Chairperson's Message

by Lynne Hare

Five o'clock this morning, I took my son to the airport. He is going to Colorado Springs to visit his grandparents and play golf with his grandfather who is 83 and shoots lower than his age. I could tell by the bleary eyed look of the agent that Tim's clubs would end up in Omaha.

Then one Skycap said to another, "Man, you got no business walking around that ug-ly!" And I lightened up.

Those incidents made me think about quality and about the Statistics Division reason for existence. That's a bit far fetched, . know, but hear me out.

If you have a pulse you know about the state of our economy and the importance of productivity and quality issues. (The pro-blem with the GNP is the MBA and the BMW) You may even have pushed for a Total Quality Management philosphy in your own organization — being careful, of course, to be a champion but not a martyr. Have you been as effective as you would like? How do you get management's attention without disembowel-ing yourself outside the CEO's office?

Well, I'm glad you asked that question because I think the Statistics Division can help. The Statistics Division's reason for existence is to help. We have no business walking around this ugly if we don't help. And all of our leadership since Bill Hunter have worked to assure that the Division presents unique growth opportunities to its membership and to anyone else who will listen.

The growth opportunity in this case is the Statistics Division sponsored short course by Dr. Ron Snee. [See announcement in this Newsletter] Ron's a guy who has been t" the city and seen the elephant. He has struggled successfully with implemen-tation of TQM at Du Pont, and be is willing to share his ex-periences and insights. I can't think of anyone more highly qualified.

We will continue to try to meet our customer's (your) needs by providing growth opportunities such as Ron's course. We're

rking on a special "Haw To" booklet offer, a membership arvey, the Annual Quality Congress Program, and another short

course, just to name a few. So I went home and told my dogs about all these exciting

developments. They're the only people in the house who love me. Did you know that dogs can sleep with their eyes open?

BEYOND THE TAGUCHI CLASSICAL WARS

Or How The Alpha - Beta Wars Open A Door of Reconciliation

by Bill Goodenow

Rock Valley College - Rockford, IL

The battle lines remain drawn: Classical, traditional, western or, as George Box now prefers, "mainstream" statisticians on one side, and Taguchi proponents on the other. Even with a widening DMZ, as increasing numbers of quality practitioners seek ways to have the best of both sides, confusion still prevails when the opposing camps try to discuss their differences.

The paper by Joseph Matar and Robert Lochner, 1988 — ASQC Quality Congress Transactions — Dallas, is a good ex-ample of the growning DMZ school. Titled ENHANCED TAGUCHI METHODS USING WESTERN DESIGNS, its authors propose that Taguchi's concept of robust design can be improved by using foldover models to increase flexibility in estimating main effects and their interaction.

This combining of eastern philosophy and western methods has strong appeal to those of us who sense that both sides may be blinded by their respective traditions or recent conversions. After all, isn't synthesis the byword today in all fields, including politics, religion and science?

In spite of the appeal of synthesis, something seems to be vaguely out of place in these efforts to "enhance" Taguchi's methods. It's almost as if our rational mind is trying to improve that which can only be understood intuitively, and in most cases doesn't need improvement. (Did I say that?)

I had been wrestling/wresting with this paradox for several weeks when I had a close encounter with Beth Propst's article, THE ALPHA - BETA WARS, in the Spring 1988 issue of this Newsletter. Subsequent conversations with her and Bob Lochner have helped me, if no one else, put the Alpha-Beta Wars to rest and resolve one aspect of the TAGUCHI - CLASSICAL WARS, as well.

My intention in this article is to highlight a plausible connec-tion between the two wars and how a deeper understanding of Taguchi's philosopy might bring both sides closer to mutual understanding. Future articles could explore other fronts in the Taguchi-Classical Wars, such as the DEDUCTIVE-

Cont. on page 2...

PAGE 2

STATISTICS DIVISION NEWSLE FALL, 1988

Coro. from page 1... INDUCTIVE DEBATE, the PASSIVE-ACTIVE DELEMA and the thorny ANALYTIC-ENUMERATIVE PARADIGMS, if I haven't had my word (thought) processor tarred and feathered by then.

Most QC/SPC analysts will agree with Ms. Propst's contention that we have tend-ed to focus on Alpha and let Beta fall where it may. 'Taguchi advocates are also quick to point out western preoccupation with Alpha, allowing us to surmise that their highly fractionated factorials must somehow be Beta-oriented. When we realize that their statistics are just as Alpha-oriented as ours, we may feel nettled by what has all appearances of the proverbial "primrose path." We must ask, however, have we been led astray by Taguchi or our possible predisposition to A PRIORI statistical rigor.

I propose that the Taguchi approach to robust design is inherently Beta-oriented, not by its statistics but by its design engineering philosophy. The selection of

- process/product parameters which are on the flat part of the response curve automatically assures insensitivity to noise factors; i.e., robustness. This, in effect, is the same thing as having a low Beta risk built into the process, so that we no longer need to be concerned about Alpha/Beta trade-offs further on down the line. In other words, why should we worry about the pro-bability of falsely accepting defective pro-duct when there is no way that it can be produced. Thus it appears that Taguchi has ended the ALPHA - BETA WARS, open-ed a door of reconciliation in our war with him and further demonstrated for us that...

"There is no way to tell whether the pat-terns extracted by the right hemisphere are real without subjecting them to left-hemisphere scrutiny On the other hand, mere critical thinking, without creative and intuitive insights, without the search for new patterns, is sterile and doomed. To solve complex problems in changing cir-cumstances requires the activity of both cerebral hemispheres; The path to the future lies through the corpus callosum,"

Carl Sagan THE DRAGONS OF EDAN

Could it be that Taguchi also represents the corpus callosurn between the INDUC-TIVE and DEDUCTIVE approach to problem-solving? Check the next issue to find out.

IMPLEMENTATION OF TOTAL QUALITY

A STATISTICS DIVISION SHORT COURSE

Presented by DR. RONALD D. SNEE

• October 19, 1988

• Meadowlands Sheraton Hotel

East Rutherford NJ 07073 •

(Prior to the 32nd Annual Fall Technical Conference)

For Registration: Dr. R. B. Barrett Morton Thiokol, Inc. 335 MacLean Boulevard Patterson, NJ 07504

Ron Snee, an ASQC Shewhart Medalist, will provide an overview of the implemen-tation of total quality including the ingre-dients of total quality, systems for manag-ing total quality, and tools useful for achiev-ing total quality.

