Common cause and special cause variation
The two kinds of variation, how to respond to each, and the cost of tampering.
No two parts are exactly alike. SPC separates variation into two kinds, because they need completely different responses.
Common cause variation
Many small, ever-present sources: slight differences in material, temperature, operator technique, machine vibration. Together they form a stable, predictable pattern. A process with only common cause variation is in statistical control.
Response: if the stable process is not good enough, you must change the system itself (better material, a new fixture, a revised method). Reacting to individual points does not help.
Special cause variation
A specific, identifiable event outside the normal system: a broken tool, a new material lot, a wrong setting, a new operator, a power dip. Special causes make the process unpredictable.
Response: find the cause, fix it, and prevent it from returning. Record what you found in a cause log.
Two mistakes
- Treating common cause as special (tampering): adjusting a stable process after each reading. Deming's funnel experiment shows that this increases variation, in one common adjustment rule by about 41% (the standard deviation grows by ).
- Treating special cause as common: ignoring a real signal and missing the chance to fix a problem.
A control chart keeps both mistakes rare by placing limits at ±3 sigma.
Stable vs capable
- Stable (in control): predictable. Says nothing about meeting specs.
- Capable: the stable output fits inside the specification limits.
A process can be stable and not capable, or capable for now but unstable.
Questions and answers
Our process is in control but makes 2% scrap. What now?
The scrap comes from the system. Use capability analysis, then Pareto, fishbone and designed experiments to reduce common cause variation or re-center the process.
Can a special cause be good?
Yes. A sudden improvement is also a special cause. Find out why and make it permanent.
