[Eeglablist] Average re-reference

[Makoto Miyakoshi]

Dear Chiara,

> I get a different pattern, such as late potentials  being negative.

ICA has inherent uncertainty on polarity of the decomposed data.
For example, ICA cannot distinguish 1 = 1*1 = (-1)*(-1) so your P300 at Pz
may be represented by positive ERP with 'red' scalp topo or negative ERP
with 'blue' scalp topo.
I guess you got stumbled on this. I don't know how EEGLAB handles this
issue, but difference in reference is not the inherence cause of the issue
but it superficially affected the polarity reversal by affecting some
conditions upon which EEGLAB determines IC polarity.

For example, if you want the IC that most explains P300 at Pz to show
positive P300, unfortunately we need to use that prior knowledge to correct
the IC polarity.

> And how much not doing so would bias my results?

At the group-level analysis, EEGLAB STUDY has a solution that aligns IC
polarities maximally consistent within each cluster (that does not
guarantee you see negative P300 with 'blue' topo, but at least consistent
across subejcts). If you don't use STUDY to perform group-level analysis,
I'm afraid you need to implement some solution to check IC polarity, which
could be pretty messy in the sense that you need to use some heuristics
from time to time (unless you use prior knowledge.)

Sorry I'm not showing the solution but a problem to reply to you.

Makoto



On Tue, Oct 20, 2020 at 1:15 PM Chiara Terzo <Chiara.Terzo at iit.it> wrote:

> Dear all,
>
>
> I estimated evoked-potentials of an EEG data set collected with 32
> channels. With the reference to the right mastoid I get a specific pattern
> of evoked that are consistent with the previous literature (i.e. positive
> late potentials around 400ms). However, if before doing ICA I re-reference
> my data to the common average (as it's highly recommended), I get a
> different pattern, such as late potentials  being negative. Which one of
> the two should then be reliable? Is it necessary to re-reference to average
> before doing ICA? And how much not doing so would bias my results?
>
>
> Thank you for your help,
>
>
> Chiara
>
>
> _______________________________________________
> 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
>