[Eeglablist] running ICA a second time

[Tarik S Bel-Bahar]

Greetings Ross:

Some brief responses to your questions below, I hope they are helpful. I've
numbered the responses according to your questions.

0. if you have note please search eeglab list archives where you will
definitely find some past discussions.
PLease read through the articles about cleaning with eeglab, which you can
easily find on Google Scholar.
Check out the eeglab tutorial and wiki, particular guidelines.

1a. You can use PCA (see post within last 2 months on this subject based on
a "should I PCA and if so what is best method")
Reducing by PCA may decrease the "validity" or "truthfullness" of your ICs.
Check Joseph Dien's PCA toolkit for some principled methods to do PCA.

1c. You should find similar ICs after 2 ICAs, as the second ICA
decomposition. The ones in the second ICA should be cleaner and more
accurate.
Some people just publish their first ICA results, as it is not always the
case that you gain much from a second ICA.

2a. I assume you have properly cleaned your data before the first ICA.
My recommendation is, if you want to do a second ICA, first consider doing
what eeglab documentation recommend:
after your first ICA, then do artifactual epoch rejection using ICA-based
rejection, then do a second ICA.

2b. it's not clear what your question is, perhaps there is a word missing
in your question.
You could remove noisy or blink or other artifactual components if you want
to, go ahead
and do a second ICA, and compare the results to doing it via the method
suggested in the response to 2a above.

also:
Check out the the ADJUST plugin (amongst others for a variety of cleaning
techniques).
&
Note there are least three camps:
a. those who use ICA just to deblink or otherwise clean their data, and
then they reconstruct the EEG and do their analyses outside of ICA space.
[ergo, ICA is a cleaning tool]
b. those who use ICA to get ICs that reflect brain dynamics and (often)
established ERP components, they do their analyses on ICs [ergo, ICA gives
real brain dynamics]
c. those who use PCA (with or without ICA) to decompose ERPs into spatial
(and/or) temporal components, they do their analyses on PCs
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20130219/7c709e2a/attachment.html>