The full range of the canonical data for Indiana appears as below together with the statistics from the various algorithms. It is seen that the plateau alogrithm detects anomales between 6/20/04 and 7/02/04, and also one at 7/23/04. The HW X^2 is very sensitive to missing data.

Looking more carefully at the anomalous events seen by the plateau algorithm between 6/22/04 and 7/2/04, see below, we believe these are caused by the diurnal nature of the data. The KS algorithm on the other hand shows no such anomalies.

The event at 7/24/04 is examined in more detail in the figure below. It is seen that all three methods detect the change in bandwidth.

Maxim made some corrections to the X^2 method with HW, the previous large vales of X^2 for missing data was due to a bug (forgot to use forecasted values instead of zeroes). The X^2 is now calculated by normalizing the data using the variance. The new graph appears below: