## Historical Note on Statistical and Technical Control

For many years, I’ve thought about two types of process control. I learned about statistical process control first: the application of control chart theory and methods to monitor and improve a process.

A bit later, I came to appreciate technical or management process control: the application of the Plan-Do-Study-Act cycle to a process to maintain performance. I’m still learning about this type of process control.

The Plan step describes what to do and how to do it to get the desired results from a process.

People or machines do the work.

Someone or some device should study the difference between what is done and the plan—is the output good enough? Does the work method need adjusting?

Finally, what action(s) should be taken based on the study step?

This PDSA cycle sounds very simple but seems to be very difficult to do reliably.

In some industries, people weave statistical process control into management process control—control charts are used to detect important changes in the Study step. This is the case at Tata Steel, the source of the examples in the Ando and Kumar book on Daily Management (discussed in this post).

Stability of one or more control charts relevant to the process implies at least short-term predictability of the process, which is a basis for rational management.

Appreciation of the power of effective control charts depends on management insight and technical control of the process. Thus, technical or management control appears to be a prerequisite for effective statistical process control.

Last week, I read a 1943 article by W.E. Deming that makes the same point.

“A constant cause system may be described as one that produces numerical results that vary from one to another like a sequence of numbers that are drawn from a bowl of physically similar numbered chips, blindfolded, and with replacement and shuffling. Drawing from a bowl is the limiting attainable state of knowledge, which is to say, this is the limiting attainable state of stability.”

“Any observation, measurement, or table of frequencies, comes about as the result of applying an operation. Whatever the operation, a repeated application thereof gives a new observation. Continued repetition gives a sequence of terms-a population of results. Thus, suppose an article is being manufactured, one after another, and let each be measured for its outside diameter. The situation is something like this--

Article: F_{1}, F_{2}, F_{3}, ..., F_{n}

Diameter: X_{1}, X_{2}, X_{3}, ..., X_{n}

The manufacture and the testing and the recording of the diameters constitute the operation of obtaining the n values of X.“

“In the state of randomness or stability, the terms of the sequence behave as if they were being produced by a constant cause system--they behave as if they were being drawn from the bowl. Without stability, probability tolerances for future terms cannot be set purely on the basis of past terms. The attainment of randomness is much more difficult than is commonly supposed. It may require months. Statistical control is more than technical control. In the latter, the terms of the sequence are merely kept within certain bounds. Be they ever so narrow, this is not statistical control. ** Of course, technical control comes first, and is often sufficient** [emphasis added].”

(W.E. Deming, “Opportunities in Mathematical Statistics, with Special Reference to Sampling and Quality Control”, *Science*, Vol. 97, No. 2514 (Mar. 5, 1943), pp. 209-214.)