He will show how the concepts and methods of quality technology, statistics. and organizational effectiveness are team-ed to initiate and sustain total quality. Topic such as Process Analysis will be addressed.

Management and technical personnel in-volved in the total quality process will benefit from the material presented and the associated discussions.

Ron Snee is Manager, Quality Systems, for the Project Engineering Division of Du Pont's Engineering Department. His previous assignments at Du Pont have in-cluded Statistical Consultant; Supervisior of Consultants in the fields of statistics, software engineering, plant engineering, environmental engineering; and Senior Consultant, Corporate R&D Planning. His contributions to quality technology and ap-plied statistics have received several awards and prizes from ASQC, including the Shewhart Medal. He is also a Fellow of ASQC, ASA, and AAAS.

SOL + REAL-TIME RELATIONAL DATABASE MANAGEMENT SYSTF"c• AVAILABLE FOR PC/A1 This news release showed up in my mail. I know nothing about this software or these people, but here it is for what it's worth.

(Editor) by Greg Lawson

Advanced MicroSolutions has announc-ed a version of SQL+ the real-time rela-tional database management system, for the PC/AT and compatibles running the HP BASIC Coprocessor. SQL+ now is sup-ported oa the HP 9000 Series 200/300 workstations and PC's. SQL+ is available today. Modules are also available for SQC, SPC and Data Acquisi-tion and Control. For more information on SQL+ and other CIM products and ser-vices from Advanced MicroSolutions, con-tact Greg Lawson at (415) 365-9880.

COMMITTEE ON AREAS OF APPLICATION

The Section on Quality & Productivi-ty has formed a new committee to address areas of application. In general, the mis-sion has been defined as:

Take Statistical methodology and quality philosphy which as already been developed and translate it into language and techni-ques specific to each area of application.

So far, three areas have been identified — Education, Manufacturing, and Research Development & Engineering. A fourth area, concerning applications in Latin America, has also been identified. This group will be working on awareness case studies, publications and presenting papers in Spanish at the 1990 WINTER CONFERENCE.

A related, but separate committee ad-dressing the statistics and quality in Health Care has also been formed.

Right now all of these groups are look-ing for input from the general public on the direction they should be taking Their next step will be to tailor the mission statement-select key objectives, and develop plan: achieve the objectives. Please contact then, with your ideas. (They are also looking for volunteers to help!)

Cont. on page 3...

FALL, 1988

Cont. from page 2... The people to contact are:

EDUCATION: Reider Peterson Math Department So. Oregon State College Ashland, OR 97520

MANUFACTURING: Paige Miller Quality Consultant Eastman Kodak Company 4165 Kodak Park Rochester, NY 14650-1814 (716) 477-2946

RESEARCH, DEVELOPMENT AND ENGINEERING:

James L. Hansen Development Scientist Union Carbide Corporation P.O. Box 8361, KTechnical Center South Charleston, WV 25303 (304) 747-4862

LATIN AMERICA: Dr. Victor Aguirre Torres Centro de Invesigacion en

Maternaticas Apartado Postal 402 36000 Guanajuato, Gto. (473) 2-58-05 or 2-02-58

HEALTH CARE: Randy Spoeri Senior Corporate Statistician Human Inc. The Humana Building 500 West Main Street P.O. Box 1438 Louisville, KY 40201-1438

Anyone with comments or suggestions about any other areas of application should contact:

Beth Propst PMT (612) 893-0313 732 Parkview Avenue Rockford, IL 61107

Postscript on Malcolm Baidrige National

Quality Award Since several of you called to ask, I

though the information would be of general interest. Those who are interested in ap-plying should write to:

Malcolm Baidrige National Quality Award

National Bureau of Standards Gaithersburg, MD 20899

They will dispense applications on request.

SHORT COURSES FOR 1988-89

Seminars offered by University of Wisconsin - Center for Quality & Pro-ductivity Improvement

An Explanation and Critique of Taguchi's Contributions to

Quality Improvement October 17-20, 1988

May 22-25, 1989 September 11-15, 1989

Instructors: George Box, Soren Bisgaard, and Conrad Fung of the Center for Quality and Productivity Improvement.

Kaizen: The Key to Japan's Competitive Success November 17-18, 1988

Instructors: Masaaki Imai and associates from the Cambridge Corporation; and Soren Bisgaard and Conrad Fung of the Center for Quality and Productivity Improvement.

Designing Industrial Experiments: The Engineer's Key to Quality

December 12-16, 1988 April 24-28, 1989

Instructors: George Box, Soren Bisgaard, and Conrad Fung of the Center for Quality and Productivity Improvement.

Dynamic Statistical Process Control: Going Beyond Shewhart

Instructor: J Stuart Hunter, Professor Emeritus, Princeton University, In Madison, Wisconsin; to be scheduled.

1989 William G. Hunter Conference on Quality

April 19-21, 1989 Plans are underway for the 1989 William

G. Hunter Conference on Quality, which will take place April 19-21, 1989 in Madison, Wisconsin. The conference is be-ing organized by the Madison Area Quali-ty Improvement Network (MAQIN). For information, or to volunteer your help, con-tact Barb Hummel, MAQIN, 1010 Mound Street, Madison WI 52715, (608) 267-6581.

Thomas J. Snodgrass Engineering Professional Development 432 North Lake Street Madison, Wisconsin 53706 (608) 263-3371 or (800) 262-6243 (800-362-3020 in Wisconsin)

SPC Really Stands For . . The response to my request for your ver-

sions of what the letters SPC really stand for has been gratifying. As promised, I am printing all contributions. In fact, many of the letters were so terrific I am printing all or portions thereof. Those of you who didn't get your contributions in this issue — it's not too late. I will continue to print suggestions as I receive them. STOP POSTING CHARTS

Rich Postrosny Midas International Corporation

STOP PRODUCT CONTROL START PROCESS CONTROL STAY PURPOSEFULLY CENTERED START, PROCEED, CONTINUE

Bill Goodenow, Instructor Rock Valley College

SIGNIFICANT PROFITABILITY CONFORMATION

Kevin J. Stoneback QA Engineer

American Meter Company

SOLVING PROBLEMS CONTINUOUSLY

Gene L. Coleman . Quality Consultant

SOLVING PROBLEMS COOPERATIVELY

James R. Gill Consultant

STABLE — PREDICTABLE -CONSISTENT

Jerold W. Ferree Q.A. Manager

SKF Bearing Industries

STATISTICAL PROCESS CONTROL SERVICE POTENTIAL

CUSTOMERS STOP POSSIBLE COMPLAINTS SHOWING PEOPLE CAN SHOWING PROFITS CLIMB SHOWING PEOPLE CHOICES

Teresa Snowden SPC Technician

The Sherwin-Williams Company

STATISTICS DIVISION NEWSLETTER

PAGE 3

PAGE 4

STATISTICS DIVISION NEWSLE1

FALL, 1988

Editor's Corner This issue marks my fourth issue as

editor, and I have finally gotten the process under control! It's even starting to be fun. The best part is the response I have receiv-ed from you — the readers.

