbook, Cambridge, Mass: MIT
   Press.


Best,
Steve



Stephen Politzer-Ahles
New York University, Abu Dhabi
Neuroscience of Language Lab
http://www.nyu.edu/projects/politzer-ahles/

On Fri, Oct 31, 2014 at 11:46 PM, Marco Montalto <montaltomarco at onvol.net>
wrote:

> Dear List,
>
> I have conducted a study in which 20 participants (10 female and 10 male)
> were tested during 2 conditions using 32 electrodes placed on the scalp. I
> ran a mixed model ANOVA with within-subject factors condition (2 levels)
> and electrode (32 levels) and between-subject factor gender (2 levels). I
> got no significant main effects or interaction terms and therefore I
> collapsed across electrodes. Now each subject was represented by 32 values
> and therefore the degrees of freedom looked funny (they were huge) but the
> results were exactly as expected. If I took an average of each
> 32-value-cluster, again, I lost all significance. Does the method of having
> each subject represented by 32 values seem sensible to anyone and, if not,
> can anyone suggest a different method of analysis I could adopt?
>
> Thanks in advance for any advise/suggestions!
>
> Regards,
> Marco
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to
> eeglablist-request at sccn.ucsd.edu
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20141102/282306c9/attachment.html>