Está en la página 1de 24

Volume 16, No.

Winter, 1996

Chairs Message
by Nancy Belunis
The Statistics
Division was
awarded division
status over 16
years ago in the
summer of 1979.
One of the
major projects of
the new Division
was the development of The Basic References in
Quality Control: Statistical
Techniques, also known as the How
To series. The aim of the series is
to present the latest statistical techniques in a form which is easily followed by the quality control practitioner so that these procedures can be
readily applied to solve industrial
quality problems. The first volume
in this series, How To Analyze Data
with Simple Plots, appeared in 1978
while the Division was still a
Technical Committee. At present, the
series contains 16 volumes and covers
topics like sampling, mixture experiments, reliability data, outlier testing
and sequential methods. In the
series history there have only been
four editors, Ed Dudewicz, Sam
Shapiro, John Cornell and Ed
Mykytka. With the assistance of these
editors, the authors have driven the
creation of a successful series.
Continued on page 3

Basic Reference in Quality Control:


Statistical Techniques1
Volume 1: How To Analyze Data
with Simple Plots
Wayne Nelson
Volume 2: How to Perform
Continuous Sampling
Kenneth S. Stephens
Volume 3: How to Test Normality
and Other Distributional
Assumptions
Samuel S. Shapiro
Volume 4: How to Perform Skip-Lot
and Chain Sampling
Kenneth S. Stephens
Volume 5: How to Run Mixture
Experiments for Product Quality
John A. Cornell
Volume 6: How to Analyze
Reliability Data
Wayne Nelson
Volume 7: How and When to
Perform Bayesian Acceptance
Sampling
Thomas W. Calvin
Volume 8: How to Apply Response
Surface Methodology
John A. Cornell
Volume 9: How to Use Regression
Analysis in Quality Control
Douglas C. Crocker

Volume 10: How to Plan an


Accelerated Life Test-Some Practical
Guidelines
William Q. Meeker and Gerald J.
Hahn
Volume 11: How to Perform
Statistical Tolerance Analysis
Neil D. Cox
Volume 12: How to Choose the
Proper Sample Size
Gary G. Brush
Volume 13: How to Use Sequential
Statistical Methods
Thomas P. McWilliams
Volume 14: How to Construct
Fractional Factorial Experiments
Richard F. Gunst and Robert L.
Mason
Volume 15: How to Determine
Sample Size and Estimate Failure
Rate in Life Testing
Eduardo C. Moura
Volume 16: How to Detect and
Handle Outliers
Boris Iglewicz and David C.
Hoaglin
1Available through ASQC Quality Press
at 800-248-1946

Editors Corner

VISION

Over the next several months, many of you will have the opportunity to
speak to Falguni Sharma. In her new role as the Acquisitions Coordinator for
the Statistics Division, Falgunis main responsibility is to develop a list of
potential topics consistent with the needs of Division customers, and to identify potential authors for Division Publications. Please support her efforts.
Readers should consider submitting either a basic tools or mini paper for
publication. The criteria for the basic tools and mini paper columns appear
on the following page. Papers can be submitted with one hard copy and
one copy on a 3-1/2 diskette. Since I use Microsoft Word for Windows, the
files should be sent in either Microsoft Word, ASCII Text File, or
WordPerfect. Figures should be properly identified and labeled. (Please do
not embed figures in the text.)
You will notice that there is no deadline for the Spring 1996 issue. In the
place of the Spring newsletter, the Statistics Division will present a special
publication on Statistical Thinking. The edition will focus on Statistical
Thinking concepts as presentation by Lynne Hare, Roger Hoerl and Ron Snee
at the 1995 AQC Conference in Cincinnati.
Janice

Inside This Issue


Letters........................................................................p. 4
Call for New Regional Councilors ...........................p. 4
Youden Address .......................................................p. 6
Tactical Planning Meeting......................................p. 14
Annual Quality Congress Activities .......................p. 16
Hunter Award .........................................................p. 19
Job Openings..........................................................p. 20
Deadline for Newsletter Contributions .................p. 24

- Our customers needs will be continuously anticipated and met.


- Our members will be proud to be a
part of the Division.
- Our Divisions operations will be a
model for other organizations.
- We will be a widely influential
authority on scientific approaches to
quality and productivity improvement.

MISSION
- Promote statistical thinking for quality and productivity improvement.
- Serve ASQC, business, industry,
academia and government as a
resource for effective use of statistical methods for quality and productivity improvement.
- Provide a focal point within ASQC
for problem-driven development
and effective use of new statistical
methods.
- Support the growth and development of Division members.

STRATEGY
- Our primary customers are Statistics
Division members. Other key customers are:
- Management,
- Users and potential users of statistical methods for quality and
productivity improvement,
- Educators of the above customers.
- Our orientation to customers is customer focused.
- Our markets, within which we
intend to offer products, are weighted as follows: greatest weight on
intermediate statistical methods,
nearly as much weight on basic
methods, and much less weight on
advanced methods.
- Our primary products are educational services.

PRINCIPLES
- Focus on a few key things.
- Balance short-term and long-term
efforts.
- Recognize that we exist for our customers.
- Value diversity (including geographical and occupational) of our membership.
- Be proactive.
- View statistics from the broad view
of quality management.
- Apply statistical thinking ourselves
(that is, practice what we preach).
- Uphold professional ethics
- Continuously improve

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

CHAIRS MESSAGE
Continued from front

For the last few years, the Division


has been discussing improvements to
the series. We needed to develop
explicit strategic and tactical plans to
guide the future activities of the How
To series. Specific consideration
were the types, levels and formats of
products and its audience and future
topics. This was broken into smaller
tactical plans that could be completed
within a year. The first plan was to
document and improve How To
series infrastructure. At the same time
that Ed Mykytka was putting together
a straw-proposal for this plan, the
Division was looking at revising their
bylaws. The Model Division bylaws
included a publications committee as
a standing committee. While the
Division had several publications, we
did not have a formal infrastructure
linking them together or an avenue for
other publication areas such as briefings. Hence, the birth of the
Publications Committee.
The groundwork for this committee
was laid by the tactical planning team
last Spring. The committee consists of
Publications Committee Chair, How
To Series Editors, Briefings Editor,
Glossary Editor, Acquisitions
Coordinator and in the future, New
Products Coordinator. The Newsletter
Editor is an ad-hoc member of this
committee.
Ed Mykytka has decided to retire
from the role of How To Series
Editor. We thank him for his time and
dedication to the series. Replacing Ed
as How To Series Editors are Walter
Liggett and Bob Brill. Bob and Walter
are already hard at work in their new
roles. They along with the rest of the
Publications Committee met during
the Tactical Planning Meeting at the
Fall Technical Conference (FTC) and
established short-term and long-term
goals. You will find articles discussing
this meeting as well as other items
from the FTC throughout the newsletter.

