Edwards Deming's speech in Japan in 1978.
"A Quick Look at Some New Management Principles"

The following is a transcript of a forgotten speech given in Tokyo in 1978 by W. Edwards Deming for the Union of Japanese Scientists and Engineers (JUSE). Because the original was a poor photocopy, there are small portions of text that cannot be deciphered. Transcript kindly provided to the Deming Institute (USA) by Mike McLean. Source: www.qualitydigest.com

The translation of the article is supplemented with comments and links to supporting materials on our website, highlighted in italics in separate blocks. Translation and comments: scientific director of the AQT Center Sergey P. Grigoryev

Free access to articles does not in any way diminish the value of the materials contained in them.

The spectacular leap in quality of most Japanese-made products, from third class to superior quality and reliability, with astonishing economies of production, began with a rapid explosion in 1950 and continues to this day. The whole world knows about Japanese quality and the sudden surge that began in 1950, but few understand how it happened.

Photo. In the 1950s, Eizaburo Nishibori, a member of JUSE, and Shigeichi Moriguchi of the University of Tokyo invited W. Edwards Deming to lecture on statistical business methods at a session organized by Keidanren, Japan's most prestigious executive society, under its chairman Ichiro Ishikawa (also president JUSE). Source: The Deming Institute .

It seems worthwhile to collect in one place the statistical principles of management that made the quality revolution possible in Japan, since even at the present time most of these principles are not generally understood or practiced in America.

The relative importance of some of the principles described here has, of course, changed over the years since 1950. Some of the principles stated here arose as corollaries of earlier principles. Other consequences could be added, almost endlessly.

The qualitative leap in Japan is not accidental. It was a success that followed a concerted, determined, methodical effort throughout Japanese industry, at all levels of production, including, of course, management, to make statistical methods work. Repeat visits, as well as courses at various levels for managers, engineers and craftsmen, organized by the Union of Japanese Scientists and Engineers, guided and consolidated these efforts. Perhaps such a concerted movement is only possible in Japan.

Obviously, recognizing problems of quality, uniformity and economy, and evaluating attempts to solve them, requires statistical methods and statistical thinking. Statistical methods cover every step of the production line, from the specification and testing of incoming materials, to in-use product testing, to customer research, product design, and redesign. That's why, as I said Walter A. Shewhart , statistical quality control is the broadest term possible for lean manufacturing problems.

Principle 1.

Bringing a production worker's attention to a defective product he has made while he is in a state of statistical control as to the cause of that defect is demoralizing and costly. ¹ . This will not help him at all, since he is, in fact, blindfolded, taking samples of beads, white and red, out of a box of carefully mixed red and white beads. He cannot control the random occurrence of red beads in his samples, his product; he cannot beat the system. He is hampered by the share of red beads in the system. Only management can change the proportion of red beads in the box.

See description E. Deming's experiment with red beads .

In fact, if you call his attention to a defective element while his performance is under statistical control as to the cause of that defect, he will try to change his procedure... [text lost], hoping in vain for improvement. The result of his modification will be... [text lost] an increase in the variations of his results - a backfire that negates his efforts.

See the explanation in the article by Donald Wheeler that we translated. Correct and incorrect ways to use tolerance fields. Should products be sorted according to tolerance margins for defective and non-defective, or should we try to customize the process?

The argument that a production worker has a right to know the state of his affairs and that he will not know until he sees the defective items he produces is one of the common management misconceptions taught in business schools today. The error explained here arises from a failure to understand the best principles of management.

Principle 2.

If an employee has achieved a state of statistical control in his work, but his performance in terms of quality or quantity is unsatisfactory, it is better to transfer him to a completely different job.

See explanation in open solution A critical look at the use of KPIs in the personnel motivation system. Or how management deprives itself of the most important information for managing the company and destroys teamwork .

Principle 3.

Training and retraining of a worker who has not achieved a state of statistical control of his work will still be effective for him.

Principle 4.

The upper and lower limits of a product's tolerance range alone provide a costly and unsatisfactory reference point for the production worker. For example, outside diameter tolerance limits in the range of 1.001 to 1.002 cm tell a production worker that a diameter of 1.0012 cm is within the tolerance range, but this does not help him produce fewer defects and improve productivity, unlike what he might achieve with less effort using statistical methods. (Statemented several years ago by Dr. Shewhart.)

See the explanation in the article by Donald Wheeler that we translated. Correct and incorrect ways to use tolerance fields. Should products be sorted according to tolerance margins for defective and non-defective, or should we try to customize the process?

Principle 5.

Therefore, in order to save money, his job description should require him to achieve statistical control of his work with the distribution of individual elements within the tolerance zone. Under this system, his products will meet the tolerance range without the high cost of detailed inspection. Workers who are under statistical control but whose performance is unsatisfactory may be transferred and trained for another job.