Previous editors tell stories of the dif-ficulties they have had trying to wring ar- ticles out of the other members of the Statistics Division Board. My first ex-periences were similar. First of all, you play telephone tag with these people. A couple of times it got so bad that I left the simple message "Tag — you're it!" I am getting to know the secretaries of the board members intimately.

In all fairness to the members of the board — these people are over-committed. In ASQC as in any other volunteer Organization, it seems that the same peo- ple volunteer over and over again. Just as, in your community, the same people who coach the Little League team are on the United Fund, the same people who are scoutmasters are members of the Booster Club, so in ASQC. The members of the

' boards of the division often hold multiple responsibilities on that board, as well as be- ing active in their local sections, other areas of ASQC, and other professional organiza-tions, such as the American Statistical Association. Not only that, they often are speakers at meetings and conferences or write papers for journals such as Quality Progress.

This sounds like an accolade of my fellow board members, but it isn't. It is a tribute to you, the readers, whose contribu- tions over the last few months have allow- ed me to short cut the process of trying to wring blood out of already wrung turnips. Letters, articles, responses to my queries, all have made ray job easier. I also look for- ward to reading my mail and thus meeting new colleagues. Comments from other people indicate that they find your con-tributions useful and enjoyable. Please con-tinue to write. We all thank you.

I have a special request to make of you. (Notice how I have at least one of these every issue?) This one has to do with the Statistics Division as a supplier to you, the membership. Let's call this a brainstorming-by-mail session. What kinds of things would you like to see from the Statistics Division in the next five years? What are your needs as Quality profes-sionals? Remember, anything goes in a brainstorming session. Look at all areas

courses, books, recommendations, con-ferences. I won't guarantee that we will act on all suggestions, but we need to know what you need if we are ever going to meet your needs.

The deadline for submissions for the Winter issue is 30 November, so let me hear from you by then.

Beth Propst Editor

Dear Beth: What, indeed, does SPC stand for?

Perhaps "so poorly communicated". Inter-preting it to be just Xbar - R charts would be similar to calling JIT an inventory con-trol technique. Both are as much manage-ment philosophies as they are tools.

In this light, I'd like to propose that we call it "Strategically Planned Change", as it recognizes that the impact when using the Deming philosophy of SPC will be a different way of managing the business. We will no longer react to each and every bit of information as though it is an exception, but will instead ensure that the system, responsible for the majority of errors, receives our attention when appropriate.

If we must leave the statistics in it, perhaps "Statistically Prioritized Changes" would be appropriate. That may prevent the pendulum from swinging to far to the philosophical side, as I may have done once during a training session. An engineer decided that giving their operators as much voice in their work as was recommended would result in "Social Problem Control".

Duke W. Okes Consultant

Aplomet Applied Logical Methods

Dear Beth: What does SPC really stand for? Our first process/product to have the

benefit of on line use of SPC literally shut down the process. Our production person-nel quickly came up with SPC stands for "Stop Production Completely". Our response was that we are not stopping anything that should not have been started in the first place.

The good news is that with much effort by all, problems were quickly resolved, quality is substantially improved and now a year later the operator complains of boredom as rarelydoes anything go wrong.

Cordially, W. J. Couch

Durable Wire, Inc.

Dear Beth, We at Brookside strongly agree with your

comments about the intimidation most peo-ple feel when confronted with the let'-SPC. As in all corporations since. dedicated to spreading the word of SPC, the management here at Brookside Group is doing everything possible to help employees understand and appreciate the value of Statistical Process Control.

Here is our aphorism for the letters SPC: SUPERIOR PRODUCTIVITY

CAPABILITY Tony Moffett

Sr. Quality Technician Brookside Group

Beth, I received my Statistics Division newsletter today. Enjoyed your article. Here's my contribution, although I can't claim it. It was developed by Ken Morris, QC Manager of our Mobile Products Divi-sion. We use it for our heading on our bulleting board. SMART PEOPLE CARE.

We post our Quality related articles, charts, memos, and even a U.S. map show-ing all of our customer plant locations.

I hope to attend this year's conference in New Jersey. Hope to see you then.

Kris Kr Atwood Automotive,

How to Order How-To Booklets

In the last issue of this newsletter we featured a "Call For Authors" for the Statistics Division's How To Booklet series. Since it was a call for authors, we neglected to include information on how to order these books. Silly us. Judging from the number of phone calls I received, this was a serious oversight. (Sometimes rework is necessary.) You can order these booklets in one of 3 ways. 1. Call ASQC and place your order. Their special number of placing orders is (800) 952-6587. In order to place an order over the phone you must have either a P.O. number or a credit card number. 2. Alternatively, you can use the order b • flies which appear in the Quality Press Bookshelf section of Quality Progress. (It 'malty appears near the end of the magazine.) Remember, the list appear three parts in consecutive months. 3. If you have received a copy of "ASQC Publications Catalog" you may use the order blank in there.