I would like to discuss in further


detail the role of regional councilors.
The regional councilor serves as a
liaison between the sections and the
Division. If you have looked at a map
of the regions, you will quickly realize
that we are asking one person to be
responsible for a large number of sections. We believe that the role of
regional councilor is important since
working with sections is one way for
the Division to be closer to you, our
members. The General Technical
Council, the overseeing body for all
divisions, realized the need to allow
us to modify our bylaws with regard
to regional councilors. The Statistics
Division Council voted to change the
position of regional councilor from
elected to appointed. This will allow
us to appoint more than one councilor
per region. Bob Mitchell, our
Membership Chair, is leading a tactical
planning team to work on defining the
role of these councilors. If you have
any ideas on how a Statistics Division
councilor might work with sections to
meet your needs, please contact Bob.
Finally, I want to thank Nick
Martino and Jacob Van Bowen as well
as the entire FTC committee for their
efforts in making the 1995 FTC a success. Nick served as Short Course
Chair and organized short courses on
response surface methodology and
decision and risk analysis. Van served
as Program Committee Representative
and was responsible for working with
representatives from the other sponsoring organizations to put together
the program. Van is serving in a twoyear slot and will be chairing the 1996
program committee. There are currently opportunities for individuals to
serve in similar roles for the Division
in upcoming years. If you are interested, please see the article discussing
job openings and complete the member interest form.

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

Criteria for Basic


Tools and Mini-Paper
Columns
Basic Tools
Purpose: To inform/teach the quality
practitioner about useful techniques
that can be easily understood, applied
and explained to others.
Criteria:
1. Application oriented/not theory
2. Non-technical in nature
3. Techniques that can be understood
and applied by non-statisticians.
4. Approximately three to five pages or
less in length (8 1/2 x 11 typewritten, single spaced.)
5. Should be presented in how to use
it fashion.
6. Should include applicable examples.
Possible Topics:
New SPC techniques
Graphical techniques
Statistical thinking principles
Rehash established methods
Mini-Paper
Purpose: To provide insight into application-oriented techniques of significant value to quality professionals.
Criteria:
1. Application oriented.
2. More technical than Basic Tools, but
contains no mathematical derivations.
3. Focus is on insight into why a technique is of value.
4. Approximately six to eight pages or
less in length (8 1/2 x 11 typewritten, single spaced.)
Longer articles may be submitted
and published in two parts.
5. Not overly controversial.
6. Should include applicable examples.
General Information
Authors should have a conceptual
understanding of the topic and should
be willing to answer questions relating
to the article through the newsletter.
Authors do not have to be members of
the Statistics Division.
Submissions may be made at any
time to the Statistics Division
Newsletter Editor. All articles will be
reviewed. The editor reserves discretionary right in determination of which
articles are published.
Acceptance of articles does not
imply any agreement that a given article will be published.

Letters to the Editor


Dear Janice,
Please pass on my appreciation to
Stu (Hunter) for the article on the
robustness of the EWMA. I do have
one comment that may be relevant. I
believe in Table 4 of the article, the
column labeled Yt+1 should really
be labeled Yt, and thats read: Y
forecasted for period t, made at period
t-1. Otherwise the math and rows
dont line up.
The control algorithm on page 10 is
kind of hard to follow.....e.g. the tab

misalignment in the 2nd row, and I


think the 1st 44.5 under New Obs.
should be under the EWMA column.
Trying to follow it is somewhat user
hostile.
Alex T.C. Lau
Imperial Oil
Toronto, Ontario, Canada
Thanks for your input. I contacted Larry and Stu for clarification. Their input will be published
in a later edition.

Related Events
JOINT RESEARCH CONFERENCE ON STATISTICS IN QUALITY,
INDUSTRY AND TECHNOLOGY
The 13th Quality and Productivity
Research Conference and the 3rd
Spring Research Conference on
Statistics in Industry and Technology
will be held jointly at the National
Institute of Standards and Technology
(NIST) in Gaithersburg, MD (a suburb
of Washington, DC) from May 29
through May 31, 1996. The goal of
this conference is to stimulate interdisciplinary research among statisticians,
engineers, and physical scientists in
quality and productivity, industrial
needs, and the physical and engineering sciences. The conference is jointly
sponsored by the ASA Section on
Physical and Engineering Sciences, the
ASA Section on Quality and
Productivity, IMS, E. I. du Pont de
Nemours & Co. and NIST.
The conference will feature presentations by scientists and engineers for
statisticians and presentations by statisticians for scientists and engineers.
Statistical issues and research
approaches drawn from collaborative
research will be highlighted.

Professor Vijay Nair, from the


University of Michigan, and Professor
William Golomski, from the University
of Chicago Business School, will be
plenary speakers. They will discuss
broad issues and opportunities in
interdisciplinary research from different perspectives. There will also be
both invited and contributed paper
sessions.
For information on submitting contributed papers, contact Will Guthrie
(guthrie@cam.nist.gov or (301) 9752854). The deadline for submitting
contributed papers is February 1,
1996.
For further information about the
conference, check the NIST Statistical
Engineering Division Home Page
(http://www.cam.nist.gov/caml/sed/)
or contact one of the conference cochairs: Eric Lagergren
(lager@cam.nist.gov or 301-975-3245)
or Raghu Kacker
(kacker@cam.nist.gov or 301-9752109).

Call for New


Regional Councilors
The Statistics Division has openings
for two Regional Councilors, in region
#5 (Pennsylvania, Delaware and
Maryland) and in region #9 (Indiana,
Kentucky, and northwestern parts of
Ohio). If you would like to become
more involved in the Statistics Division
and enjoy networking with fellow statistical practitioners, we would like to
hear from you.
At the October 18 Statistics Division
Council meeting, a change to the
Division bylaws was approved that
allows for more flexibility in appointing as many councilors as needed to
address the needs of our members
and customers, The appointed
Regional Councilor is a non-voting
member of the Statistics Division
Council. Regional Councilors serve as
a link between the Division and the
ASQC Sections. An excerpt of the
existing job description is summarized
below.
Purpose:
To serve as section liaison for the
Statistics Division.
Responsibilities:
Participate with the Statistics
Division Council to determine how
the local sections can help implement
Division tactical plans, communicate
Division news, and seek active member participation.
Contact Section Chairs to determine
how the Statistics Division can help
provide speakers, communicate
events, schedule short courses, etc.
Exhibit the Statistics Division display booth at local section meetings,
disseminate pamphlets and brochures,
seek new members.
Interested members are asked to fill
out the enclosed Member Interest
Record Form and/or contact: Bob
Mitchell, Membership Committee
Chair, (612) 234-1864

