These are the cumulative distributions for station PALO for one
hour each on day 135(Notremor) and day 137(Tremor).
The S statistic distributions are clearly distinct, while the MSQDFC
distributions are much less so.

shows the probability distributions for S (the derivative of the cdfs of the previous figure.

Same as previous but for MSQDFC.

The Receiver Operating Characteristic for the S statistic.


These figures indicate that S should be a better detector, however,
we have additional information, in the form of Justin Brown's picks.
The fact that the MSQDFC is bounded by 1 while S is not makes
comparison difficult.

My first attempt at a comparison is to use his picks as delta functions
to sample the values of the two statistics at the times of his picks
(actually I used the origin times, which may be problematic).
I then compared this to the mean values of S and MSQDFC for that hour.
The S sampling produced a number less than the mean while the
sampling MSQDFC produced a number higher than the mean, which would
seem to contradict the analysis using ROCs.  The plots of this
comparison are incomplete.  Any suggestions for better tests would
be appreciated.