Why CUSUM detects small shifts better.
Can some one throw some light on use of CumSum Chart and how to interpret the result. Only thing I know is, there are some better features compared to Shewart for detecting smaller shifts, but how ?- this 'how' is not clear.
Hi, maybe I can help you understand the why.
First of all I believe that Minitab uses the tabular method and not the V-mask. The two lines are probably the upper and lower cusum Ci+ and Ci- one for shift in positive direction and one for negative shift of the mean.
Now to why a CUSUM or EWMA is better than Shewhart in detecting small shifts. Its simply because they take historical data into account. On the other hand, for the same reason a CUSUM is not as good as Shewhart at detecting large shifts quickly.
Lets assume a positve one sigma shift in the mean of a process with sigma=1.
To detect this change with a Shewhart chart for individual observations would require approximately 44 datapoints. The beauty of CUSUM is that it accumulates historical data above a chosen level k, usually 1/2 the change you are interested in. The latest value is added to the stored sum of previous observations. The control limits are chosen as a multiple, H of sigma, usually between 4 and 5. Now if k=sigma/2 and H=5 it would take on average 10.4 samples before the sum Ci+ has reached UCL=5*sigma.
Hope this helps you!