Would be interesting to think for a second what would be the normative
prediction in that case (what should be in which value range after proper
whitening) since what is a signal and what is noise is less obvious then,
also potential implications for the power spectrum (could that possibly be
used to summarize the data for visualization?).
the y?? data have their original spatial meaning and the values are SNRs or ?z scores?. Since the whitened data are sphered, they can be rotated in any way. Maybe there is even another more interesting rotation than U.
If the whitening is applied to the data from which the covariance was computed (e.g. baseline periods in evoked data, the whole data set for continuous data), then I think the prediction should be the same. I would like to compare different covariance methods for empty room data, and such a plot method would be very useful.