CONV011404IL TIN WV SAAUIY Worlatld I•EALMO

ON 004 -Amgen- YLN1410111 7.4441 104:ffluAnow mom

ol•FOP•••••om FROM VOLL04444011041101.0LT 4401 DATA OTAN040414.044ny Tr40■ 44010 ALTIOTTATTO TIM MN

faTA.0.4.4

Oft WE MOM] swat wE04dgS SIN- COMPUTERIZID RIECIUMES AL ICITY 10. MIPLIMENTA 34444TV No la raliffir•riCi. n:Nd larrescilr 1.001 117•••1

• •IJUI.7

CONCEPOI will' 6-10•• pCsitalai 4011.1 0214.4.1414 TICAL EFFICIENCY 14414.14 PAD 1•0•101.(14

THE CHALLENOL 05401 AND COM44441 THE KEY 11401041100 viAl wLL OPIVT °TAUT, 0440.1 1141

One of the consequences of the new pas-sion for quality improvement is that it is all-pervasive. We become very sensitive to quality in our personal lives, and are in-tolerant of indifferent service when we ex-perience it in, say, air travel, a doctor's of-fice, or in our youngsters' schools. And, from these experiences, we resolve to do continuously better when we are the pro-viders — rather than the receivers — of goods and services. (The desire for conti-nuing quality improvement has also im-pacted this paper. I have gone through various versions in a continuing effort for quality improvement. The re::.crs may rightfully question the value t.e final product — but they can, at least, . assured that it is better than some of the earlier versions.)

"What is more important — the non-technical challenges, summarized in Table 2, or the technical ones of Table 3?" I pos-ed this question to the 205 members of the American Statistical Association's Com-mittee on Quality and Productivity — not exactly a random sample of quality practi-tioners, but yet a good source of knowledgeable opinion. Of the 107 respondents, 77 % indicated that the primary challenges were non-technical. Moreover, an industry-based subset of those surveyed had an even stronger disposition to the non-technical (84 %). In-deed, only 10% of the total group and 4% of the industry-based group felt technical issues were nac7:: important than non- technical ones. of these responses from statistic, ,,le might only conjec-

respond. As the "essay part" of the survey, I ask-

ed the 107 respondents to state what they regard to be die major challenges. And, as expected, I got about 107 different opinions. I have compiled these and would be pleas-ed to make them available to interested readers.

What follows is the 108th opinion, which has benefited much, and taken generous-ly, from the 107 that preceded it.

NON-TECHNICAL CHALLENGES Various gurus, such as (in alphabetical

order) Crosby, Deming, Feigenbaum, Golomski, and Juran, have spoken elo-quently about the non-technical challenges. So have many of the previous Youden Ad- • dressers. I have little to add to their thoughts about the importance of manage-ment commitment, SkOrker involvement, the need for a team effort, etc. Thus, I will try to add only a few specific observations from my own experience.

First, we need recognize that we have limited resources — there are just a few of us with much to do. Therefore, we need Pareto chart our activities to identify and concentrate on those with the biggest poten-tial payoffs (even though they may not be the ones that are the most challenging technically).

Next, we must recognize the paramount importance of seeking out productive part-ners — ones who are receptive to new ideas, and who like ourselves, are activists. These partners should complement our own knowledge and provide us an improv-ed understanding of the physical background of the problem. They, too, need be team players, and must have the clout to get things done. From hard ex-perience, I know we can accomplish little in a vacuum. Without effective partners, the chances of success are remote — irrespec-tive of how valiant our efforts!

We need to play an active role. This re-

M.M..111.4 A.O. L7 At111.711LJ 11....1 V 1011../11 11.C. TT 11:.1%

Statistics - Aided Manufacturing: A Look into the Future 1987 W. J. Youden Memorial Address

31st Annual Fall Technical Conference Chemical and Process Industries Division, A.S.Q.C.

Statistics Division, A.S.Q.C. Section on Physical & Engineering Sciences, A.S.A.

Gerald J. Hahn General Electric Corporate Research and Development

TABLE 3 Over seven years have passed since Ed THE CHANGING ENVIRONMENT TECHNICAL ture how a less technical group might

Deming challenged the nation on prime-time television to adopt a new vision in the now famous NBC White Paper, "If Japan Can—Why Can't We?" Much has happen-ed since then. I will review the changing environment, discuss some major technical and non-technical challenges, comment on how well we are meeting these challenges, and state my view of the statistician's role in all of this. As a starting point, I would like to recall some comments by Jack Youden, shown in Table 1. These are as valid today as when they were first written—over 30 years ago.

TABLE 1 A MESSAGE FROM THE PAST

• THE RESPONSIBILITY FOR ADOPTING STATISTICAL TECHNIQUES WILL OFTEN REST ON wfxvIDUAUI WHO ANON RELATIVELY LITTLE ABOUT THESE TECHNIQUES.

THERE IS ONE SURE WAY FOR A COMPANY TO ACQUIRE A WELL-TRAINED STATISTICIAN. SELECT A _ (PROMISING) ENGINEER OR SCIENTIST - AND SEND TO A STATISTICAL scHooi. FOR A YEAR.

• MOST STATISTICAL WURES ARE Curie SIMPLE AND Nuummay ACCEPTABLE ONCE THEY HAVE BEEN POINTED OUT IN A PARTICULAR CONTEXT

• A GOOD STATISTICAL DESIGN IS LIKE A BLUEPRINT FOR A CONSTRUC. 710N .1011

• STATISTICAL TECHNIQUES ARE MOST PRODUCTIVE WHEN INCOR-PORATED INTO PROJECTS AT THE PLANNING STAGE.

• MUCH OF THE IMPACT OF A STATISTICAL *sway CONTROL CHART DEPENDS UPON ITS PRESENTATION.

• INDUSTRIAL APPLICATION OF STATISTICAL DESIGN REQUIRES MORE ACTIVE SUPPORT FROM MANAGEMENT.

SOURCE. STATISTICAL DESIGN: COLLECTION OF ARTICLES IN INDUSTRIAL AND ENGINEERING CHEMISTRY. !NMI YOUDEN. 1504.1901.

THE CHANGING ENVIRONMENT

Quality control practitioners need no reminder of the new environment. I have tried to summarize its major non-technical and technical aspects in Tables 2 and 3.

