Meaning
Sequential analysis technique used to detect subtle changes in the mean of a process over time. Data points are compared against a target value to calculate a cumulative sum algorithm that highlights small but persistent shifts. This approach finds applications in quality control and fraud detection where single point outliers are less informative than a trend of deviation.
Detection Logic
Calculating the difference between each observation and a set baseline creates a running total of deviations. When the cumulative sum algorithm is active, these deviations cancel each other out and the sum stays near zero as long as the process remains stable. A sustained shift in one direction causes the sum to climb or fall away from the baseline with increasing speed.
The accumulation of these differences makes the tool far more sensitive than simple threshold monitoring.
Threshold Setting
Establishing a control limit prevents the algorithm from triggering on normal variance. These limits are typically expressed in standard deviations from the mean. If the running sum breaches the boundary, the system signals that the process has moved out of a controlled state.
Change Point
Identifying the exact moment a shift began is a unique advantage of this cumulative approach. Practitioners look at the slope of the plotted sum to trace the origin of the drift. Knowing when a system began to fail allows for more accurate post incident investigation.
