[Eeglablist] Questions about clustering

[Ben Cipollini]

Hi,

I'm studying some statistics at school and trying to better understand EEGLAB's methodology better.  As I read through the clustering tutorial (http://sccn.ucsd.edu/eeglab/clusttut/clustertut.html), there are some things I'm not sure about.

1. It seems that the ERPs are limited by N dimensions (just like all components used in clustering), and the tutorial says that PCA is used to limit the dimensions.  However, the ERPs are really ICA components; is it necessary to do PCA at all?  After ICA you have a list of components ordered by the variance they account for, right?  I have a feeling I'm not reading this part of the tutorial right; can somebody enlighten me please? :)

2. The different measures used for clustering have different metrics, so normalization seems smart.  I was expecting some kind of z-scoring; however, the tutorial says that for each set of measurements (spectra, ERP, dipoles, etc) used in the clustering, the components are all divided by the standard deviation of the PCA component that accounts for the largest amount of variance.  Could somebody help me understand why that computation was used?  i'm not good enough analytically with the math yet to try and derive it on paper...

Thank you so much in advance!
Ben Cipollini
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20060605/9b673780/attachment.html>