Mail.ibisuva.nl

This article was downloaded by:[Baez, Pablo]On: 8 January 2008Access Details: [subscription number 778060948]Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK Publication details, including instructions for authors and subscription information:DiscussionJeroen de Mast a; Ronald J. M. M. Does a a Institute for Business and Industrial Statistics (IBIS UvA), University of Amsterdam,The Netherlands Online Publication Date: 01 January 2008To cite this Article: de Mast, Jeroen and Does, Ronald J. M. M. (2008) 'Discussion',Quality Engineering, 20:1, 20 - 22To link to this article: DOI: 10.1080/08982110701685754URL: This article maybe used for research, teaching and private study purposes. Any substantial or systematic reproduction,re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expresslyforbidden.
The publisher does not give any warranty express or implied or make any representation that the contents will becomplete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should beindependently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings,demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with orarising out of the use of this material.
Quality Engineering, 20:20–22, 2008Copyright # Taylor & Francis Group, LLCISSN: 0898-2112 print=1532-4222 onlineDOI: 10.1080/08982110701685754 ABSTRACT The article by Steiner, MacKay and Ramberg offers a sound and an accurate description of the Shainin System for quality improvement.
Therefore, our comments will not address their representation of the Shainin System, but concern the Shainin System itself.
KEYWORDS discovery, exploratory data analysis, problem solving The type of problems that can be tackled with SS is not clearly demar- Downloaded By: [Baez, Pablo] At: 20:39 8 January 2008 cated. However, SS has some intrinsic limitations that makes it inappropriatefor many problems.
The type of problems that SS copes with can be derived from the prin- ciple: ‘‘there is a dominant cause of variation in the process output thatdefines the problem’’ (p. 4). The type of problems that is tackled is appar-ently limited to variation problems. A more general form of problems is: acertain quality characteristic (which can be a variable or an event) has aprobability distribution which does not meet the demands. This class ofproblems includes excessive variation, but also process means that are offtarget, too low yields, too long cycle times, or the occurrence of undesiredevents. Of course, with some imagination many of these problems could berecast into variation problems, but that is artificial and it is not clear what isgained by it. Moreover, it appears—as we argue below—that the assump-tion that the problem is a variation problem is quite fundamental in SS.
PROGRESSIVE SEARCH USING FAMILIES OF VARIATION Finding the dominant cause of a problem is a special case of what is gen- erally called ‘‘discovery’’: the search for possible explanations or solutions(see, e.g., Nickles, 1998). Such a possible explanation or solution is nameda hypothesis or conjecture. Discovery is usually followed up by the justifi-cation of a hypothesis, which amounts to empirical testing of the hypothe-sis’s implications.
Justification is typically a methodical activity: it follows strict criteria and procedures. For discovery, on the other hand, prescriptions take the form Address correspondence to Jeroende Mast, Institute for Business and of heuristics, rather than of methods. The idea is that discovery involves a search through spaces of possible solutions. Heuristics are rules that make Muidergracht 24,1018 TVAmsterdam, The Netherlands.
E-mail: jdemast@science.uva.nl De Mast (2003) and De Mast and Bergman (2006) the deductive testing of potential causes in a suc- discuss a number of heuristics for discovery that are ceeding stage. This testing stage is what ensures effective in industrial quality improvement. One of ‘‘objectivity,’’ not the discovery stage. In other words, these is a heuristic that is named Zooming-in strat- since potential causes are experimentally confirmed egy. The idea is to divide the class of potential causes before they are accepted anyway, we see no reason into subclasses and using the nature of the problem to declare the use of opinions, or even wild guesses, to decide in which of these subclasses the dominant cause is to be found, thus eliminating causes that arein the other subclasses. Progressive search as described by the authors is identical to the zoom- ing-in strategy, with families of variation playingthe role of subclasses. Other heuristics that are SS claims that using inductive reasoning from mentioned in De Mast and Bergman (2006) are observations is more effective than using convictions ‘‘thinking in standard categories’’ (e.g., machine, and knowledge of engineers and operators. This claim—which is an empirical claim—should be ‘‘pattern recognition’’ and ‘‘thinking in analogies.’’ substantiated with evidence. It means somethinglike: Inductive reasoning from observations succeeds more often in identifying a real cause as a potentialcause than using opinions. This seems, however, highly situation dependent. In many processes the For the sake of discovering potential causes one engineers have knowledge of many sources of vari- Downloaded By: [Baez, Pablo] At: 20:39 8 January 2008 uses—in an informal manner—various sources of ation and we see no point in refusing to take this knowledge along as hypothesised causes. Moreover, knowledge, convictions of persons who work with reasoning from observations will help identify only the process, and so on. In SS, however, the use of sources of variation that actually vary during usual convictions and opinions is rejected for the identifi- manufacturing. However, many factors in the pro- cation of potential causes. Compare, for example, cess that affect quality and could be used to arrive the statement ‘‘SS shuns brainstorming and cause- at improvements are kept constant during normal and-effect diagrams when screening possible causes’’ production, and these will not show up as patterns and the claim that ‘‘there is no place for subjective methods as brainstorming or fish bone diagrams inserious problem solving’’ (p. 6). We want to make This last points seems to us the crux. If the type of problems for which the approach is used is limited to variation problems (see our point 1.) then clue There are several approaches for making infer- generation based on data seems more effective than ences (see e.g., Maher, 1998). One of these is the using convictions and opinions. For other types of projects, however, the opposite could prove true.