TABLE 2 THE CHANGING ENVIRONMENT: NONTECHNICAL

CONVIUSTVONAL W440,

TN( 01.01 0400

0400 044600 441CTIOr TO COLL 01A1 'LP 14301- CC 10 PRIPfloqr of..PECTS 011,01 YO OVICOUCr 0.1.1•10•0•13

440011 OC DAP 6401100f-LE0 IT TOT 1110404000+114

0440101010

AT "NAT A04 -TI, roAITY W ForSHED CONTIMIKKOLN THROLICINCIUT macoUCt ISIENomm, 100.1440

01A100 APIO 00004

11,-••••••••• 0•6•Triv•A• !mufti

OFTEN “16.1.1 COTIPUDI

T1•11•110.• •••ruFACII•IWO 00001

ALL 00001 AMO I/JIVTOL1

A 044 OUT OF 44 WV VoiCole urousuaio A O 0E 41401444 10 041041A HIEL4 -44LE coomocousas 1611•10/11.0 040Quali a.w0 IMOCRICITAry

PAGE 6

STATISTICS DIVISION NEWSLEI i ER

TALL, 1965

quires us to go out into the factory and "kick the tires," as one of my (many) past managers put it. We must understand the real problem— including the political ramifications. This may not necessarily be the same as the problem which our client might describe. Helping define the problem requires inputs from many people, in-cluding management. It is something that can rarely be done by telephone — and, certainly not by just talking to other statisticians.

We need be concerned with furthering quality improvement, irrespective of the degree to which it involves statistics. And we need capitalize on all opportunities to do an effective selling job. Tom Boardman urges that we always have available a 30-second "elevator speech" for use when we ride the elevator with our chief ex-ecutive officer, and are asked, "Well, what's new in the number game?" That's our opportunity to promote, in understan-dable terms, significant opportunities for quality improvement — and their an-ticipated payoffs.

Also we need remember, that training in statistical process control is just a starting point — not an end goal. It helps provide needed background to practitioners and can lead to some useful visibility for us, but the payoff is in properly applying the ideas -not just talking about them.

As Deming and his associates have made clear, we must understand the limitations of traditional statistical inference. I have

little use for hypothesis testing. Few if any, processes are ever created exactly equal, and whether or not we establish a statistically significant difference between them usually depends greatly on the size of the available samples. It is almost always more meaningful to make interval statements than to do significance tests. However, let's not get carried away by these either. We must remember Deming's warn-ing about the difference between enumerative and analytic studies. In an enumerative study, we may sample a batch and draw conclusions about that particular batch. However, much of the time that's not the real problem. What we have is a pro-totype batch (of limited interest, per se) and we really wish to draw conclusions about future product. But many things will change in the future. So the statistical evaluation for the current batch tells only part of the story. As a result, statistical in-tervals provide only optimistic bounds (under the assumption that the past can be

used to represent the future). We also need to improve the introduc-

tory statistics courses to which our clients — past, current, and future — are sub-jected. As currently constituted, these often constitute one big strike against us. We fail to recognize, in our own enthusiasm, that some people might even think of statistics as boring! We must show the real world ap-plications that make statistics exciting. As I have tried to indicate elsewhere [Hahn (1986)], we need to focus on the importance of properly planning investigations. We should emphasize graphical methods and basic concepts, and reduce the emphasis on formal methodolody and mathematics.

We must make it easier for others to use statistical concepts effectively. Many of us are reluctant to do this. However, the ready availability of canned statistical software, gives us no choice. We can learn from Japan, where, I am told, everybody in in-dustry understands statistics. For example, computer-aided experimental design is at-tracting much interest because it claims to make experimental planning easier, and removes dependence on the statistician. Since such programs are here to stay, we must steer our clients to the right ones for their needs, point out their limitations, and help make the programs more robust against misuse.

We must also reverse our reputation as poor communicators. One important part of communicating is listening to our customers' needs and problems. Then, we must provide short, clear answers in English — not statistical jargon. We need to play the game on our clients' turf and cannot expect them to play it on ours. Moreover, the value of our good ideas, though self-evident to us, may not be so to our clients. When we write a 20-page report on our findings, let's make sure that we include, on the first page, an executive summary of what's really important! (This might even generate interest in the remain-ing 19 pages!)

Yes! We do need to be technically knowledgeable. However, a good understanding of statistics is a necessary, but not sufficient, attribute of a successful applied statistician.

SOME TECHNICAL CHALLENGES I would now like to make various com-

ments, some provocative, on key technical challenges to meet the needs for continu-ing quality improvement and to narrow the gap between theory and practice.

BUILD ON TAGUC711 FERVOR Taguchi has done us a great service by

(among other things) focusing interest on the planning of investigations (more for-mally referred to as the design of ex-periments). He has emphasized the imp( tance of finding those operating conditions that are most robust to variability — both during production and during the subse-quent product use. A natural consequence of his advocacy of minimising the (squared) deviation from target is the need to use measurement (or variable), rather than at-tribute, data. In this age of computers, it is often absurd to reduce quantitative measurements into go-no go form.

Taguchi has also further popularized the use of fractional factorial designs, albeit under a different name (orthogonal arrays), and in a slightly different form. We may quarrel with his de-emphasis of interac-tions. However, I look at highly frac-tionated plans as a means of searching the experimental space to find the sub-region with the biggest payback opportunities -and then exploring this sub-region fruther — rather than that of estimating effects or terms in a model.

We may disagree with some of the specifics of the Taguchi approach, such as the use of the signal-to-noise ratio, a some of his analyses. However, we need throw out the proverbial baby with the bathwater. As George Box and others have pointed out, we should take advantage of the useful ideas that Taguchi has provided us, and accept the challenge to make im-provements where appropriate. Also, Taguchi and his associates surely provide us a good lesson on how to get the atten-tion of our customers.

ADDRESS THE REALITIES OF STATISTICAL PROCESS CONTROL

The proper implementation of statistical process control provides another major technical challenge. In practice, few pro-cesses are exactly in statistical control. However, some come closer than others. Thus, whether or not a process is deemed to be in statistical control, depends not only on the process itself, but also on such ex-traneous factors as the type of control chart used, the group size, the sampling frequen-cy, and how common cause variability is estimated. Moreover, the differentiation between "common cause" and "spec cause" sources of variability is not as elm. cut as most texts seem to suggest. (This is why one sometimes fords that when one

STATISTICS DIVISION NEWSLEI IER PAGE 7 FALL, 1988

succeeds in removing a special cause of variation, the common cause variation is also reduced.) For example, we recently dealt with measurements on groups of 5 - nits sampled consecutively at the end of

production process. The variability among units within such groups. at first glance, seems like a clearcut example of common cause variation. However, in look-ing further, we found that these five units were really fed by a number of parallel machines upstream, each of which, in turn, was sourced by a number of different in-coming material lots. To obtain a more ap-propriate estimate of common cause variability, we proposed a specially plann-ed investigation that involved taking con-secutive units from the same upstream machine, and the same batch of raw material, as well as the same final machine. Not unexpectedly, this resulted in ap-preciably less "common cause" variabili-ty than that observed before.

