Disaster Recovery for Control Charts
So I’m standing at my desk last Tuesday, when my legs suddenly turn to Jell-O. Then I notice the overhead lights gently waving. Soon folks are popping out of their offices with quizzical, sometimes queasy expressions. A few phone calls and Internet searches confirm that we have just experienced a very rare, East Coast earthquake. One centered in Virginia but felt from Georgia to Maine. Wow.
We were all a bit distracted for a while. Even the smallest bounce in the floor from someone walking by had us thinking about aftershocks. Toward the end of the day, I found myself at my window, gazing out over the rolling ridges of central Pennsylvania, thinking about the forces at work and imagining that ripple of energy spreading out over the countryside. How it traveled so far, so fast. How it united the entire East Coast in a moment of shared experience. And how it could really mess up a control chart.
An earthquake is a great example of a “special cause of variation.” Very special, actually. Like, once-in-a-lifetime special, at least around here. But it’s not one of those special causes you are going to be able to fix.
I imagined that somebody somewhere would update their control chart the next day -- a U chart of defects, perhaps -- and watch it go from something like this:
...to something like this:
Before the quake, everything was going fine. There was a problem on August 4th, they addressed it, and the process had been in control ever since. Until the quake hit. After the quake, folks were understandably distracted, and so there were many more defects than normal. It is very unlikely to happen again anytime soon. But in the meantime, it’s making your control limits wider than they should be and your center line higher than it should be. What are you to do?
Fortunately, Minitab Statistical Software is ready to help you cope with acts of God and other disruptions to your process. If you know one or more subgroups are out of whack, and you know they are not representative of your process, you can simply omit them from the calculations. This is particularly important when you don't have a lot of data, because each point has more impact on the control limits.
In our example, the data for August 23rd happens to be the 23rd subgroup. (What were the chances of that?) So when you create your U chart, just click the U Chart Options button. Then go to the Estimate tab and enter 23 under Omit the following subgroups when estimating parameters:
Things are back to normal and the situation is again under control. And now you have a memento of the Great Virginia Quake of 2011, enshrined forever in your U chart. Suitable for framing. (If you are feeling particularly nostalgic, you could also circle the fated control group and add some text to further commemorate the occasion. Not sure how to do that in Minitab? Maybe that would be a good idea for another post...)
What were you doing when the quake hit the East Coast?