[Eeglablist] removal of event-related (time/phase-locked) variance from ongoing oscillatory activity (cf. Makeig '93)

Scott Makeig smakeig at ucsd.edu
Mon Dec 19 16:05:33 PST 2005

Re the  model:   EEG = evoked + induced      as raised in recent posts.
                                        (ERP)      (EEG)

Although this model is easily stated, it is not necessarily 
representative of the actual generation of event-related EEG dynamics 
See our recent TICS paper via http://sccn.ucsd.edu/papers/TICS04.html 
for a discussion of a more flexible modeling environment in which ERP 
phase-locking and power spectral changes (or perturbations) are axes of 
an event-related state space in which all phenomena (at the cortical 
level) occupty some region -- but rarely live at the 'pure ERP' / 'pure 
ERD' boundaries.

One place in EEGLAB where we use the model above is in computing 
event-realted component (or channel) cross-coherence using crossf(). 
There we have an option to remove the ERP from each trial -- this 
removes the possibility that observed coherence arises through common 
ERP features for the two processes being compared. This is a highly 
conservative approach -- since in fact whatever causes the common ERP 
might also cause a change (usually, increase) in coherence.

This issue is far from settled settled - I believe it will require 
modeling the trial-to-trial variations in ERP/EEG to more adequately 
address the degree of meaning in the model above in particular cases...

Scott Makeig

More information about the eeglablist mailing list