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

ASQC STATISTICS DIVISION


1995 -1996
REGIONAL COUNCILORS
Regional Councilor Coordinator
Bob Mitchell
3M Consumer Products Plant
915 Highway #22 South
Hutchinson, MN 55350
Phone: (612) 234-1864
Region 1:
Bob Gillies
32 Dartmouth Drive
Mystic, CT 06355
Phone: (203) 445-3145

Region 6:
Marilyn Hwan
LSI Logic, MS J-202
3115 Alfred Street
Santa Clara, CA 95050
Phone: (408) 433-6362

Region 11:
George Marrah
Department of Mathematics
& Computer Science
James Madison University
Harrisburg, VA 22807
Phone: (703) 568-6534

Region 2:
Mary Garfield
205 Bryant Street
Rochester, NY 14613
Phone: (716) 722-2392

Region 7:
Tom Vaden
Consultant
5765 Grand Avenue
Riverside, CA 92504
Phone: (714) 382-5525

Region 12:
Bob Dovich
Ingersoll Cutting Tool Company
505 Fulton Avenue
Rockford, IL 61103
Phone: (815) 987-6542

Region 3:
Rich Christy
American Premier, Inc.
P. O. Box 1569
901 East 8th Avenue
King of Prussia, PA 19406
Phone: (215) 337-1100

Region 8:
Bill Bleau
Picker International Inc.
1130 Stonecrest Dr
Tallmadge, OH 44278
Phone: (216) 473-2385

Region 13:
Rick Schleusener
Kodak Colorado Division
Building C-42, 3rd Floor
Windsor, CO 80551-1672
Phone: (303) 686-4530

Region 4:
Michael Cohen
Satisfied Brake Products
650 100th Avenue
Chomedy Laval
Quebec, Canada H7W-3Z6
Phone: (514) 337-3280

Region 9:
Vacant

Region 14:
Oz Godsey
303 Ridgebriar Drive
Richardson, TX 75080
Phone (214) 690-1744

Region 5:
Vacant

Region 10:
Greg Gruska
The Third Generation, Inc.
4439 Rolling Pine Drive
West Bloomfield, MI 48323
Phone: (313) 363-1654

Region 15:
Dan Dankovic
Westinghouse Electric
Corporation
700 Energy Lane
Fort Payne, AL 35967
Phone: (205) 845-9601, Ext. 6033

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

Youden Address
USING ON-LINE PROCESS DATA TO IMPROVE QUALITY
Is There a Role for Statisticians?
Are They Up for the Challenge?
John F. MacGregor
Dept. of Chemical Engineering
McMaster University
Hamilton, ON, Canada L88 4L7
ABSTRACT
In keeping with the conference theme of Tapping Diverse Data Sources to Improve Quality this paper presents a
case for using the very large number of process variables measured routinely by on-line computers. Multivariate statistical methods can be used to project the data down into low dimensional spaces where analysis, monitoring and diagnosis
are easily performed. Strong justifications for taking this approach are presented, and several examples are given.
The statistical process control community has been slow in picking up on the data explosion brought about by the
computer era. It has continued to stick with traditional univariate charts on the quality variables, and ignored this rich
source of additional information on the process. This paper explores some of the reasons for this and argues that the
SPC community must adapt rapidly to this multivariate reality or lose control of their field to the scientists and engineers.
INTRODUCTION
The period of Jack Youdens lifetime (to 1972) represents, in my opinion, the golden years of applied statistics.
Statistical methods and thinking were introduced into nearly all disciplines of science and engineering by people such as
Fisher, Youden, Box and Tukey. These statisticians were primarily scientists themselves - geneticists, chemists, engineers. They understood the way scientists and engineers thought, and they understood their problems. As a result they
developed statistical methods to treat these real world problems, not by starting with the statistical theory and trying to
find a problem it could treat, but by starting from the very real problems and developing methods to treat them.
The last two decades have seen statistics grow as a mathematical discipline. However, this period has seen much less
interesting growth in applied statistics, not because there were no new problems, but because the leadership in the statistical disciplines passed on to a new generation of mathematical statisticians.
My belief is that we are once again seeing a major shift in the leadership and direction of the statistical community.
This new era is being ushered in by the advent of on-line process, laboratory, and management computer systems.
These have totally changed the nature of the data we are seeing. We now live in a highly multivariate, data-rich society,
and are being inundated with data from all directions. The small sample univariate methods of the past are hard pressed
to handle this new situation. This has opened the door to the need for an explosion of new statistical methods to treat
the new problems associated with large volumes of messy data that are being collected routinely every second, minute or
hour. These massive data sets and the problems with interpreting them arise in all areas where computers have penetrated - communications, image and speech analysis, management information systems, chemistry, the process industries,
etc. The leadership in developing statistical methods for these data rich problems has again returned to the owners of
the problems - scientists and engineers - paralleling the earlier era of applied statistics when scientists such as Youden
developed methods to treat the problems in their areas of science.
The theme of this conference - Tapping Diverse Data Sources to Improve Quality - has motivated the main topic of
this address - Using On-Line Process Data to Improve Quality. Under
this topic I will discuss some new multivariate approaches to SPC which
acknowledge that on-line process computers have become commonplace,
and that huge amounts of process data are being collected on a routine
basis. A secondary theme concerning the role that statisticians might play
in these developing areas is then explored.
The SPC Problem
To illustrate the problem of monitoring process performance and product quality, and diagnosing assignable causes for special events, consider
the process situation depicted in Figure 1. Process computers routinely
Figure 1

