Walter Shewhart

Drew the first control chart on a one-page memo to his boss at Bell Labs — and "three standard deviations" got its operational meaning from a manufacturing compromise.

May 16, 1924. Walter Shewhart, a physicist at Western Electric's Hawthorne Works under contract to Bell Telephone Laboratories, sends a one-page memo to his supervisor George Edwards. It contains the first statistical process control chart: a center line for the process mean, two horizontal lines drawn at ±3 standard deviations, and a scatter of measurements over time. Points inside the lines: leave the process alone. Points outside: investigate. The famous three-sigma rule dates to this memo — but it was not derived from any deep theorem. Shewhart picked 3σ as a working balance between false alarms (stopping a process that was actually fine) and missed signals (ignoring a real shift). Three sigma corresponds to roughly a 0.3% false-positive rate under a normal distribution; he judged that acceptable for the cost structure of telephone manufacturing. He published Economic Control of Quality of Manufactured Product (1931); his student W. Edwards Deming carried the method to Japan after WWII, where it became the spine of the Toyota Production System. Three sigma is not a law of nature. It is an engineer's judgment about how often you can afford to be wrong. That judgment became the operational definition of "outlier" for the next century.