[Eeglablist] Full-epoch length single-trial corrections and assessing significance for frequency-bands

[Makoto Miyakoshi]

Dear Belen,

I hope the authors of the papers have already given you explanations.

I thought European Journal of Neuroscience would publish a nice technical
paper about baseline correction method that criticized the paper you cited.
Their idea was simple; when you compute a joint baseline for ERSP, make
sure that you concatenate baseline data of all single trials (more than 30?
if I remember correctly) to compute unbiased baseline data... something
like that.

Makoto

On Fri, Jan 22, 2016 at 2:19 AM, Belen De Sancristobal <bdesancr at uottawa.ca>
wrote:

> Dear all,
> I am following the Methods described in Grandchamp, R. and Delorme, A.
> Front Psychol, 2011.
> In particular, I am applying the full-epoch length single trial correction
> using eqs. 1, 3, 5 and 6 (this is what is called ERSP_Full-log in the
> paper). I follow this equations for each frequency and channel of interest
> individually. However I am interested, at the end, on grouping my results
> in 6 frequency bands (delta, theta, alpha, low-beta, high-beta and
> low-gamma). Here is my first question: Can I average my ERSP across
> frequencies within each band after applying eq. 6 and not before? This
> means that I do my average across frequencies after I average across trials.
> I am building my own Matlab code in order to make sure that I understand
> the steps.
>
> My second concern is regarding the computation of significance. Here I am
> following the baseline permutation method described in the same paper. It
> is said that prior to computing statistics, single-trial power estimates
> need to be baseline corrected. Since to compute my ERSP estimates I
> followed the full-epoch length single trial correction procedure, I did not
> divide by the baseline (actual pre-stimulus baseline) until I averaged
> across trials (classical baseline approach as opposed to single-trial
> baseline correction). Hence, I understand that to proceed with the
> statistics I have to, after doing the full-epoch length single trial
> correction, divide by real baseline on a single-trial basis as well. Is
> this correct?
> Once I have done that I compute the surrogate distribution for my baseline
> period and compute the exact p-values for the original ERSP values from a
> Wilcoxon test. However, this original ERSP values are now the ones obtained
> from a single-trial baseline correction as opposed to baseline correction
> on the resulting trial averages. Does this makes sense or I am confused?
> Because I am interested on frequency bands, for this statistical analysis
> I am already using ERSP signals that are averaged across frequencies within
> bands.
>
> Hope it is clear enough.
> Thanks a lot,
> Belen S
>
>
> _______________________________________________
> 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
>



-- 
Makoto Miyakoshi
Swartz Center for Computational Neuroscience
Institute for Neural Computation, University of California San Diego
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20160318/d359f813/attachment.html>