Continued on page 7

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

YOUDEN ADDRESS
Continued from page 6

collect measurements on a large number of process variables such as flows, temperatures, pressures, etc. Let us denote
these process variables by xj; j = 1, 2, ... ,k, and let the (N x k) matrix X contain observations on them over some time
period (i = 1, 2, ...,N). In addition to these frequently measured process variables we also usually measure several quality and productivity variables ( ; = 1, 2, ...,m). These variables are often measured much less frequently on samples
collected and analyzed off-line in a quality control lab.
Traditional approaches to statistical process/quality control established a set of univariate control charts on each of the
quality variables (y ). Occasionally multivariate extensions of these charts based on Hotellings T2 statistic are
employed on a small number of the most highly correlated ys. The process data (X) are almost never used directly in
monitoring the process, but only in searching for an assignable cause once one of the control charts on the ys has indicated the presence of a special event.
(iv) Missing Data
Most large process data sets contain missing data (sometimes up to 20%). Therefore, if we are to analyze such
process data, and if we are to establish multivariate control charts to monitor the future behavior of the process, then the
Projection to Latent Structures (PLS) is another multivariate projection method which can be applied when one
has two data matrices - a process data matrix (X) and a quality data matrix (Y). Now one is interested in the high variance directions in the process data (X), but more specifically the high variance directions in X that are related to the variation in the quality data Y. In the first dimension PLS accomplishes this by extracting that linear combination of the
process variables t1 = w1T x which maximizes the covariance of X and Y. Subsequent PLS latent variables (t2, t3...) are
again required to be orthogonal to preceding ones. Latent variables (ui = ciT y) are also obtained for the Y space which
are most correlated with their corresponding tis. The analysis, interpretation, and monitoring can again be more easily
performed in the reduced dimension defined by the orthogonal latent variables (ti), but unlike PCA, PLS is focusing attention on that variation in the process variables that is most explanative of the quality data.
Problems with Traditional Approaches
Although such an approach was well suited to information-poor systems of past decades, there are two major flaws in
it when applied to todays information-rich systems.
(i)

The Multivariate Nature of Quality


The first problem is that product quality is a multivariate property and must be treated as such. By this I mean
that a high quality product must simultaneously have the right combination of all the individual ys. Each individual y
has little meaning by itself. Furthermore, with the improvement in measurement and material characterization methods, it
is now common to obtain many more quality measures on a product. These additional quality measures are usually
highly correlated. For example with synthetic fibers up to 10 or more quality variables are typically measured, but a
Principal Component Analysis (PCA) often reveals that they are so highly correlated that the true dimension of the quality
space is more likely about four. In such situations monitoring univariate control data on all the individual ys can be very
misleading. This is illustrated in Figure 2 with two correlated ys (correlation coefficient 0.9). The point indicated by
the x, although well within the control limits on the individual charts, is clearly very unusual when plotted in the multivariate space. It is unlikely that product with this unusual combination of ys would perform in the same manner as the
other product in a customers end use application. Extending the univariate quality control charts to multivariate ones
based on Hotellings T2 would normally handle this problem well as long as the dimension is not too large. Any point x
lying outside the elliptical contour in Fig. 2 would be detected as unusual by multivariate T2 charts.
It is surprising that more multivariate charts are not advocated by applied statistics and SPC groups. This is in sharp
contrast to the current practice in the design of experiments (DOE) where, horrified by the desire of some scientists and
engineers to perform experiments on one variable at a time, statisticians have been quite successful in introducing the
ideas of designed experiments in which many variables are changed simultaneously. And yet many of these same statisticians appear to be content to meet with quality control groups and recommend that factors be monitored one at a time
via univariate charts, thereby leading to the situation depicted in Figure 2. Is not the use of univariate SPC charts in multivariate situations directly analogous to one factor at a time experimentation in DOE? The presence of variable interactions in DOE leads to the same difficulties in interpreting the results of one factor at a time experimentation as does the
presence of correlation among variables in interpreting univariate SPC charts. Perhaps, as applied statisticians, we should
try to be more consistent in our approaches, or at least explain the reasons for our lack of consistency if we are to be
considered credible by those we are advising.
Continued on page 8
ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

YOUDEN ADDRESS
Continued from page 7

(ii)

Ignoring the Process Data


The second, and by far the most important problem with current SPC approaches, is that they ignore most of the
available data - namely the massive amounts of data that are routinely collected on the process variables (X). These
process data (X) must play an extremely important role in SPC for the following reasons.
(a) Many more process variables (xs) are usually measured than quality variables (ys). In chemical processes it is
not uncommon for the ratio of process variables to quality variables to be in the order of a hundred to one.
(b) The process variables are almost invariably measured much more frequently, usually in the order of seconds or
minutes as compared to hours for the quality variables.
(c) The process variables are measured on-line by process computers as opposed to quality variables which are usually measured off-line in a quality control laboratory. As a result the process variables are available instantaneously while the quality variables are often available only after a considerable delay (often several hours).
(d) The process variables such as temperatures, pressures, flows, etc. are usually measured more precisely than the
quality variables since the latter often involve complex off-line chemical preparation and analysis.
(e) Any faults or special events that occur in the process will leave their fingerprints in the process data as well as
the quality data. It is rarely the case that events affecting product quality will not show up in any of the measured process variables. However, in my experience the converse is sometimes true - that events affecting product end use quality usually show up in the process variable measurements, but perhaps not in the limited set of
quality measurements (y) made by the manufacturer. This of course implies that the quality data (y) being measured by the manufacturer are an insufficient characterization of end use performance.
(f) Eventually, the process data will be needed in any case. If a fault or special event is detected by a control chart,
then one must usually go back to the process data in order to look for an assignable cause. As shown later this
task will be much easier if the X data are used in the control charts in the first phase.
From the above points it should be apparent that both the process data (X) and the product quality data (Y) should be
used in any Statistical Process Control scheme. Of the two sets of data (X, Y), in my opinion the process data (X) is usually of much greater value to SPC than the quality data (Y).
Difficulties with Multivariate Process Data
Having made the claim that we should be using all of our process data in SPC schemes, it is also important to
point out that there are many difficulties with these data.
(i)

Dimensionality
The first problem is that the dimension of the problem is usually very large. It is not uncommon in the chemical
industry for on-line process computers to measure hundreds or even thousands of process variables every few seconds,
and for ten or more quality variables to be measured off-line every few hours. Recent trends in process instrumentation
are also allowing for on-line measurement of some quality variables every few minutes. The shear magnitude of the
problem leads to a DATA OVERLOAD situation in which, overwhelmed by the massive amount of data, operators and
engineers have resorted to following only a few key process variables on their computer screens.
(ii)

Colinearity
Even though hundreds of process variables are measured there are not hundreds of independent events happening in the process. Only a few underlying common cause events (eg. normal raw material variations, etc.) are usually
operating under in-control situations and most faults or special events are the result of a single cause or only a few
simultaneous causes. The hundreds of measured process variables are all related to these small number of underlying
events, and are therefore highly correlated with one another. Hence, process data is very difficult to use because its true
rank is very much less than the number of variables measured. However, each measured variable is valuable because it
contains a little different information on each of underlying events.
(iii) Noise
All of the process variables (as well as the quality variables) are measured with errors (measurement error, sampling error etc.) and the signal-to-noise ratio in any one measured variable is very small during in-control operation.
This is understandable when one realizes that the objective of operators and process engineers is to make the signal-tonoise ratio as small as possible during normal operating conditions. As a result plotting univariate control charts on many
process variables is of little value. However, if each of these process measurements contain a small amount of signal
and each of them contributes some new information on the underlying events, then by using suitable multivariate statistiContinued on page 9

