Common and special causes
🧩EXPLANATION
Every process varies. The question is which kind of variation you are looking at.
- Common cause: variation that belongs to the system. Many small influences, always present, producing a stable spread. This month's number differs from last month's and nothing in particular happened.
- Special cause: variation that does not belong. Something identifiable arrived from outside the usual run: a strike, a new depot, a broken scanner.
Walter Shewhart drew the distinction at Bell Laboratories in the 1920s and called the two chance and assignable causes. The names I was taught are the ones 🏛️ W. Edwards Deming made common later.
Why it matters: the two need opposite responses.
| Common cause | Special cause | |
|---|---|---|
| Right response | Change the system | Find the event and deal with it |
| Wrong response | Chase each point | Redesign the system for a one-off |
Deming's word for chasing each point is tampering, and his claim is that it makes the spread wider. A control chart is the tool for telling the two apart: points inside the limits are the system talking.
Often confused with: good and bad results. A terrible month can be common cause. The improvement method that goes with all this is 🧭 Plan-Do-Study-Act.
Where I have seen it
📞 Gideon's objection. On 22 October Gideon Asante of Harbourline told me that small carriers will be punished by noise: one late lorry in ten is a 10 percent swing, and one in a thousand is nothing. I had no answer. He was describing common-cause variation, and a scorecard that ranks carriers monthly would be a tampering machine: warn the bottom five, watch them recover, credit the warning.
It is why 🧪 Report every metric with a range for a month exists. If the range is on the slide, a point inside it does not get a meeting.
The human version of the same idea is 💎 The structure of a system explains more than the people in it. Most of what looks like one person's bad week is the system's ordinary spread.