[Eeglablist] Fwd: Is short epoch necessary for all types of experiment ?

[Tarik S Bel-Bahar]

---------- Forwarded message ---------
Date: Fri, Sep 21, 2018 at 1:35 AM
Subject: Re: Is short epoch necessary for all types of experiment ?
To: Vibra Lab <labvibra at gmail.com>


Hello Ihshan!  here's a set of quick notes below regarding your last
questions. In general, epoching is a usual thing in EEG science, and you
should be finding plenty of good examples in the literature already of how
to work with continuous data. Lean on the might of Google scholar to get a
feel for best/usual choices in the field :)




*About your question 1. *One could analyze full length of the time window.
But you still have to decide if you take an average across the window, or a
moving window, or some other estimate. Here is a good place to defer to
studies that look at Continuous EEG signals, and attempt to pull metrics
from these.

*About your question 2. *The reason why epoching is often used is a basic
fact that EEG students learn before they try to run EEG studies. Try googel
and google scholar first for some basic definitions and reasons.  If you
haven't had a chance to, please be sure to check out the basic readings and
tutorials that are recommended for people just starting out with EEG. The
reasoning for having "epochs" is found in the tradition & history of eeg,
handbooks of eeg, and guidelines/standards for eeg.  Epoching is also use
in various ways in other fields. Essentially, many EEG metrics are computed
from discrete periods that are usually single trials/epochs, regardless of
the total length of the session. Sometimes, it's not even the user that
creates epochs, but the function that is run, whcih needs epochs. For
example many spectral metrics require windows (which are similar to epochs)
across the total time period (when examining non-epoched continuous data).

Also, most EEG studies since the beginning have been event-related studies,
where stimuli are presented every 1 or 2 second. Then the epochs are
averaged to make an ERP which has a good SNR.
However there have been more continuous studies and "resting" studies for
some time.
At the frontiers of continuous data analysis methods, for advanced
researchers, there are also ways to "convolve" events into continuous data
(or more generally, to find ways to cut up the continuous data).
For example, one could examine the covariation between some continuous
event like music, and the continuous EEG. If one had markers for the onset
of different changes in the music, then once could
event-lock and epoch the continuous data to these musical events.
Remember that there are many different kinds of event-related metrics, not
just ERPs.

*About your question 3. *This is kind of traditional
experimental/perceptual psychology, and tradition in EEG research.
Traditionally single epochs (also called trials/segments) are averaged up.
Over the last decade or two there has been a growing interest in
single-trial analyses.

*About your question 4. *Use eeg_regepoch functon, in eeglab.  This will
allow you to make a set of epochs with the length of your choice in a
dataset that does not have events for it. I belive the function actually
makes the events and then epochs the data. Google past
eeglablist discussions on similar topics.


Because your participants are answering a question, you may want to look at
other studies where people give verbal or written or gestural answers.
These should give you good examples of how to minimize artifacts in the
data. You may also want to look at studies where continuous data was
convolved with continuously changing events (such as music, stress, etc..)


... Have you considered breaking up the periods into "each answer period"
instead of the "whole answer period" ? It is important for your analyses
what kind of responses they are providing, and whether they were also
hearing questions provided by a person. If so, then the onset of a specific
question (or the onset of a specific response)  could be useful  "events"
for you to create event-related potentials. This is often done in
continuous EEG studies with ongoing events that have some
regularity/repeatability.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20180922/543fbd1c/attachment.html>