P300 correlation

[paolobernardis]

Dear All,

I read on BCI2000 wiki that "The coefficient of determination, or r^2 value, is a statistical measure computed over a pair of sample distributions, giving a measure of how strongly the means of the two distributions differ". So, if the two conditions (i.e. expected vs unexpected stimuli) show on the spectra plot a large difference at, let's say, 320 ms we should observe a r square peak at 320 ms as well, is it correct? Sometimes we observe a different behaviour: spectra with higher differences between the two conditions show, in the r^2 plot, lower peaks, and vice versa.
Kind regards,

Paolo, Marcello e Joanna

[mellinger]

Hi,

it seems you are talking about wave form plots, not spectra.

If the r^2 would just follow the squared difference between conditions, it would not carry any additional information. Instead, it is a correlation, and as such, it takes variances into account.

When the difference is small compared to variances, the r^2 will be small. Only when the difference is large compared to variances, it will approach 1.

Picturing two distribution peaks that represent data from condition 1, and data from condition 2, you will immediately understand that it is difficult to tell the two apart when they are close together. Also, it becomes obvious that it is not the absolute distance between the peaks that determines how difficult they are to tell apart, but the distance in relation to the width of the peaks.

Regards,
Juergen