Another great article by Wheeler. A little more detail is in his Advanced Topics in Statistical Process Control book. I thank him once again for repeatedly reminding us that 3 sigma limits are usually enough. In my organization it is easy to get a time series of data that shows a sign of trouble with just 3 sigma limits. The problem isn't a lack of signals, but a lack of action to identify and resolve the causes, preferably with prevention and not just reaction. So when people in my organization want (like I once did) more sensitive charts via runs rules (including the additional ones out there), CUSUM, EWMA, etc. I try to convince them otherwise. If you already have enough signals to work on, why do you think you need more? If a person is engineering a process I can see the desire for lots of signals, but one needs to remember the pain of chasing down the cause of something that was random (false alarm). Another statistician recommended DOE for understanding a process -- finding out what factors it is sensitive to -- so it can be made more predictable by acting on the lessons from such a DOE.
Another great article by Wheeler. A little more detail is in his Advanced Topics in Statistical Process Control book. I thank him once again for repeatedly reminding us that 3 sigma limits are usually enough. In my organization it is easy to get a time series of data that shows a sign of trouble with just 3 sigma limits. The problem isn't a lack of signals, but a lack of action to identify and resolve the causes, preferably with prevention and not just reaction. So when people in my organization want (like I once did) more sensitive charts via runs rules (including the additional ones out there), CUSUM, EWMA, etc. I try to convince them otherwise. If you already have enough signals to work on, why do you think you need more? If a person is engineering a process I can see the desire for lots of signals, but one needs to remember the pain of chasing down the cause of something that was random (false alarm). Another statistician recommended DOE for understanding a process -- finding out what factors it is sensitive to -- so it can be made more predictable by acting on the lessons from such a DOE.