[Eeglablist] synchronization: a problem for ICA?

[Scott Makeig]

Ruijiang Li wrote:

>Hi, all:
>
>I have a simple question about using ICA for ERP analysis.
>  The experiment was designed such that every time the stimulus (a
>very loud 50ms white noise) is presented, the subject would blink. So,
>the blink becomes stimulus-locked and almost always co-occur with the
>N100 wave. My job is to isolate these two activities using ICA, which
>I'm not comfortable with because temporal independence is not
>statistically significant for these two sources (eyeblink and N100). My
>question is, according to your experience, is it possible to separate
>the two sources at all?
>
>Thank you very much!
>
>
>  
>
If you decompose a set of average ERPs from such an experiment, ICA 
might indeed not be able to separate overlapping blink and other 
activity. However, decomposing the concatenated single trials from the 
same experiment (or the recorded continuous data, after removing unique 
artifacts) would likely give the separation, since (1) eye blinks might 
occur at other times as well, and (2) other event-related activity might 
likely vary across individual trials.

Scott Makeig