There are also some further technical complications with which we need be con-cerned in applying statistical process con-trol, especially in the chemical and process industries. Applications in these industries differ in many ways from the parts in-

,---dustries for which the Shewart methods ere originally developed. Some of these

differences, such as auto-correlations, are the consequence of automated (and essen-tially continuous) process measurements. These auto-correlations have always been there for continuous processes. When we take measurements every four hours, they are often of little concern. However, when an updated reading is obtained every se-cond, they generally are.

I feel, moreover, that — especially in chemical processes — there are situations where legitimate batch-to-batch variabili-ty is present. In such cases, this should be included as part of the common cause variation [see Hahn and Cockrum (1987)]. Indeed,, sometimes, we have a hierarchy of common cause effects with batch-to-batch variation just one part of that hierarchy.

Also, we need recognize the inherent dif-ference between the usual attribute control charts and variable control charts. In par- ticular, for a p chart and a c chart, the values of p and c define the common cause variability (because the distribution mean

------'=:!fines the distribution variation). In con-ACt, for an X-bar chart, the data from the

process itself is generally used to separately estimate the process mean. and variation. This difference has some important prac-

tical consequences. The preceding comments are not meant

to put down the control chart. It is still one of the most important analytical tools available to the quality analyst, even though it may not be the simple cure-all that some advocates may suggest. My purpose is merely to promote sobriety and selectivi-ty in its use. One of the major virtues of the Shewhart chart is that it leads to a plot of observed data against time. So, instead of paying too much attention to the control limits — whose determination may be somewhat arbritrary — I recommend that we shift emphasis to assuring that the data are plotted in the most informative man-ner. This includes the capability of plotting the results over selected time windows of varying length, allowing side-by-side com-parisons of parallel processes, etc.

GUIDE THE SEARCH FOR ROOT CAUSES OF VARIATION

A major goal in the continuing effort to reduce process variability is to identify and then eliminate root causes of variation. This, however, requires that the right data are collected in the first place, We play an important up-front role by providing guidance to help make sure that the need-ed data are obtained. This includes a careful study of the process to determine when and where to take measurements, and how many measurements are needed. For example, for a chemical process, we recently desired to evaluate the apparent impact of raw materials, feeding into the hopper, on product performance. However, we found that the material measurements that were available really pertained to the material at incoming inspection and cor related poorly with those for the material in the hopper. Thus, statistical analyses are unlikely to bear fruit until more pertinent data is obtained.

We need also push for adequate "memory." Many systems retain informa-tion on what was happening during the past week, and even the past month, but the in-formation about that happened a year, or even a few months ago, is difficult to retrieve, or lost entirely.

However, such information, even in sum-mary form, is often important to have readily accessible.

Similarly, we should urge, wherever possible, that samples of current material and product be retained for potential future evaluation. On numerous occasions, changes in observed performance leads us

to question the reasons for the change; for example, are the changes in product per-formance due to raw material, process, or measurement error, or some combination of these? Assuming material properties do not change over time, it is often invaluable to have some past product against which the performance of current product can be compared directly. In most programs with which I have been involved, past product was unavailable. In many of these cases, the retention of even small amounts of such product and/or material could have paid handsome dividends.

We need also push for pro-activeness in searching for the root causes of variabili-ty. This means that we should not hesitate to propose special studies or experiments — either on-line or off-line — when this is what is necessary to get the needed in-formation. Such studies might be directed at measuring the "bottom-line" root cause variability, and at evaluating the impact on both average performance and variability of various process or materials variables. In on-line studies, especially in the chemical industry, we need to recognize the potential existence of the "master-chef' -the knowledgeable operator, who adjusts the process, based upon a review of the available data, current goals, and past ex-perience. This phenomenon can create havoc, if not properly recognized, in any statistical evaluations.

RESPOND TO THE NEEDS OF A COMPUTERIZED, DATA-RICH ENVIRONMENT

More and more factories are becoming fully automated, as computer-integrated manufacturing continues to play an ever-increasing role. This often means automated data gathering with essentially a continuous flow of process and perfor-mance data. In fact, for many processes, the data-gathering capabilities have far outstripped the ability to digest the infor-mation and to appropriately act thereon.

This presents a challenge if there has ever been one! The methodology developed for a data-poor environment is no longer relevant, but what do we have to put in its place? Clearly, time series methods will become of ever-increasing importance. So will the ability to develop incisive summary statistics and comparative graphical displays from vast volumes of multivariate data.

With essentially continuous and instan-

PAGE 8

STATISTICS DIVISION NEWSLLI 1 k.R

FALL, 1988

mucous data flow, our mission will more and more be to identify and help correct problems BEFORE they occur. And, as the volume of data multiplies with increased automation, the available staff to process it is often diminished! Thus, we will need systems that provide computer-aided alarms to trigger corrective action at various levels of the organization — from the process engineer to the plant manager. In this case, we had better properly con-trol the false alarm rate if we wish to main-tain credibility! This will require us to re-consider the relative merits of signaling an out of control situation (i.e., a statistically significant change from past performance) — something that is often easily detectable with large masses of data — versus mere-ly identifying a change that is of practical importance.

INTEGRATE SPC AND AUTOMATIC CONTROL THEORY

As MacGregor (1987) has pointed out, the statistical quality analyst and the automatic control theory specialist, rather than being oblivious of one another (or operating in a competitive mode), need unite to further the common goal of con-

• tinuous quality improvement! The term "statistical process control" is itself a misnomer, as Stu Hunter and others have pointed out. It might better be referred to as "statistical monitoring," since its major purpose is to signal significant changes from the past. In contrast, automatic con-trol theory is concerned with how to ad-just a process, often with the goal of com-ing as close as possible to its target value. This is frequently our real goal. In contrast, we have found that, at least in some ap-plications, process changes are made only when "out of control" performance has been demonstrated by a control chart. However, this may be far from an optimum (or even a good) approach when there is no direct cost for process adjustment, as is often the case for many chemical pro- cesses. Instead, how to best proceed depends upon such things as the underly- ing model, our knowledge of how changes in process conditions impact performance, process dyndamics, etc.