method says that it is of no importance how potential We give one example of such an improvement causes are found, as long as their effect is empirically verified (by deducing testable consequences from The objective of a certain improvement project was them). An other method is ‘‘inductive generalisation’’ the reduction of the cycle time of a caffeine extraction (or ‘‘enumerative induction’’). In this method poten- process while keeping the resulting caffeine percent- tial causes are derived from observations and also age of the coffee safely below a certain limit. This justified by these same observations.
is not a variation problem, and the principle (‘‘there SS seems to combine the data-based identification is a dominant cause of variation in the process output of potential causes of inductive generalisation with that defines the problem’’) does not hold. The main influence factors in this process (such as the extrac- where a criterion was needed that indicates which tion time, the number of extractions and the amount variation patterns should be an incentive for action, of solvent) were all known to the engineers. Their and which should not. Useful as the division may precise effect onto cycle time and caffeine percent- be in this context, we agree with the authors age, however, were not, and the project focussed that the distinction is of limited use in the context on experiments to model these effects. Using brain- storming the main influence factors were rapidly dis-covered. Had we used observational data, however, these influence factors would not have been discov- ered, because their constant settings during pro-duction have as a result that their effect does not Also in the Six Sigma program a process capability show up as variation patterns in observational data.
study is performed in the early stages of an improve-ment project (step 4 of the DMAIC strategy, see e.g., De Mast et al. 2006). In our experience, one of thereasons why this study is so valuable, is that it gives a verification of the problem statement. The trans- The authors remark that ‘‘However, in comparison lation of a problem into a measurable characteristic, to other approaches, in SS, the use of experimentation the related measurement procedure, and the posed is subordinated to observational investigations’’ specifications are in many projects nothing more (p. 12). Again, this position limits the applicability of than intelligent guesses. The process capability study the approach. There are many improvement projects verifies that the problem definition actually manages Downloaded By: [Baez, Pablo] At: 20:39 8 January 2008 in which the focus of the project is the identification of the dominant cause of the problem. However, inmany projects identification of dominant causes is one thing, but finding out the exact relationship withthe quality characteristic under study is a second.
Jeroen de Mast is associate professor at the Univer- Experimentation is the cornerstone for learning about sity of Amsterdam, and senior consultant at the Insti- relationships; obervational data are typically less tute for Business and Industrial statistics of the suited for this purpose. The project that was given as University of Amsterdam (IBIS UvA). Ronald Does an example in our point 3.3 may illustrate this point.
is professor at the University of Amsterdam, andmanaging director of IBIS UvA.
‘‘For variation reduction problems, using families De Mast, J. (2003). Quality improvement from the viewpoint of statistical of variation and the method of elimination is a method. Quality and Reliability Engineering International, 19(4):255–264.
De Mast, J., Bergman, M. (2006). Hypothesis generation in quality more effective way to partition the causes than is improvement projects: Approaches for exploratory studies. Quality the classical Statistical Process Control (SPC) division and Reliability Engineering International, 22(7):839–850.
into common and special causes’’ (p. 7).
De Mast, J., Does, R. J. M. M., De Koning, H. (2006). Lean Six Sigma for Service and Healthcare. Alphen aan de Rijn, The Netherlands: Divisions of influence factors, causes or types of variation are only sensible in their context. The com- Maher, P. (1998). Inductive Inference. Routledae Encyclopedia of Philosophy, E. Craig (Ed.), Vol. 4, Routledge, London.
mon causes=special causes distinction was intro- Nickles, T. (1998). Discovery, Logic of. Routledge Encyclopedia of duced in the context of on-line process monitoring, Philosophy, E. Craig (Ed.), Vol. 3, Routledge, London.

Source: http://mail.ibisuva.nl/assets/files/discussion-shainin.pdf