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

YOUDEN ADDRESS
Continued from page 8

cal methods, we can obtain SPC charts having an extremely high information content or signal on the underlying events
occurring in the process.
(iv) Missing Data
Most large process data sets contain missing data (sometimes up to 20%). Therefore, if we are to analyse such
process data, and if we are to establish multivariate control charts to monitor the future behavior of the process, then the
multivariate methods used must not require that the data on all variables be present at all times. The procedures must be
able to handle missing data in a very transparent manner and still extract as efficiently as possible the information from
the remaining measurements.
In the next sections I describe some multivariate methods capable of addressing the above difficulties, use them
to develop multivariate SPC charts, and illustrate their application to process data.

observations

Multivariate Statistical Projection Methods


For treating large, ill-conditioned data sets that are less than full statistical rank there exists a number of very useful multivariate projection methods. Among them are well known multivariate statistical methods such as Principal
Component Analysis (PCA)(Jackson,1991), and Canonical Correlation Analysis (Johnson and Wichern, 1988) as well as
some perhaps less well known methods such as Reduced Rank Regression (or Redundancy Analysis) and PLS (Partial
Least Squares or Projection to Latent Structures - Martens and Naes, 1989). All of these methods simplify the data analysis
and the subsequent process monitoring problems by projecting the data into low dimensional latent variable spaces.
Burnham et al. (1996) present an objective function framework that shows how all of
variables
these methods are related. In this paper I will only be considering PCA and PLS.
PCA is a method extracting the major variance components from a single matrix
of data (X). The projection aspects of PCA are well treated by Wold et al. (1987), and
are illustrated in Figure 3. After mean centering and scaling X, the first principal component (t1) is that variable defined as a linear combination of the x variables (t1 = p1T x)
which has greatest variance. The second principal component is that linear combination
t2 = p2T x with next greatest variance subject to the condition that it is orthogonal to the
first component. The vectors of observed values of t1 and t2 are referred to as the
scores and the coefficient vectors p1 and p2 as the loadings. The scores show the size
of the variation in any particular multivariate observation, while the elements of loading
vectors show the relative importance of each variable to the corresponding principal
component. One can extract as many principal components as one has either variables
or observations. However, in SPC situations the number of statistically significant comFigure 3
ponents is usually very small (two to five) reflecting the fact that under common cause
variation only a small number of events are affecting the process. Cross-validation is often used to determine the number
of dimensions necessary. In Figure 3 the projection aspect of PCA is illustrated assuming that only two components are
necessary. The variation in X explained by the first two principal components is represented by the projection of the
data onto the two dimensional plane defined by the loading vectors p1 and p2. The scores (t1 and t2) define the position
of the projected data points on this plane. If most of the important variation in the data lies in the plane, then one can
display this plane on the computer screen, and examine the movement of the process in the space of the latent variables
t1 and t2. This provides us with a two-dimensional window onto our hundred dimensional process, analogous to our
viewing the projection of a three-dimensional world on a two-dimensional television screen. The analysis, interpretation
and monitoring of the process data is thus greatly simplified by working in a much lower dimensional space which contains most of the important process variations.
Projection to Latent Structures (PLS) is another multivariate projection method which can be applied when one
has two data matrices - a process data matrix (X) and a quality data matrix (Y). Now one is interested in the high variance directions in the process data (X), but more specifically the high variance directions in X that are related to the variation in the quality data Y. In the first dimension PLS accomplishes this by extracting that linear combination of the
process variables t1 = w1T x which maximizes the covariance of X and Y. Subsequent PLS latent variables (t2, t3...) are
again required to be orthogonal to preceding ones. Latent variables (ui = ciT y) are also obtained for the Y space which
are most correlated with their corresponding tis. The analysis, interpretation, and monitoring can again be more easily
performed in the reduced dimension defined by the orthogonal latent variables (ti), but unlike PCA, PLS is focusing attention on that variation in the process variables that is most explanitive of the quality data.
Continued on page 10
ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

YOUDEN ADDRESS
Continued from page 9

Multivariate SPC
If ones objective is to explore a process database or monitor the behavior of a process by using all the available
process data (X) and quality data (Y), then using individual plots and univariate control charts is not a feasible approach.
Based on the discussions in the previous sections, an obvious approach is to project the data into the low dimensional
spaces defined by the latent variables (tis for the X space) and (uis for the Y space). Since the process data is available
much more frequently and has many other advantages as discussed earlier, score plots of t1 vs t2, etc. provide the most
Condenser
useful windows into the behavior of the process.
Exploring Process Databases
To illustrate the utility and simplification of such projection plots consider data from an industrial batch polymerization process such as the one illustrated in Figure 4. The analysis of
batch processes via multivariate proMonomers
Catalyst Emulsifier
jection methods is discussed in
Nomikos and MacGregor (1994, 1995)
Steam
and Kourti et al. (1995). The data
base from such discontinuous batch
Jacket effluent
processes is of the structure shown in
Water
Figure 5. For each of the batches (55
Figure 4
in this example) data is available on
the raw material data and initial operating conditions (Z), and on the final
Figure 5
product quality (Y). Also available are on-line measurements of process variables, X (temperatures, pressures, agitator power, etc.) over the entire history of the batch.
Fault I
These process variables attempt to track preprogrammed setpoint trajectories. In this
Normal batches
example the trajectory measurements on 10 variables at 200 time periods was used. A
score plot for an MPCA model in the space of the second and third latent variables (t2, t3)
is shown for the 55 batches in Figure 6. It is clear from this plot that there are several disFault II
tinct clusters of batches that differ from the main cluster of batches. The main cluster contains all those batches that yielded good quality product while the other three clusters show
Fault III
batches from which poor quality material was obtained, each cluster representing a different type of fault in the process operation. (Note that the clear distinction between the bottom two clusters is apparent in the other latent variable dimension - ie. one of these clusFigure 6
ters falls well behind the other in that dimension). By examining the loading vectors (wi)
of the model the process variables contributing to the shift in these clusters away from the main in-control group can
be determined, thereby suggesting assignable causes for these poor quality batches. In this particular industrial example
it led the company to make important process modifications to all their batch plants to eliminate the identified source of
the major
problem.
Multivariate SPC Charts
Once this phase of the analysis and exploration of past historical data has been completed, and any assignable
causes for problems that have been observed in the historical data have been corrected, it is desirable to set up SPC
charts to monitor the future behavior of the process. To accomplish this one must built a PCA or PLS model for the
process when it is operating well and producing only good quality product, and then reference future behavior against
this model. This is exactly analogous to Shewharts SPC philosophy where univariate charts are devised based on a
model for the in-control mean and variance of that process variable and future behavior referenced against the model.
The only difference in the multivariate case is that we are using all variables (X and Y) simultaneously to develop the
model, and will reference future multivariate data against it. To accomplish this we need to collect a reference data set
on the process which we believe represents the various modes of process operation which lead to good product quality.
Such a data set might be the central cluster of batches in Fig. 6 of the last example.
On-line multivariate SPC charts which use multivariate projections of the process data (X) and quality data (Y)
have been developed for both continuous processes (see for example Kresta et al., 1991; MacGregor et al., 1994;
MacGregor and Kourti, 1995), and for batch processes (eg. Nornikos and MacGregor, 1994, 1995; Kourti et al., 1995).
Using the good reference data a PCA or PLS model is built and the variation in the scores and residuals calculated from
Continued on page 11