All of this has been long recognized [see Box, Jenkins and MacGregor (1974)]. However, it is more pertinent today than ever before in light of both our renewed in-terest in quality improvement, and the in-creased availability of automated data col-lection schemes.

Incidentally, classical statistical process control (or monitoring) has a place even when automatic control is used. In a re-cent application, dealing with a modem machining factory, we applied statistical process monitoring to the inputs and out-puts of an automatic process control scheme. This, among other things, helps assure that the automatic control continues to do its job, and help prevent it exceeding its capabilities.

GUIDE PRODUCT LIFE ANALYSIS Product Life and in-servicce reliability

are important quality performance characteristics that warrant our attention. Failure to recognize this can have disastrous consequences. However, product life analysis presents some technical challenges different from those en-countered in other applications. Foremost among these is the obvious fact that it takes time to measure product life. (Exactly how much time depends on the product. Thus, five years of operating experience can be obtained more rapidly on a product that is used only infrequently, such as a toaster, than, say, a steam turbine, which may be in almost constant use.) To obtain informa-tion faster, testing is frequently conducted under accelerated conditions, e.g., high temperature, increased voltage, etc. However, accelerated testing requires:

• An appropriate physical model to relate performance at the accelerated stress to normal in-service conditions,

• A proper test plan, • A valid analysis of the resulting data.

Moreover, the analysis of product life data — accelerated or not — is further complicated by the following:

• The normal distribution is often not an appropriate model for the random -varia-tion. Instead, the less familiar Weibull distribution, or some other specialized representation, may be required.

• The available data are often censored (due to unfailed units),

• It is frequently desired to estimate distribution percentiles or survival prob-abilities to a specified age.

None of these problems are insurmoun-table. However, they do require the use of special methods with which most engineers, and many statisticians, are not familiar. [See Hahn and Meeker (1984) for some books in this specialized area].

HOW ARE WE DOING? In the Survey of the ASA Committee on

Quality and Productivity, referred to earlier, I also asked the respondents characterize our progress on various fn The results, which speak for themselves, are summarized in Table 4. Again, I'd be pleased to provide further details upon request.

TABLE 4 SUMMARY OF FINDINGS

• SE% 190% AMONG INDUSTRIAL RESPONDENTS) FELT THM THERE IS A MEM COMMITMENT 10 GUALTry IMPACrif WENT Ni NORTH AMERICAN INDUSTRY TODAY

• 71I% (54%) FELT COMMITMENTAWS aniunA THAN 5 YEARS AGO

• 6356 (7o%) FELT STATISTICIANS' CONTRIBUTION IS MIXED

• 70% (TImI FELT STATISTICIANS CONTRIBUTION WAS GREATER THAN 5 YEARS AGO

• 39% pir91 FELT STATISTICIANS' CONTRIBUTION WAS GREATER THAN EXPECTED S YEARS AGO

• 37% CB%) FELT STATISTICIANS CONTRIBUTION WAS LESS THAN EX. PECIED 5 YEARS AGO

THE STATISTICIAN'S ROLE Many of us have faced what I like to call

"the Deming contradiction." Ed Deming talks fervently about the key role of statistics and statisticians. Yet, when we ex-amine his specific proposals, such as his 14 points and his "deadly sins," we might conclude that these have very little to with statistics, as we know it. And that, course, is the point. Dr. Deming calls for statisticians very different from those trained by conventional methods. Whether or not such persons are still really statisti-cians, is more a matter of definition than of essence. Clearly, Deming's concept are very much in line with the observation of the majority of our respondents that today's challenges tend to be more non-technical than technical.

Deming and his associates also claim that statisticians are uniquely suited to lead the transformation of the Western management style. I think that's somewhat of an overstatement. I don't think that I or most of my colleagues (even with different train-ing) have the knowledge or the hutzpah to tell management what their style should be. On the other hand, I do agree that we must look at the big picture and that we need be an important part of the quality improve-ment team.

What then are our major contributions? I have summarized the major areas, as I se.- them, in Table 5. My views are enca sulated in the words of one of the respondents to my survey, who said, "We don't hold all the cards, but we sure hold some pretty good ones!"

STATISTICS DIVISION NEWSLETTER PAGE 9 FALL, 1988

TABLE 5 THE STATISTICIAN'S ROLE: A PERSONAL VIEW

. FACING THE DEMING CONTRADICTION

— DEMING ET AL TALK ABOUT KEY ROLE OF STATISTICIANS

_ MANY OF DEMINGS CONCEPTS HAVE LITTLE TO DO infirm sunsTICS

DEMINGS VISION CALLS FOR A VERY DIFFERENT STATISTICIAN

. ARE WE U/NOUELY SUITED 10 LEAD --NE TRANSFORMATION OF AmERiCAN MANAGEMENT STYLEr

— PROBABLY NOT

— BUT WE CAN PLAY AN IMPORTANT ROLE ON THE OUAuTy IMPROVEMENT TEAM

CONCLUDING REMARKS The quest for quality is now an integral

part of our lives. It, and the needs of our customers, enters all aspects of our work. We have made some progress, and so has Western industry as a whole — but so too have our off-shore competitors! We still have a long way to go (especially when we realize that quality improvement is a never-ending goal). The non-technical challenges are of paramount importance. There are some significant technical challenges also, but we should not limit our attention to these. The value of our work will be measured by its long-term impact, and not by the sophistication of the underlying mathematics. The new computer-intensive environment provides many opportunities. As we move further from a data-poor to a data-rich environment, we will become more and more absorbed by the challenges of compressing data to gain information and of anticipating and correcting problems before they occur. The future indeed, looks exciting!

REFERENCES Box, G.E.P., Jenkins G.M., and

MacGregor, J.F. (1974), "Some Recent Advances in Forecasting and Control," Applied Statistics, 23, 158-179.

Hahn, G.J. (1986), "Improving Our Most Important Product," Editor's Invited Column, Institute of Mathematical Statistics Bulletin, 16, 354-358, November 1986.

Hahn, G.J. and Cockrum, M.B. (1987), "Adapting Control Charts to Meet Prac-tical Needs: A Chemical Processing Application:' J. of Applied Statistics, 14 (1), 35-52.

