No subject

and PCA are different. In PCA the PCs are orthogonal, i.e., uncorrelated,
but in ICA the ICs are independent, defined in terms of probabilities: p(a)
= p(a|b).

The question whether for ERP data one method has advantages over the other,
is an open question, IMHO, and can (should) be empirically tested.

Best,
Geert van Boxtel

-----Original Message-----
From: eeglablist-admin at sccn.ucsd.edu
[mailto:eeglablist-admin at sccn.ucsd.edu]On Behalf Of Scott Makeig
Sent: donderdag 9 oktober 2003 02:09
To: M D
Cc: eeglablist at sccn.ucsd.edu
Subject: Re: [Eeglablist] PCA vs ICA

>
> Why conduct ICA analysis on continous EEG vs conducting PCA analysis on
> continious EEG data?
>
> Margery Doyle

Margery -

ICA and PCA have quite different objectives.

PCA tries to gather together the most possible channel activity into
each component.  Usually, this activity sums activity in multiple
independent sources, leaving the remaining variance to be accounted by
subsequent principal components.

Therefore PCA is efficient for squashing the maximum variance in the
data (regardless of source) into any specified (lowered) number of
dimensions.

ICA, to the contrary, tries to split the channel activity into as many
independent components as possible. Independent EEG components are
usually quasi-synchronous activity within a cortical patch, or sometimes
within two cortical patches connected by corpus callosum -- or else,
artifact sources - eyes, muscle, ECG, line noise, electrode noise, etc.

So ICA is (super-)efficient for splitting EEG data into simplest
possible physiological components.

Note that separate scalp channel records are NOT separate physiological
components, since no EEG (except electrode noise, usually quite small)
is generated at any electrode.

Scott Makeig

_______________________________________________
Eeglablist mailing list Eeglablist at sccn.ucsd.edu
Eeglablist page: http://www.sccn.ucsd.edu/eeglab/eeglabmail.html
To unsubscribe, send an empty email to eeglablist-unsub at sccn.ucsd.edu