10

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

YOUDEN ADDRESS
Continued from page 10

this reference data are used to define control limits in the score space, and in the residual space. The future behavior of
the process is then monitored by plotting the projection of the data in the score space (either as joint t1 - t2 plots, or as
individual Shewhart
plots on the scores since the score vectors are orthogonal), and a control chart on the Squared
m
Prediction Error SPE = (xi - x i)2. Any value of the SPE above its control limit suggests that one has moved off the plane
i=1
defining good process behavior, and hence that a new event not in the reference data base has occurred. Any values
of the score (t1, t2, ..) outside their control limits also signal an event. If an accompanying increase in SPE has not
occurred then large scores simply imply that the process is operating according to its normal model (ie. on the plane),
but that unusually large variations have occurred.

Figure 7

To illustrate these charts I use an example from Nomikos and MacGregor (1994) on a batch process for the production of styrene-butadiene rubber (SBR) by emulsion polymerization. A reference data base of 50 batches, 9 variables
and 200 time intervals/batch was used to build an MPCA model having three latent variables. Control charts (t1 vs. t2; t1
vs time; SPE vs. time) for a new batch producing good quality product are shown in Figure 7. The control limits represent those for approximate 5% and 1% type I errors. These control charts reveal no unusual events over the history of
this batch. With a result such as this it might be reasonable to allocate or ship the product from this batch as in-control
material based on this process variable control charts, rather than waiting many hours for the laboratory results on Y to
come back. In Figure 8 control charts are shown for a new batch which resulted in poor quality product. It is obvious
from the SPE chart and the score plots that a problem developed half way through the batch. To help diagnose an

Contribution %

Contribution Plot

Variables

Figure 8

assignable cause for the event, one can interrogate the underlying PCA model at point 106 where the SPE chart detects
the event, and plot the contribution of the process variables to the SPE at that point in time. This plot (in Fig. 8) shows
that variables 4, 5 and 6 are the main contributors to the large SPE. Variable contribution plots for the t scores and
Hotellings T2 can similarly be calculated (see Miller et al., 1993; MacGregor et al., 1994; Kourti and MacGregor, 1996).
Some Statistical Issues
When dealing with large multivariate data sets and multivariate control charts in the projection space, a number
of statistical issues arise, many of which are not yet resolved. A few of them are discussed below.
(i)

Control Limits on the Charts


To determine control limits for the SPE and t-score charts we do not have to make the usual assumptions of
Normality and independence. With large data sets available from the historical databases of on-line computers, the
model can be built and the control limits can be established directly from the reference distribution of scores and SPE
provided by the database. The philosophy behind this approach using reference distributions is well described in chapters 2 and 3 of Box, Hunter and Hunter (1978), and was employed to obtain multivariate control limits in Nomikos and
MacGregor (1995). The only important assumption needed is that the historical data set provides a representative sample
of common cause variation.
Continued on page 12
ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

11

YOUDEN ADDRESS
Continued from page 11

(ii)

Hypothesis Testing and Control Charts


The basis of most univariate control charts lies in hypothesis testing. We usually assume a distributional model
for the process and established a null or in-control hypothesis that specifies parameters such as the mean and variance
of the distribution. The alternative hypothesis is usually that either the mean or the variance has shifted.
In the multivariate industrial processes the null or in-control hypothesis is that new data are generated from a
process with the same PCA or PLS model and that the distributions of the scores and SPE from this model are the same as
their reference distributions obtained from the historical data. The alternative hypothesis is simply the process no longer
follows this model and its reference distribution. In general it is not possible to be more specific and hypothesize that
just the process mean has shifted, or the variance-covariance matrix has changed. The exact nature of the effect of a special event on the process is not easily specified. Consider, for example, the special event occurring in the polymerization
process in Figure 6 where an increase in the level of impurities in one of the feedstreams occurred. Such an event
affects not only the means of many of the process variables, but also their variances and covariances in an unknown
way.
The complexity of the multivariate processes also makes it difficult to find assignable causes for events once they
have been detected. For this reason they need to be supplemented with diagnostic methods such as contribution plots
which greatly simplify the subsequent search for assignable causes.
(iii) Degrees of Freedom
In evaluating a multivariate hypothesis test or determining control limits based on distributional theory the
degrees of freedom must be determined. In multivariate analysis we usually assume that a matrix of data X with k variables and n observations has nk degrees of freedom, and if we estimate p parameters from the data we are left with (nkp) degrees of freedom. Most of our concern is usually with determining the number of degrees of freedom lost through
estimating certain terms such as the scores and loadings in PCA. However, the real difficulty lies in estimating the number of degrees of freedom we have in the original X matrix. Consider the batch reactor examples discussed earlier. Both
had in the order of 100,000 observations in the reference data set. Is it reasonable to assume that we are starting with
100,000 df.? If so, then a Likelihood ratio test to compare two PCA or PLS models developed from different data sets
would imply the use of an F(100,000; 100,000) distribution and the null hypothesis would be rejected in essentially all
cases.
In reality we never have (nk) degrees of freedom in such process data. The batch reactor data sets probably
have at most a few hundred degrees of freedom. The difference between the effective statistical rank of a matrix and the
numerical rank is an important issue yet to be addressed adequately by statisticians, as is the development of statistical
methods to treat these rank deficient data. Even the notion of an effective number of degrees of freedom has not been
adequately defined. Methods for treating these real world reduced rank problems are not generally available in the literature or in multivariate statistical texts.
Will Statisticians Play a Useful Role?
I return now to the secondary theme of this address - Is there a role for statisticians to play? and Are they up to
the challenge? The answer to the first question is clearly yes. We have entered a new era that has been ushered in by
the use of on-line computers to collect data on an ever increasing number of variables. Many new approaches are needed to efficiently analyse these data. Whether the statistical community is up to this challenge is still an open question.
As an example consider the recent history of statistics in chemistry. Ten to twenty years ago analytical chemistry was still
based on off-line chemical analysis of samples. Very few measurements were made on each sample and univariate statistical methods were quite adequate. The switch over to automated instrumental analysis such as spectrophotometric and
chromatographic methods provided hundreds to thousands measurements on each sample. Statisticians offered little help
and showed little interest in these problems.
As a result the field of chemometrics sprang up in which chemists developed their own set of statistical tools and
procedures. This resulted in the creation of two new journals - the J. of Chemometrics and the J. of Chemometrics and
Intelligent Laboratory Systems - which now attract most of the applied statistics articles in chemistry that once might have
gone to Technometrics. It is interesting to note that the ASA recently rejected a move to create a chemometrics division,
and that this rejection led directly to the creation of a North American Chemometrics Society with no connection to the
ASA.
The area of Statistical Process Control appears to be headed in a similar direction. Although Shewhart presented
a philosophy for SPC which is as valid today as it was in his time, we seem to be constrained by the type of traditional
Continued on page 13

