[Eeglablist] quick question about use/interpretation of "duplicate/split" ICs

[Tarik S Bel-Bahar]

Greetings, any quick thoughts from the eeglablist tribe would be
appreciated.

We are having the problem that ICA decomposes into several very similar
ICs,
with one being active the first half of trials, and the other IC being
active in
 the second half of trials. Similar topics may have come up earlier on the
list.

This issue  is probably due to electrode rewetting during the session,
but also shows up as triplicate ICs when looking at ICA decompositions
from merged data including sessions from different days.

Attached are two images, one of each IC,
with the erpimage plot showing the trials "split-up" between ICs.
It's not the best example, but this kind of splitting shows up
quite a bit in our data.

Has anyone come across this issue before ?
Do you happen to have a practical solution to
merge the the duplicate ICs in a principled manner,
so as to create one IC from, for example the two ICs in the attached images
?
When I attempt statistical analyses across or within-subjects ,
is there a problem to watch out for, or a standing known solution,
in such ICA situations ?

Thanks for any help or thoughts you can share! Would be happy to reciprocate
in any way possible.

***EXTRA NOTE: some have suggested that single or multi-subject
ICA clustering would not be
badly affected by such splitting, and using 2-model amica
decompositions we have had a modicum of success
finding singular ICs that once were split.

best wishes, Tarik
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20120111/8cf564c6/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0027-fred-BCRed2_FILE2_CompoP3.png
Type: image/png
Size: 125377 bytes
Desc: not available
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20120111/8cf564c6/attachment.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0027-fred-BCRed2_FILE2_CompoP4.png
Type: image/png
Size: 116002 bytes
Desc: not available
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20120111/8cf564c6/attachment-0001.png>