Hahn, G.J. and Meeker, Jr., W.Q. (1984), "An Engineer's Guide to Books on Statistics and Data Analysis," J. of Quality Technology, 16 (4), 196-218, Oc-tober 1984.

MacGregor, J.F. (1987), "Interfaces Be-tween Process Control and On-Line Statistical Process Control," CAST Newsletter, A . I .Ch. E.

ACKNOWLEDGEMENTS I would like to express my appreciation

for their numerous and significant inputs to my colleagues Fred Falun and Bill Tucker and to the 107 members of the ASA Committee on Quality and Productivity who responded to my Survey. Thanks are also due to Deborah Hopen for transcrib-ing and editing the manuscript.

Editor's Note: Deb Hopen is editor of the CARL Division Newsletter. The ASA Committee on Quality and Pro-ductivity referred to herein has now been approved for section status -see article elsewhere in this issue.

ASA QM' Update by David Fluharty, Chair

As indicated in the last issue of the Statistics Division Newsletter, each issue will contain a summary of recent activity of the American Statistical Association's Quality and Productivity Committee. In the last several months there have been three major developments:

The Committee has achieved perma-nent Section status within ASA (equivalent of an ASQC Division) with 1,200 members — quadrouple its previous membership. As a sec-tion, Q&P will be sponsoring ses-sions at major ASA meetings, par-ticularly the Joint Statistical Meetings in August.

The annual Joint Statistical Meetings in New Orleans provided an oppor-tunity for a number of people to become actively involved in work of the new Section's committees and for these committees to make plans for the coming year.

- Because of these huge influx of members and volunteers the primary locus of activity has shifted to the working committees described below. More reliance will be placed on ASA's Amstat News to provide members with ongoing information.

One of the most dramatic decisions taken was by the Education Committee which has narrowed its focus. It will 1) support efforts

of engineering educators who are trying to get statistical and quality tools into engineering education, 2) help to broaden the support for a Q&P component of the business school statistics curriculum, and 3) point out the common areas of interest between the two. This group's efforts will be designed to augment two major efforts in these areas, a) that of the Accreditation Board for Engineering and Technology which will sponsor a Conference on Statistics in Engineering Education January 11-13, 1989 in New Orleans, and b) the an-nual Conference on Making Statistics More Effective in Schools of Business.

The Research Committee will attempt to identify areas in quality and productivity where new research is needed, disseminate this information to statistics community to foster this research, and sponsor forums to bring researchers and practitioners together. This group is exploring the possibility of publishing an annual report of research highlights of the prior year.

With increasing complexity of processes and products (including software) improv-ing reliability becomes increasingly impor-tant. Unfortunately the tools of reliability analysis which can assist in this effort have not received as much recent attention as other quality technologies. The Reliabili-ty Committee will work for broader adop-tion of these methodologies.

The Awareness Committee has developed a draft of a brochure to spark the interest of managers and executives in the uses of statistical tools. It is also ex-ploring a means to make key printed material, such as major speeches by leading corporation executives, available. Finally, with the theme of quality and pro-ductivity, the 1990 ASA Winter Conference will provide an outstanding opportunity for increased awareness.

The Publications committee will be working with the Journal of Quality Technology on another special case study issue for January 1990. It is also in-vestigating the possibility of an annual award for a case study.

Finally, committees on specific applica-tion areas are being formed. The first is Health Care Delivery. Another group will attempt to improve communications on Q&P issues among Latin American statisti-cians. In addition, a general Applications Committee will serve to coordinate the in-terests of people in specific applications areas which do not yet have enough active

ASQC STATISTICS DIVISION 1988-89 OFFICERS

Chairman: Lynne B. Hare Thomas J. Lipton, Inc. 800 Sylvan Avenue Pnelewood Cliffs, NJ 07632 (201) 894-7441

Chairman-Elect: Steven P. Bailey E.I. du Pont de Nemours

& Co., Inc. Engineering Department 655 Paper Mill Road Newark, DE 19711-7515 (302) 366-2580

Secretary: Conrad A. Fung Center for Quality and

Productivity Improvement University of Wisconsin WARP/1109B 610 Walnut Street Madison, WI 53705 (608) 263-2652

Treasurer: Roger W. Hoer! Scott Paper Company Scott Plaza I Philadelphia, PA 19113 (215) 522-5011

1

PAGE 10

STATISTICS DIVISION NEWSLETTER

FALL, 1988

The ASQC Statistics Malan Nainies. ter is a publication of the Statistics Divi-sion of the American Society Or Quali-sy Cantrell.

All communications regarding this publication, EXCLUDING CHANGE OF ADDRESS, should be addressed to:

Annabeth Propel, Ecr.or ASQC Sods' ties Division Newsktur Process Managment Institute 732 Parkview Ave. Rockford, IL 61107 (612) 893-0313 Other communications relating to tic

Statistics Division of ASQC should be addressed to:

Lynne B. Hare, Chairman ASQC Statistics Division Thomas J. Lipton, Inc. BOO Sylvan Avenue Englewood Cliffs, NJ 07632 (201) 894-7441 AU communications regarding mem-

bership including change of address, should be addressed to:

American Society for Quality Control 310 West Wisconsin Avenue Milwaukee, WI 53203 (414) 272-8575

STATISTICS DIVISION AMERICAN SOCIETY FOR QUALITY CONTROL 732 Parkview Ave. Rockford, IL 61107

Non-ProAt °Immolation I US. Rafts

BUD Perak No. 942

Rockford, Mac&

ASOC STATISTICS DIVISION NEWSLETTER participants to be a stand-alone committee. These areas include. manufacturing, R&D, quality of education institutions, and government service agencies.

One way this new section will seek to help focus efforts where they are relevant is by interfacing closely with the Statistics Division - hence the regular appearance of an article in this newsletter. In the past, per-sonal contributions have been made by Division Board Members who happened to be members of ASA. In the future (start-ing now) we are actively soliciting input from non-ASA members. Much of the work being done by committees transcends ASA, ASQC, or any other professional society. Why duplicate efforts to maintain an essence of territorialism which does not exist?

All input is welcome, whether ideas or assistance. However, assistance is especial-ly welcome. Individuals interested in working on a specific committee should write ASA Q&P's secretary, Jack Wood, Ford Motor Company, P. 0. Box 2053, EEE Building Mail Drop 57, Dearborn, MI 48121