I never met Walter A Shewhart, but his insight of how to “see” process behavior through data is truly profound. W. Edwards Deming along with H. Alan Lasater and David S. Chambers of the University of Tennessee, Knoxville were instrumental in introducing me to the techniques Shewhart’s used to analyze process behavior. Through this experience my understanding of all statistical techniques and my life as well were forever changed.
Once, in a seminar for managers and supervisors primarily from manufacturing companies, David Chambers said, “If a predictable process is shown to produce 10% nonconforming units, then that process was doing what it was ‘designed’ to do.” I was flabbergasted! Why in the world would anyone “design” a process to produce 10%, or any other %, nonconforming? David assured me that this was not intentional. But what all too often happens is that people with the best knowledge at the time, making the best decisions possible at the time, will not necessarily build a process that produces 100% conforming results.
So, if perfection out of the shoot is not what typically occurs, what to do? Before answering this question, let’s consider another question. What in the world did Shewhart do that was so profound?
Shewhart worked at the Western Electric Company in Hawthorne, IL in the inspection engineering department. Western Electric manufactured telephone hardware where improving reliability of components was a concern. Reducing failures, repairs and the variability associated with the hardware components was the focus. They demonstrated that reacting to out-of-specification product by making process adjustments led to an increase in variation and reduction in quality. To overcome these problems, a rigorous and costly inspection program was in place to ensure that customers received only “in-specification” product.
If you can’t adjust your way to quality, how do you get there? Shewhart came to the realization that the causes of variation in the process could be categorized as “chance” or “assignable” respectively. (Other terms are “common” causes and “special” causes. I call them “routine” causes and “exceptional” causes.) Regardless of the terms you use, the critical distinction is that routine causes of variation are essentially “designed” into the process as a result of decisions about raw materials, processing techniques, equipment, and maintenance schedules to mention a few; whereas, exceptional causes of variation manifest themselves as departures from the routine. The amount of routine variation determines the “best” a process can do and consequently the “best” the product from that process can be. Causes of exceptional variation on the other hand, always introduce extra variation in the process and the product leading to reduced quality as well as the need for costly additional inspections, rework, missed shipments to customers and the like.
On May 16, 1924, Shewhart distributed a memorandum which outlined the concepts of routine and exceptional causes of variation. The technique he proposed to determine if the variation in a process came solely from routine causes or from routine plus exceptional causes became known as a “control chart.” Many practitioners use the term “process behavior chart” to emphasize that the technique reveals how a process behaves over a period of time.
Using data from a process, limits of variation attributable solely to routine causes were calculated and drawn on the chart. A chart with all data points inside these limits and exhibiting no unusual patterns indicated that the process was “in statistical control” and doing exactly what it was designed to do. If this reality was not good enough, the process needed to be “redesigned.”
Any chart that had points outside the calculated limits or exhibited unusual patterns revealed the presence of exceptional causes of variation and indicated a process the is “out of statistical control.” A process showing exceptional causes of variation is not doing what it was originally designed to do; the process has changed. By signaling the presence of exceptional causes of variation, the control chart serves as a messenger for employees who can then investigate the cause and take action to prevent it from recurring. Such activities lead to process improvement.
Shewhart’s approach was profound because he recognized:
- There were two categories for causes of variation in a process – routine causes of variation that are designed into a process and exceptional causes of variation which indicate a process change.
- Reducing variation from exceptional causes required identification and elimination of the specific cause while reducing variation from routine causes required a redesign of the process.
- Treating a routine cause of variation as if it were an exceptional cause added variation to the process and degraded quality. Treating an exceptional cause of variation as if it were a routine cause prevented timely process improvement.
As it turns out, quantifying “routine” variation and isolating “exceptional” variation is fundamental to every analysis of data and is present in all statistical analyses. Thank you Walter Shewhart, W. Edwards Deming, Al Lasater and Dave Chambers for opening my eyes to this PROFOUND concept.