Principle 6.

Good quality does not necessarily mean high quality. This mainly means uniformity and reliability at low cost and quality that meets market requirements.

Principle 7.

Deviation of the quality characteristic from the nominal value causes losses, even if the deviations occur within the tolerance range. With greater uniformity, a manufacturer can make savings in the production process, such as reducing some steps or using cheaper raw materials, and still meet specifications. Thus, greater homogeneity allows a) the producer and his consumer - both - to achieve greater savings; and b) provides a better basis for doing business together ² .

Examples abound, but may not be as well known, for example the uniformity of agricultural products is also important for the economical processing of food and other derivatives.

See description Taguchi quality loss functions .

Principle 8.

In a state of chaos (poor supervision, poor management, lack of statistical control), the production worker cannot develop his potential capabilities and ensure product uniformity and productivity.

Statistical methods such as Shewhart control charts provide signals indicating the presence of a specific cause of heterogeneity that requires corrective action.

Principle 9.

It is advisable to divide the reasons for high production costs with loss of competitiveness of enterprises into two categories:

Systemic causes (general or environmental causes) - 85%.
These failures remain in the system until management corrects them. Their combined effect is usually easy to measure. Some individual causes can be identified based on judgment. The rest can be determined experimentally; some by examining records of operations and materials.

Special reasons - 15%.
These reasons are specific to a particular worker or machine. The Shewhart control chart detects a signal of a special cause that the worker can usually identify and correct.

The percentages given only indicate that, in my experience, systemic causes outshine special causes. Product design and service testing are part of the system and are the responsibility of management.

Common causes get their name from the fact that they are common to an entire group of workers: they belong to the system.

No improvement in the system or any reduction in the special causes of variances and problems will occur if management does not address the common causes.

Confusion between the two types of reasons leads to frustration at all levels and leads to greater variability and higher costs - exactly the opposite of what is needed.

Principle 10.

Fortunately, this confusion can be eliminated almost without fail. Simple statistical methods such as distributions, process flow charts, Shewhart control charts, all of which are explained in books, provide signals that tell the operator when to take action to improve the uniformity of his work. They also tell him when to leave the process alone.

Principle 11.

These simple statistical methods minimize the cost of two common errors:

1. Over-correction (over-regulation), searching too often for a specific cause and taking measures that only increase variability and deviation from the goal.

2. Inaction (doing too little or nothing) when there are signs of special causes.

Any of the mistakes are easy to avoid completely. You can avoid error #1 (of the first kind) by doing nothing about special reasons, thereby making #2 as often as possible. The result is chaos. Or you can avoid error No. 2 (type 2) by taking action at the slightest sign of upward or downward deviation. The result is increased variability and even greater chaos.

Statistical methods are the only cost-effective way to achieve stability and minimize losses from both errors.

What is not in the books, unclear or generally unknown to quality control engineers is that the same control charts that send statistical signals to the production worker also provide management with a measure of the totality of problems related to the system itself (common causes) ³ .

Explanation of the previous paragraph of the article by Edwards Deming (Sergey P. Grigoryev)

When a Shewhart control chart shows evidence of the presence of special causes of variability according to Western Electric zonal criteria , this indicates unstable (unpredictable) behavior of the process and indicates to the production worker that the variations observed in the “red” points are due to special reasons that should be eliminated and, if it cannot be eliminated at once, taken under control - this in most cases can be done at the shop level.

At the same time, the control limits (upper and lower red lines) of an unstable process demonstrate the minimum potential of the process state to which it will come after eliminating special causes at the shop level. These same control limits and process position (Central Line, CL) demonstrate a condition due to general (systemic) reasons, when further improvements in the vast majority of cases depend on how the process is designed, and this is within the competence of the company's top management. But systemic changes should be undertaken only after the specific causes of variation have been eliminated! Otherwise, you will not be able to evaluate their effectiveness.

If the Shewhart control chart demonstrates stable (predictable) behavior of the process - the control limits and average of the process demonstrate that the process is in the best state of which it is capable in the existing system of common causes. And if this state of the process does not satisfy the requirements for it, then only systemic changes can improve the state of affairs, and this is subject only to the top management of the company, but not to the shop floor. The stable state of the process will allow you to track the effectiveness of attempts made to improve the system in which this process operates.

Demanding that shop staff independently improve a stable process means creating demotivating conditions for workers in which small and big lies are born. And any sincere attempts by the working staff to improve such a process only lead to increasing its variability , which only worsens the process and causes disappointment in one’s own abilities.

Principle 12.

