Download as PDF, PPTX
Uploaded byAGILEMinds
Don reinertsen is it time to rethink deming
The document discusses rethinking some of W. Edwards Deming's ideas in the context of product development. It notes that while Deming's work was influential for repetitive manufacturing, his statistical process control ideas must be adapted when applied to other domains like product development. The document questions if organizations should respond to all variation, try to eliminate variability, and if problems are always best prevented rather than corrected. It suggests Deming's views assume independent and identically distributed outcomes but product development may involve dependent stochastic processes.
In this document
Powered by AI
Exploration of Deming's relevance, importance in manufacturing, and a call to adapt his ideas to product development.
History of Deming’s contributions to statistics and quality control; recognition in Japan and industry.
Questions regarding variability, process control, and differences in approaches to problem-solving in product development.
Explanation of statistical control limits, standard deviation, and implications of adjustments within these limits.
Deming's perspective on process performance, economic views on variability, and decision-making in fixing defects.
Discussion of random processes, how outcomes vary, emphasizing the difference in control dynamics in product development.
Necessity of timely interventions when reaching control limits; potential benefits of higher variability in certain contexts.
How to make economic choices regarding variability and losses; utilizing Taguchi Loss Function for decision-making.
Difference between sampling strategies and experimental designs; importance of efficient data generation.
Designing tests for defective modules and understanding probabilities of failure and success.
Evaluates the effectiveness of prevention methods versus finding and fixing problems in systems.
Discussion on Deming's Red Bead Experiment; the importance of system management and historical performance as benchmarks.
Comparison of decision-making frameworks by Boyd and Deming; relevance of rapid response in development.
Critique on Deming's knowledge of Lean principles; relationship between batch size and quality.
Summary of main learning points including the need for statistical methods, experiments, and understanding randomness.
More Related Content
Similar to Don reinertsen is it time to rethink deming
Don reinertsen is it time to rethink deming
- 1. Is It Timeto Rethink Deming? Lean Kanban Benelux Antwerp, Belgium October 3, 2011 No part of this presentation may be reproduced without the written permission of the author. Donald G. Reinertsen Reinertsen & Associates 600 Via Monte D’Oro Redondo Beach, CA 90277 U.S.A. (310)-373-5332 Internet: [email protected] Twitter: @dreinertsen www.ReinertsenAssociates.com
- 2. Perspective • Deming’s workis extremely important and it has had great influence on repetitive manufacturing. • His ideas are relevant outside of this domain, but they must be used with some knowledge of the target domain. • This involves rethinking a little bit of the mathematics and a lot of the implications. • Deming did not claim that he had optimized his ideas for product development. 2
- 3. Who Was Deming? 1927 US Department of Agriculture 1939 Adviser to US Census Bureau 1945 1950 Taught SPC in Japan, Deming Prize Created 1960 Awarded Japan’s Order of the Sacred Treasure, Second Class Statistics Professor at New York 1900-1993 University, Consultant, Celebrity 1993 Legitimized relevance of statistics to industry. Made SPC a household term. A 1980’s celebrity. 3
- 4. Some Product DevelopmentQuestions 1. Should we respond to random variation? 2. Should we try to eliminate as much variability as possible? 3. What is the essential difference between process control and experimentation? 4. Is it always better to prevent problems than correct them? 5. Is the system, as Deming states, the cause of 94 percent of our problems? 6. Are there other useful approaches? 4
- 5. 1. Statistical Control •For Deming, bringing a process under statistical control is indispensable. • This state occurs when the outcomes of the process lie between upper and lower control limits. • These limits are set at 3 times the standard deviation of the process. • Standard deviation is calculated from the sampled output of the system. • Thus, a process can be classified as in statistical control even when it has very high variation. • This inherently stabilizes the status quo. 5
- 6. Statistical Control Upper Control Limit Value 3 Mean 3 Lower Control Limit In Control Time 6
- 7. Deming’s World View 3 Upper Variation and Lower Process Control Limits under statistical control Common Cause Process Special Cause not under statistical control Shewhart used the terms chance (random) cause and assignable cause. 7
- 8. Inherently Recursive Sample System Output Set Control Limits 3 from Mean Inside UCL and LCL Outside UCL or LCL Common Cause Special Cause No Action Take Action Output Doesn’t Change Output Changes or Drifts Randomly 8
