Shewhart and Control Charts
In this chapter, Deming discusses Shewhart's concepts of variation, common causes - variation caused by the system and special causes - variation caused by something that is not part of the system of common causes.
He mentions the two mistakes, i.e., reacting to an outcome as if it came from a special cause when it came from a common cause, or reacting to an outcome as if it came from a common cause when it came from a special cause.
One of Shewhart's contributions was to develop control charts to minimize the loss from the combination of both mistakes. When the chart indicates no special causes, the process is in statistical control, or stable. In a stable system the performance of the system can be predicted within a range of variation. The performance of an unstable process cannot be predicted. After statistical control is achieved, the process may be improved. An improvement is either a reduction in variation or a movement in the average, up or down, closer to the optimum level. "The control chart is the process talking to us."
There are many potential applications of the control chart concepts, or techniques in industry, education and government. The most important application is in the management of people.
Some managers set specification limits where they think the limits should be. However, there is no logical connection between control limits and specification limits. Using specification limits based on intuition causes loss either from mistake 1 or mistake 2, but no one could know which or the extent of the losses.
Deming discusses some examples where common causes are often confused with or interpreted as special causes. These include accidents on the highway, fires, absenteeism, and malpractice suits. Highway accidents arise mostly from common causes such as drunk driving and unintelligible road signs. These, he says, are not special causes. Malpractice suits in medicine, engineering and accounting all treat the event as a special cause - somebody is at fault. Study and knowledge of variation leads to the conclusion that the event could have come from the process itself. The system may be at fault.
The purpose of this chapter is to illustrate the losses that are caused by tampering with a system or process. At the beginning of this chapter, Deming defines tampering as management by results. Other ways to define tampering include trying to improve the performance of a process or system based on an individual observation or result, or trying to improve the process or system without theory.
The funnel demonstration includes a funnel, a marble and a table, preferably with a cloth on it to record the results. A dot is drawn on the table cloth to represent the target. Then four rules or procedures are used and the results are recorded on the cloth.
Rule 1: Hold the funnel over the target and drop the marble through the funnel 50 times marking each spot where the marble stops. A distribution of spots or plots will occur.
Rule 2: After each drop, move the funnel from the previous position to compensate for the last error. The last error is the basis or reference point for each new drop. Record the spots with a different symbol. A wider distribution will occur. Deming calculates the diameter of this rough circle will be 41% wider than the circle based on rule one. This is tampering, i.e., trying to improve the performance of process each time based on an individual result.
Rule 3: After each drop, adjust the funnel using the target as the reference point. The results will be worse than before.
Rule 4. After each drop, set the funnel over the spot where the marble stops. The results continue to spread out and are even worse than in rule 3.
Tampering with the process (Rules 2, 3 and 4) only makes things worse. Deming says that possible improvements in this process include lowering the funnel, using a thicker table cloth, and using a steel ball rather than a marble. A magnetic target and marble will also improve performance of the process.
Deming provides seventeen examples of tampering based on rule 2. Some of these include reactions to a complaint of a customer, adjustments in interest rates made by the Federal Reserve Board, a reaction to stock market news, changing company policy based on the latest attitude survey, continuous tax law changes that try to correct a previous mistake, and price wars.
Examples of rule 3 include nuclear proliferation, barriers to trade, illicit drug enforcement, and a gambler increasing their bet to cover losses.
Examples of rule 4 include workers training other workers in succession, a group of players in an orchestra tuning their instruments sequentially not against the same source, hanging wallpaper, and copying examples with no theory.
A process may be stable and produce defects and errors. To adjust the process based on a single defect or error is tampering with the process and will make performance worse. Improvement in a process requires studying the process to understand the capability of the process, including the mean outcome and range of variation. If the process is stable, then a planned change can be developed based on theory, then tested, studied, then implemented or rejected.
The purpose of this chapter is to provide: 1) some easy lessons in variation including examples of situations where common cause variations are confused with special cause variations, and 2) some illustrations based on the concept of a loss function.
Deming explains that variation is life. Life is variation, but those who have no knowledge of statistical theory tend to attribute every event to a special cause. One qualification useful to anyone, and definitely needed by anyone in management, is to understand the concept of variation. This understanding of variation will help them understand the system and to stop asking people to explain the day to day, month to month, and year to year ups and downs that come from the variation that is built into the system.
Loss Functions
Deming explains that a loss function shows the losses that a system suffers from different values of some adjustable parameter. A loss function is useful to help one change from the idea of meeting specifications to continual reduction in variation and improvement in the mean outcome through improvement of the process or system. Each individual has a loss function and a combination of people have a loss function. Loss functions are usually not symmetrical but may look something like the illustration below.
Two distributions are shown in the extended graphic illustration below to convey my interpretation of the concept Deming explains in Chapter 10. The distributions are identical except for their means, i.e., the range of variation is the same in both distributions. However, the one on the left creates more loss than the one on the right. The mean of the process described by the distribution on the left is further away from the optimum or minimum loss. The mean of the process on the right is very close to the optimum value. Improvements in both mean and range of variation are possible for the process on the left. Improvement in the range of variation, i.e., reduction of the amount of variation, is possible for the process on the right.
Deming was critical of the zero defect philosophy because it is associated with the idea of meeting specifications as opposed to continual reduction in variation and improvement in the mean outcome through improvement of the process or system.
This point was not entirely clear to me when I read the last chapter of The New Economics. Several articles by Albright and Roth helped to clear up my confusion (See references below). According to these authors, there are two philosophies associated with quality. One concept is the zero defects philosophy and the other concept is the robust quality philosophy based on the Taguchi loss function. According to Roth and Albright, the zero defects philosophy is associated with defining quality as conforming to specifications where the only costs attributed to variation are those that fall outside the specification limits. They refer to this as the goalpost view. However, the robust quality philosophy views any variation from a target value as undesirable because it causes unnecessary costs to be incurred by the manufacturer, the customer or society. The lost function provides a way to estimate these costs. Deming subscribed to the robust quality philosophy as indicated by his discussion of the loss function in Chapter 10. See the summaries below for more on this issue.
No comments:
Post a Comment