[Eeglablist] How can I figure out correlation between original data and reconstructed data

Arnaud Delorme arno at salk.edu
Sun Oct 2 12:38:00 PDT 2005

Dear Dr. Jung,

>I'd like to know how correlate between original data and reconstructed data
>of decomposed independent component
>in continuous EEG data, not epoched data.
>
>
You may use corrcoef() to compute the correlation for each scalp
channel. For instance to compute the correlation between channel 1 after
removing artifactual components (i.e., blink component 3) of the
tutorial dataset (ALLEEG(2)) compared to the raw data (ALLEEG(1)).

chan =1;
tmp = corrcoef(ALLEEG(2).data(chan,:),ALLEEG(1).data(chan,:))
tmp(2) % contains the correlation factor.

You may then average across all data channels. Another measure is the
relative variance

1-var(ALLEEG(2).data(chan,:))/var(ALLEEG(1).data(chan,:))

A last measure, that we judge to be more precise than the percentage of
variance, is the percentage variance accounted for. This is the measure
we prefer using because it tells us how much of the variance in the
original signal a group of component account for. Basically for each channel

var(ALLEEG(1).data(chan,:)-ALLEEG(2).data(chan,:))/var(ALLEEG(1).data(chan,:))

Once more, you may then average across all data channels.

Hope this help,

Arno