The first step to improving any manufacturing or measurement process is to achieve statistical control of one key product characteristic, then another, and then another. Once statistical control of the main qualitative characteristics of the process has been achieved, the process is ready for the next stage - improvement, and this is the responsibility of management.

Attempting to estimate the effect of steps taken to reduce common causes will be risky and misleading unless the processes have first been brought under statistical control.

Systemic (general) causes remain even though statistical control of the most important quality characteristics of a product has been achieved (as Dr. [Joseph M.] Juran long ago taught). Again, systemic (general) causes are challenges for management.

Principle 13.

A mechanical regulator(s) that simply maintains quality characteristics within the tolerance range does not provide better product uniformity and no economic benefits from it. It doesn't improve the system. Mechanical controllers can be successfully used in conjunction with computer equipment to create control charts that will provide signals to the operator indicating when a specific cause of deviation has occurred and process adjustments are required.

Principle 14.

Management is not doing its job unless the plant maintains Shewhart control charts to show what proportion of recognized problems are attributable to the system and therefore management's responsibility.

Management's usual assertion, in the absence of statistical methods, that "we do everything we know to improve quality and reduce costs," while true, is merely wishful thinking without understanding and using statistical management methods.

A typical management statement might be, “I understand enough about quality control to manage it effectively from my office as president.” Superficial understanding only appears when problems arise ⁴ .

American management generally assumes that production workers are responsible for all production and quality problems. In the absence of statistical methods, management proceeds from the theorem that if workers did not make mistakes, there would be no problems. This elegant theorem does not improve product uniformity or reduce waste. This is a costly attitude. It hides management's responsibilities and capabilities and ensures that trouble will continue.

The reason for recalling cars for correction, familiar to everyone, occurs whenever something was obviously wrong, is not related to the quality of workmanship, but to the quality of the design, the system, hence the error of management.

The boost to the production worker's morale if he saw a real attempt on the part of management to improve the system and hold the production worker accountable only for what the production worker is responsible for and can control, and not for the deficiencies assigned to him by the system, cannot be overstated .

Principle 15.

A process has uniformity and predictable capabilities only if it is in a state of statistical control. In this state, the basic quality characteristics of the product tomorrow will reliably fall within predictable limits. The volume of production can be predicted, as well as the cost of production.

Drawing. Production line, from design, raw materials to consumer. Data from customer research and service requirements provide the basis for product redesign and changes in production input requirements.

Principle 16.

The consumer is the most important point on the production line. Consumer needs research and service testing are statistical tasks.

Principle 17.

Product performance is the result of the interaction between three components: 1) the product itself; 2) the user and how he uses the product, how the customer installs it, how he cares for it, and the conditions of use (example: the customer allowed dirt to get into a roller bearing); 3) operating instructions, customer training, repair services, training of repairmen and availability of spare parts.

The manufacturer of the equipment for which service records are maintained can continually review the records, learning how faults are distributed among the three components and how to most cost-effectively improve the performance of their product.

Offers:

1. Make frequency distributions for fault diagnosis by type of customer, by service personnel, by type of equipment. Some random switching of service personnel between clients and equipment types would provide the basis for improved training of service personnel.

2. The same is at the request of the customer, according to the type of equipment. The results will show which types of customers and which terms of use are most likely to be satisfied and which are most likely to cause dissatisfaction.

3. Monitor the dynamics of the number of faults by type of fault and type of equipment.

Principle 18.

The criteria for testing a product to declare that it meets or does not meet specification requirements (technical tolerances) are completely different from the criteria for assessing the performance of a product. ¹⁸ .

Principle 19.

On-time delivery of a product demonstrates early delivery for a few days and late delivery for another few days.

This principle occurred to me one day in Japan when I stepped onto the train platform for an arriving train and noticed that there were 10 seconds left before the scheduled arrival time. “Of course,” I remarked, “he would have to be early half the time and behind half the time if his arrival rate was to be defined as on time.”

See the explanation of the operation in the article:
1. Operational definition .

Principle 20.

When producing and marketing a product, management faces two types of problems (challenges): ⁶ :

A. What to do with an already manufactured product. Is it suitable to be sent to a market or to a specific buyer?

B. How to improve the future product.

Problem A always arises for the manufacturer: what to do with today's product? Working on problem B is an investment. Each degree of success in task B makes task A easier in the future.

There was a time when quality control consisted of Problem A to the near exclusion of Problem B. Inspecting the final product to identify defective units was supposed to ensure that only good products were sent to the market or to a particular customer.

It is now known that no amount of end-of-line product testing can guarantee the quality of the product. Some defective items may slip through and cause ill will among buyers. In addition, at the end of the line there is a constant need to waive requirements to avoid wasted time and costs for rework or replacement. The more we rely on inspection, the greater the proportion of defective goods that will reach the market.