12

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

YOUDEN ADDRESS
Continued from page 12

univariate statistical approaches that worked well in the past. Faced with the large multivariate problems of today
applied statisticians appear immobilized by the immensity of the problems, and by peer pressures under which they are
more worried about the detailed assumptions and correctness of the methods they use than providing even an approximate solution to the problem. On the other hand, they havent fully recognized the tremendous advantages that arise
from having such large data sets. A further inhibition stems from a lack of general knowledge of processes. As one
starts to use data on process variables, a greater understanding of the process itself, and the nature of process data, is
required in order to use and interpret the data in an efficient manner. Another very real barrier to the statisticians
involvement is that the engineers have ownership of the computers and hence of the data they generate. They are often
very reluctant to part with these data. In the absence of involvement by statisticians, they have also developed their own
approaches. An extensive literature exists in fault detection and identification using both theoretical and empirical models, in artificial intelligence approaches, and in some of the multivariate projection (PCA/PLS) approaches highlighted earlier in this presentation.
I have also observed a tremendous resistance by statisticians to using more complex and alternative approaches
to SPC. It is almost as though a certain set of classical univariate methods have been officially sanctioned and no alternatives will be considered until they are also sanctioned by some higher authority. Furthermore, I cannot remember how
many times I have heard the statement: We have difficulty getting people to use even simple methods. How can you
expect them to use multivariate PCA/PLS methods? My standard answer to this argument goes as follows. If simple univariate charts, etc. are not being used, it is probably because they often reveal little useful information, and are much too
difficult to interpret. Personally, I have a lot of difficulty trying to interpret a set of univariate charts on more than four or
five correlated variables. Multivariate projection methods are much simpler to use. All the data are projected down into
low dimensional orthogonal latent variable spaces that contain most of the information. The information is easily presented and easily interpreted in these spaces. Operators dont need to know PCA/PLS theory in order to use these charts
any more than they need to know Normal distribution theory in order to use univariate charts. The only important characteristic of SPC charts are their simplicity of presentation and simplicity of interpretation. The complexity or simplicity
of the underlying statistical methods used to develop the charts is not an issue for the operator or engineer using the
charts.
Summary
I hope that I have been able to make a case in this paper, that a new and exciting era for applied statistics has
been opened up by the advent of on-line computers. The massive data sets we now collect contain a wealth of information that can be extracted by multivariate statistical methods. This is particularly true in the area of statistical process control where process computers and information management systems have improved sufficiently that we can obtain data
on hundreds of process variables on a frequent and regular basis. By using these data in effective multivariate SPC
schemes we have the potential to achieve major improvements in the understanding and control of our processes. I
hope that the statistical community will accept the challenge that this new computer era presents, and work with the
engineers and scientists in realizing this potential.

References
G.E.P. Box, W.G. Hunter and J.S. Hunter, 1978, Statistics for Experiments, John Wiley & Sons, N.Y.
A.J. Burnham, R. Viveros and J.F. MacGregor, 1996, Frameworks for Latent Variable Regression, J. of Chemometrics, In Press.
J.E. Jackson, 1991, A Users Guide to Principal Components, Wiley-Interscience, John Wiley & Sons, Inc., New York.
R.A. Johnson and D.W. Wichern, 1988, Applied Multivariate Statistical Analysis, Prentice Hall, N.J.
T. Kourti, P. Nomikos and J.F. MacGregor, 1995. Analysis, Monitoring and Fault Diagnosis of Batch Processes using Multi-Block, Multi-Way PLS, J.
Proc. Control, 5 No. 4, 277-284.
T. Kourti and J.F. MacGregor, 1995. Process Analysis, Monitoring and Diagnosis using Multivariate Projection Methods, J. Chemometrics and Intell.
Lab. Systems, 28, 3-21.
T. Kourti and J.F. MacGregor, 1996, Multivariate SPC Methods for Monitoring Process and Product Performance, J. Quality Tech., In Press.
J.F. MacGregor, J. Jaeckle, C. Kiparissides and M. Koutoudi, 1994, Monitoring and Diagnosis of Process Operating Performance by Multi-Block PLS
Methods with an Application to Low Density Polyethylene Production, Amer. Inst. Chem. Eng. J., 40, 826-838.
J.F. MacGregor and T. Kourti, 1995, Statistical Process Control of Multivariate Processes, Control Eng. Practice, 3, 403-414.
H. Martens and T. Naes, 1989, Multivariate Calibration, John Wiley & Sons, N.Y.
P. Nomikos and J.F. MacGregor, 1994, Monitoring of Batch Processes using Multi-Way Principal Components Analysis, Amer. Inst. Chem. Eng. J.,
40, 1361-1375.
P. Nomikos and J.F. MacGregor, 1994, Multivariate SPC Charts for Batch Processes, Technometrics, 37, No. 37, 1995.
S. Wold, K. Esbensen, P. Geladi, 1987, Principal Component Analysis, Chemometrics and Intelligent Laboratory Systems, 2(1987), p. 37-52.
ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

