Thinking further, I suppose one way the congruency effect could appear to get split off like that is if some subjectsí N400 have a congruency effect and some donít and they have different scalp topographies, so that might be worth looking at.  I also donít meant to claim that PCA always works, just that when it doesnít it generally takes the form of not successfully separating ERP components.  The part of the N400 variance with the congruency effect should not normally go wandering off into a distinct factor by itself unless (for a spatial PCA) it has a distinct spatial distribution from the rest of the N400 variance, in which case it is indeed arguably a different ERP component.  Keep in mind that, within a given subject, a different scalp topography means that a different part of the brain generated it, although the reverse cannot necessarily be claimed.

Cheers!

Joe

On Apr 10, 2014, at 8:57 PM, Joseph Dien <jdien07@mac.com> wrote:

The spatial step shouldnít be splitting apart the effects in the way you describe.  I certainly donít see anything like that in my N400 datasets, although from time to time I do get datasets where the PCA results are noisy, probably due to the dataset itself being too noisy (too few trials/subjects).  Iíd have to see the data myself to say more.  For example, it may be that there are ERP components involved that youíre not familiar with.  But anyway, statistically the portion of the ERP component associated with the semantic effect should have the identical spatial distribution as the portion that does not and so they should end up being described by the same spatial factor.  I canít think of any statistical mechanism by which they might be split off like that.

Regarding the PARE, when I use NetStation to preprocess, I do use it.  Iíve been meaning to add it to the Toolkit.  Itís a low priority because my sense is that it really doesnít make much of a substantive difference to the results.  Nonetheless, I do intend to implement it at some point.

Cheers!

Joe


On Apr 10, 2014, at 3:40 AM, Petter Kallioinen <petter.kallioinen@gmail.com> wrote:

Hi Joe!
I use your toolkit all the time. I like the structured

I find that the temporal PCA-components usually have nice correspondance to candidate ERP-components (as you know N400 is my primary interest). However in the spatial decomposition things get problematic: Very often the first spatial component is the best in terms of topography and amplitude, but hardly ever show differences between conditions. Condition differences is usually I usually find in 2 to 4th spatial components, with topographies that does not look like typical ERP-components. (Later spatial components usually captures noise, remaining artifacts etc which is fine). So the mapping between the two-step PCAs and ERP components does not seem to be direct. N400-incongruency effects does not appear, in my data, as a centroparietal negativity. The typical N400-negativity is caputerd in the first component while, incongruency-effect is captured in components with less typical topographies. Is this stuff you recognize Joe? Should I evaluate topographical components as related
subcomponents of the same underlying ERP-component rather than as independent ones? And, if the matching between PCA-components and ERP-components is more nested than direct (a unrecognized spatial PC evaluated as part of known and predicted temporal PC) do you have suggestions about how to should do the bonferroni correction?

A second question: What are your views on PARE-correction? I've seen you using it in articles, but it is not a part of the toolkit.

Hope to have you in Stockholm again in the future!

Best regards!
/Petter Kallioinen











------------------------------------------------------------------------------
Put Bad Developers to Shame
Dominate Development with Jenkins Continuous Integration
Continuously Automate Build, Test & Deployment
Start a new project now. Try Jenkins in the cloud.
http://p.sf.net/sfu/13600_Cloudbees
_______________________________________________
Erppcatoolkit-support mailing list
Erppcatoolkit-support@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/erppcatoolkit-support


--------------------------------------------------------------------------------

Joseph Dien,
Senior Research Scientist
Maryland Neuroimaging Center
University of Maryland 

Phone: 202-297-8117
















--------------------------------------------------------------------------------

Joseph Dien,
Senior Research Scientist
Maryland Neuroimaging Center
University of Maryland 

Phone: 202-297-8117