The idea that the solution lies in control and additional testing has in some places given way to the idea that something needs to be done about the process to reduce the rate of defects produced.

Statistical acceptance procedures should be the first priority subtask in Task A.

Statistical methods also offer a cost-effective approach to Problem B, process improvement.

Principle 21.

The cost of producing multiple units and testing them does not provide sufficient information to predict product costs, even when combined with a market forecast. It is also necessary to know how the customer will check the batches for compliance with the requirements. Does every part have to meet specifications? (Not possible, of course, if the test is destructive.) Or will the client use an acceptance plan that does not require 100 percent compliance?

Principle 22.

Measurement, simple or complex, is a manufacturing process ⁷ . The product is characterized by numbers obtained using a measuring system. There is no identifiable measurement system unless it is subject to statistical control, including exchange of observers. Statistical control is not a matter of opinion, but a matter of compliance with certain statistical tests of chance.

In my experience, unreliable instruments and unreliable measurements are the source of many problems and disputes in manufacturing. Is the product defective or the measurements are incorrect?

Every manufacturer faces daily challenges related to differences between his own measurement results and those obtained by his customer; between his standard measures and reagents and his client's standard measures and reagents.

Any food manufacturer can recall the horrific experience of having a product, already distributed or ready to be distributed, found to be defective, at least temporarily, due to false positive test results that were later traced to contamination in its own laboratory.

Principle 23.

Due care in manufacturing cannot be defined operationally; therefore, any requirement of due care in production cannot be legally enforceable. However, care in production can be defined and measured. Evidence of care is provided by records of tests in the form of meaningful data (which may take the form of charts and statistical calculations), supplemented by records of corrective actions and results. Instructions for product use and warnings regarding misuse are part of a protocol that indicates a degree of caution on the part of the manufacturer.

See the explanation of the operation in the article:
1. Operational definition .

Principle 24.

The practice of giving a so-called merit award to a production employee or group of employees for the best performance (highest sales, fewest defects, highest production) during an accounting period may well be demoralizing unless the award is based on a satisfactory statistical measurement that will distinguish good performance from simple luck. If the award is not based on a statistical measure of performance, then the award system is just a lottery. As far as I know, there is nothing wrong with lottery if it is called lottery, but lottery can do a lot of harm if it is called merit. The result will be decreased productivity, poor quality, and job dissatisfaction. Confusing luck with merit is costly.

See the explanation in the articles on our website:
1. Don't confuse luck with success ;
2. Bonuses or depreciations for company employees based on the results of the reporting period within the system are the same as rewarding for winning the lottery or punishing for losing .

Principle 25.

Statistical methods are not implemented in companies or government organizations. Statistical methods come from knowledge and experience. Knowledge of statistical theory is fundamental. There are no guaranteed cookbook recipes that you have to follow.

Links

1. Principles 1–17 were described in more detail in my article “On Some Statistical Aids to Economic Production,” Interfaces, Vol. 5, No. 4, August 1975: p. 1–15; also in “My View on Quality Control in Japan,” Reports of Statistical Applications Research, vol. 22, No. 2, June 1975: p. 73–80.

2. Smith, Richard D. “Rescuing Superfoods,” The Sciences, Vol. 16, 1976: p. 13–18.

3. For examples, see W. Edwards Deming, “On Some Statistical Aids to Economic Production,” Interfaces, vol. 5, August 1975: pp. 1–15.

4. Golomsky, William. “Department Size,” Quality Progress, August 1976, p. 13.

5. This principle was constantly emphasized by Harold F. Dodge, of course, back in 1944.

6. This principle was formulated by George Edwards of Bell Telephone Laboratories back in 1942. contributions to acceptance sampling and during his long service as Chairman of the E-11 Committee of the American Society for Testing and Materials.

7. Shewhart, Walter A. “Statistical method from the viewpoint of quality control” , The Graduate School, Department of Agriculture, Washington, 1939; pp. 110–119. (Shewhart, Walter A. "Statistical Method from a Quality Control Point of View," USDA Graduate School, Washington, 1939; pp. 110-119).

Ku, Harry H. “Precision measurement and calibration” , National Bureau of Standards, Washington, vol. 1, Publication 300, 1969. (Koo, Harry H. “Precision Measurement and Calibration,” National Bureau of Standards, Washington, Vol. 1, Publication 300, 1969).

Cameron, Joseph M. “Measurement Assurance” , National Bureau of Standards, Washington, Bulletin no. NBSIR 77.1240, 1977. (Cameron, Joseph M. “Measurement Assurance,” National Bureau of Standards, Washington, Bulletin No. NBSIR 77.1240, 1977).