Stats tip: Control limits need to be calculated using the correct method

Matt SavageI often tell others that a control chart is one of the most effective and easy-to-use quality tools. Some argue that experimental design is more effective. Maybe so, but can you teach a novice experimental design as quickly as you can teach him a control chart?

A control chart is a simple tool that works well for many applications. One key component of a control chart is the control limits. Without control limits, you don’t have much … unless, of course, you like run charts. So if the control limits are such a key part to a control chart, why do so many problems and questions exist related to control limits?

Most often I find problems exist because someone was taught incorrectly or applied what was taught incorrectly. Many times people don’t understand that control limits are set at +/- 3 sigma from the mean for a specific reason. This reason is that they simply work well. At +/- 3, control limits do a good job signaling an alarm (out-of-control point) when it is warranted and not creating an alarm when it should not be created. Control limits help you to avoid what statisticians call undercontrolling and overcontrolling. Undercontrol is not reacting to a set of data when the data is showing a reason to investigate. Overcontrol is an over reaction to a set of data, or making a change when there is no statistical reason for the change.

The lesson is resist the temptation to calculate control limits at two standard deviations. If you don’t, you will end up with many false alarms.