- 9. Making Adjustments • Whenthe output of a process lies randomly between its upper and lower control limits it is under statistical control. • If we make adjustments to a process that is under statistical control it will increase variation and hurt performance. • If the output falls outside its limits this is defined as a special cause and the operator should investigate and correct this cause. • Control limits are not specification limits! 9
- 10. Deming’s Funnel +1 +1 No Adjustment Variance = 1 -1 -1 +2 Offset to +1 +1 Offsetting 0 Adjustment 0 Variance = 2 -1 Offset to -1 -2 10
- 11. Statistical Control Now it’stime to put on your critical thinking hat. “The aim of a system of supervision of nuclear power plants or anything else should be to improve all plants. No matter how successful this supervision, there will always be plants below average. Specific remedial action would be indicated only for a plant that turned out by statistical tests, to be an outlier.” - Out of the Crisis p.58 11
- 12. An Economic View Cost/Benefit No Remaining Variation Analysis Economic Opportunity Not Economical Economical to Correct to Correct Economic Opportunity Fixing or mitigating a defect is a tradeoff between the cost and benefit of fixing it, regardless of the cause. 12
- 13. Deming’s Frame ofReference • As you might expect, Deming views each outcome as an independent identically distributed (IID) random variable — the classic statistics of random sampling. • But, what would happen if we had a Markov Process, where the outcome was a function of both the current state and a random variable. • This is common in product development, e.g. when a second stochastic activity can’t start until the first one finishes. 13
- 14. A Random Walk •We flip a coin 1000 times, add 1 for each head, subtract 1 for each tail, and keep track of our cumulative total. • How many times the cumulative total will return to the zero line during the 1000 flips? Cumulative H T T H T H H Total Time 14
- 15. One Thousand CoinTosses 1st Half Crossings = 38 Cumulative 2nd Half Crossings = 0 50 Average Time Between Crossings = 25.6 40 Maximum Time Between Crossings = 732 30 20 10 0 0 250 500 750 1000 -10 Note: +1 for each head, -1 for each tail Based on example from “Introduction to Probability Theory and Its Applications”, by William Feller. John Wiley: 1968 15
- 16. Cumulative Totals Diffuse Early Probability Late Value of Random Variable Notes: 1. Zero is always most probable value. 2. But, it becomes less probable with time. 3. For large N a binomial distribution approaches a normal distribution. 16
- 17. It’s Not Deming’sFunnel • The randomness that causes a problem will not fix this problem in a reasonable amount of time. • We must intervene quickly and decisively when we reach the control limit. • It is precisely this control of high queue states that is exploited by the magical Kanban approach. (Blocking can be viewed as a M/M/1/k queue.) • And when we intervene we should return to the center of the control range not its edge. • Think of a Drunkard’s Walk on top of a skyscraper. 17
- 18. 2. Eliminating Variability •In manufacturing we try to minimize the variability of a process. • There is a underlying economic reason why this works. • In product development variability plays a very different economic role. • Consider a race with ten runners. 18
- 19. Asymmetric Payoffs andOption Pricing Expected Price Payoff vs. Price Probabilty Payoff x Strike Price Price Price Expected Payoff Expected Payoff = Strike Price Price 19
- 20. Higher Variability RaisesThis Payoff Strike Price Expected Payoff Price Payoff SD=15 Payoff SD=5 Option Price = 2, Strike Price = 50, Mean Price = 50, Standard Deviation = 5 and 15 20
- 21. Manufacturing Payoff-Function* Gain Target Payoff Loss Performance Larger Variances Create Larger Losses *The Taguchi Loss Function 21
- 22. Making Good EconomicChoices Economic Probability Payoff Economic Expectation Function p( x ) Function E ( g ( x )) g ( x ) p( x )dx g( x ) Deming’s Another critical What we want Focus leverage point. to maximize. 22
- 23. 3. Sampling vs.Experimentation SAMPLING EXPERIMENTATION • The population you are • Identify the question you sampling is given. are trying to answer. • Devise efficient sampling • Determine what data you strategies to balance need to answer the accuracy vs. cost. question. • Here sampling design is a • Develop an efficient way to key skill. create this data. • Here experimental design is key skill. 23
- 24. Inferential Statistics Input Output How many modules are defective? Design a sampling strategy to answer this question at the required confidence level. 24
- 25. Design of Experiments Input Output 16 Modules with 1 defective Which, if any, modules are defective? Design a testing strategy to quickly and efficiently answer this question. 25
- 26. Information and Testing Information Probability of Failure Pf Probability of Success Ps Information Generated by Test I t 1 I t Pf log 2 Ps log 2 1 P P f s 0% 50% 100% Probability of Failure 26
