What to do if you see a possible shift in a control chart?

This is the second most often asked question about control charts by students. What if you believe you see a shift (it does not matter if it created a run rule violation or not) in the data.

1. The first action is always to investigate the process to see if something may have occurred. You should check at the period you see the shift and back a few sample periods. A process may show a shift some period after the change has occurred, especially if it is not a big change.

1a. If you find out a process change did occur, then define a new stable region with a stage or a new control chart since the change occurred.

1b. If you do not find any process changes in your investigation. You should keep an eye on the process to see if the shift was a random event or an unidentified change. If the shift appears to continue, even if no cause has been identified, create a stage and consider it a new region of stability. This action should be after enough data was taken (more than 2 or 3 points)

2. If you have a shift that also identifies as a special cause event, then you should investigate to find when the change occurred. Again, look at the moment of the shift and them back in time for a few data points. Generally, you have the special cause detection after a number of data points have shown a pattern, so you should stage the chart and define a new stable region.

Most practitioners would agree on the prior guidance. There are some that believe you should define a new stage (stable region) whenever you see a shift by eye alone. As long as you investigate and look for process changes, you are probably OK. One of our Master Black Belt students showed an internal system that automatically created a stage when ever a special cause detection occurred. This was an extreme of the actions in this paragraph. All of this can be beneficial as long as you investigate all apparent special causes and do not use shifting stages to hide an unstable process.

Many practitioners would define a separate stage for a period of drift or process change that separates two stable regions. This makes good sense because it separates out the unstable region from being considered into the control limits.