Here is the device variance, treated the same way as the
network. I have replotted the network, overlaying the smoothed curve with the
original. You can tell me if you have a pref for the
overlay presentation.
Note: The device variance is the time ordered sample
variances of normalized trial scores for 1 minute resolution. The variance, of
course, is chi-squared (or nearly so, to the accuracy of calling the
normalized trial scores normal random variables) but the degrees of freedom
increases with the number of trials in a minute (i.e., increases with number of
online regs). The variances are all normalized to 1, but the variance of
the variances increases with df. Thus the device variance time series
should be thought of as a series drawn from a changing underlying
distribution. This is why the fourier components show a skew, as you can see in
the plot - the values are slightly more extended to the high side of the mean
(the smoothed curve gives a good indication of where the mean is).
-pb