- 27. 4. The Cultof Prevention • Is it always better to prevent problems than it is to find and fix them? • This will be quick. • NO. • Minimizing the cost of failure is always a local optimization. 27
- 28. 5. The SystemDominates “I should estimate that in my experience most troubles and most possibilities for improvement add up to proportions something like this: 94 % belong to the system (responsibility of management) 6 % special” - Out of the Crisis p.315 (Responsibility of leadership) “A third responsibility is to accomplish ever greater and greater consistency of performance within the system, so that apparent differences between people continually diminish.” - Out of the Crisis p.249 These statements have terrifying implications. 28
- 29. The Red BeadExperiment • Deming’s epic work is an entertaining con. • It demonstrates vividly that a set of behaviors (that he disapproves of) do not work to improve performance. • How does he work this magic? • The output of the Red Bead Game is a random variable that is completely independent of the applied treatment. • It will demonstrate that NO management method can EVER influence the output of a process. 29
- 30. The Red BeadExperiment Input System Various Workers Treatments Output Rewards Slogans Random Percent Posters Number White Beatings Generator Beads Anything Experimental Design 30
- 31. 6. Deming: Maintainthe Status Quo • For Deming the past history of the system represents the goal and reference point defining whether the system is under statistical control. • Action is not taken when the system is under statistical control. • We react to deviations outside the control range because they indicate that the system is no longer in statistical control. • Thus, we look at the road behind us, through the rear view mirror, and use control limits to prevent ourselves from deviating from our past course. 31
- 32. The OODA Loop •Originally developed by Col. John Boyd, USAF. • F-86 achieves 10:1 kill ratio vs. the technically superior MiG-15. • There are time competitive cycles of action. • The effects of faster decisions are cumulative. • So, complete the loop faster than the competition. Orient Observe Decide Act 32
- 33. Boyd: Influence theFuture • For Boyd we are always walking into new terrain in the fog. The situation changes and we must quickly make choices to exploit these changes. • This means it is critical to detect new information, determine what it means, and take action. • Decision loop closure time is a critical metric. • Boyd is focused on the road ahead and on reacting quickly to obstacles and opportunities. • Which model is most relevant to the way we add value in product development? 33
- 34. Lean Start-Up • TheBoyd model is, in fact, the approach of the Lean Start-up movement. • Start with a testable hypothesis. • Construct a fast, cheap experiment to test this hypothesis. • Use this information to make the best economic choice: persevere or pivot. • Lean Start-up looks much more like Boyd than Deming. 34
- 35. Did Deming UnderstandLean? • There is actually little evidence that Deming had deep understanding of how Lean works. • There are six passing references to Kanban in his book. • He doesn’t appear to understand the critical relationship between batch size and quality. • He has little focus on the speed of feedback loops. 35
- 36. Deming on Kanban (Whena process is in statistical control…) One may now start to think about Kanban or just-in-time delivery. – Out of the Crisis p.333 Kanban or just in time follows as a natural result of statistical control of quality, which in turn means statistical control of speed of production. – Out of the Crisis p.343-344 • Actually, WIP constraints work whether or not a process is in statistical control. • In fact, it is precisely when a process is out of statistical control that high queue states are most likely, and WIP constraints produce the greatest economic benefit. 36
- 37. Conclusion • Cumulative randomvariables behave differently. • Payoff asymmetries change the role of variability. • Sampling is not experimentation. • For the product developer design of experiments is more important than statistical inference. • Statistical control may be unnecessary. • Understand the OODA loop vs. the Deming cycle. • Lose the Red Bead Experiment. • Learn more about probability and statistics. 37
- 38. “The three fields,calculus, probability, and statistics are all in constant use. Mathematicians in the past have tended to avoid the latter two, but probability and statistics are now so obviously necessary tools for understanding many diverse things that we must not ignore them even for the average student.” R.W. Hamming, (1968 Turing Award) from “Methods of Mathematics” 38
- 39. And the BadNews... “...it has long been observed that the mathematics that is not learned in school is very seldom learned later, no matter how valuable it would be to the learner.” Very Seldom != Never R.W. Hamming, (1968 Turing Award) from “Methods of Mathematics” 39