13

Division Council Meeting


The Statistics Division Council
meets formally twice per year - in the
Spring at the Annual Quality Congress,
and the Fall at the Fall Technical
Conference.
A council meeting was held in conjunction with the Fall Technical Conference in St. Louis on October 18,
1995. This article briefly summarizes
the meeting. Anyone interested the
full minutes of the meeting may obtain a copy by contacting Don Emerling, Secretary. The meeting was well
attended by Council members and
other interested Division members.
Bob Mitchell reported that membership was 12,410. A full membership
report appears elsewhere in the
Newsletter.
The Council voted on a bylaws
revision. This revision would change
Regional Councilors from elected to
appointed positions. The revision was

approved and will take effect on the


1996 ASQC ballot.
Nancy Belunis announced that the
Division will hold several meetings in
conjunction with the Annual Quality
Congress. A Tactical Planning
Meeting will start at 12:00 p.m. on
Saturday, May 11 and continue on
Sunday, May 12th until 12:00 p.m.
The council meeting will be held
Sunday, May 12 from 8:00 - 10:00 p.m.
The Annual Business Meeting will be
Monday, May 13 from 5:30 - 7:30 p.m.
All interested members are invited to
attend any of these meetings.
Ray Waller, incoming Executive
Director of American Statistical
Association, discussed possible interaction between ASA and Statistics
Division. Possible areas of synergy
are continuing education and computer networks.
Ed Mykytka announced that the

new publications committee would


hold its first meeting during the
Tactical Planning Meeting. A new edition of the glossary will be available
through Quality Press this Spring.
Jacob Van Bowen, Program
Representative for the 1995 and 1996
Fall Technical Conference, announced
that the theme for next years FTC is
Leveraging Data for the Quality
Transformation. The site for the 1996
FTC is Scotsdale, Arizona.
The Standards Committee held a
meeting at the FTC. See report elsewhere in the Newsletter.
Mark Kiel, Bulletin Board
Administration, reported on ASQC Net
and encouraged all members to participate. Mark has set up two libraries
for Division members only. Other
forums like the Statistics Corner are
open to all ASCQ members.

Division Tactical Planning Meeting


A one day tactical planning meeting
was held following the Fall Technical
Conference in St. Louis. This meeting
was held to work on existing tactical
plans. These plans are developed to
implement the Divisions strategy.
Assessing Members Needs
(Leader: Bob Mitchell)
The Assessing Members Needs
Tactical Plan is complete . A survey
has been developed to measure member satisfaction. The survey information will be used to identify areas of
need and opportunities for continuous
improvement based on member input.
Initially, surveys were conducted via
telephone but future surveys will be
conducted via mail on a quarterly
basis.
Enable Broad Application of
Statistical Thinking
(Leader: Roger Hoerl)
This Tactical Planning Committee is
working on a special edition of the
Newsletter on Statistical Thinking.
14

The special edition should be published Spring 1996.

The group will select one or two


sites as pilots.

The group is also working on developing a presentation on Statistical


Thinking that would be made available to interested individuals to present.

Two other areas which were identified


at the AQC Tactical Planning Meeting
were worked on in small groups.

Publications Committee
(Leader: Ed Mykytka)
The new Publications Committee met
for the first time and reviewed the
background, purpose and objectives
of the committee. The group identified and prioritized important policy
issues and developed short-term
objectives and long-term goals.
Integrating Statistical Thinking
into Education
(Leader: Chris Ayers)
The team developed a vision of what
a educationally integrated curriculum
would look like. A process was flowcharted for developing a pilot area.

Division Dashboard
(Leaders: Division Officers)
The team reviewed the concept of a
balanced scorecard for the Division.
A straw proposal was developed and
includes goals and measures in four
perspectives: financial, customer,
internal business and innovation and
learning.
Define role of Regional Councilor
(Leader: Bob Mitchell)
A straw proposal for this tactical plan
will be developed by January 1. The
purpose is to define responsibilities of
Regional Councilors or Section
Liaisons, identify support mechanisms and measurements of performance.

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

1995 FTC Short


Course Summary
The two pre-conference short
courses presented on October 18th
were attended by 80 people. Short
course participants represented more
than 25% of the total number registered for the conference.
Dr. Douglas C. Montgomery presented the short course Response
Surface Methods and Designs based
on his new book, Response Surface
Methodology (co-authored with
Raymond H. Myers and available from
John Wiley.& Sons, NY). A short
course entitled The Fundamentals of
Decision and Risk Analysis was presented by Dr. Gerald A. Bush of
Decision Strategies. Both courses
were well received based on participant feedback.
These Division sponsored courses
cost only $100 to attend, yet are
taught by leading professionals.
Thats why they are consistently complimented as the best value in a short
course." Pre-conference courses will
also be offered at the 1996 FTC in
Scottsdale, AZ.

Joint Research Conference on Statistics


The 13th Quality and Productivity
Research Conference and the 3rd
Spring Research Conference on
Statistics in Industry and Technology
will be held jointly at the National
Institute of Standards and Technology
(NIST) in Gaithersburg, MD (a suburb
of Washington, DC) from May 29
through May 31, 1996. The goal of this
conference is to stimulate interdisciplinary research among statisticians,
engineers, and physical scientists in
quality and productivity, industrial
needs, and the physical and engineering sciences. The conference is jointly
sponsored by the ASA Section on
Physical and Engineering Sciences, the
ASA Section on Quality and
Productivity, IMS, E. I. du Pont de
Nemours & Co. and NIST.
The conference will feature presentations by scientists and engineers for
statisticians and presentations by statisticians for scientists and engineers.
Statistical issues and research
approaches drawn from collaborative
research will be highlighted.

Professor Vijay Nair, from the


University of Michigan, and Professor
William Golomski, from the University
of Chicago Business School, will be
plenary speakers. They will discuss
broad issues and opportunities in
interdisciplinary research from different perspectives. There will also be
both invited and contributed paper
sessions.
For information on submitting contributed papers, contact Will Guthrie
(guthrie@cam.nist.gov or (301) 9752854). The deadline for submitting
contributed papers is February 1,
1996.
For further information about the
conference, check the NIST Statistical
Engineering Division Home Page
(http://www.cam.nist.gov/caml/sed/)
or contact one of the conference
co-chairs: Eric Lagergren
(lager@cam.nist.gov or 301-975-3245)
or Raghu Kacker
(kacker@cam.nist.gov or 301-9752109).

Scholarship Formation
Division representatives are
working to establish procedures for awarding scholarships to deserving, recent college graduates and graduate
students who are pursuing
careers in applied statistics
and quality management.
Original funding for these
scholarships was provided by
a grant from the Ellis R. Ott
Foundation with the understanding that scholarships
would be given to promote
studies consistent with the
philosophies of Professor Ott.
These include applied statistics, quality management,
quality engineering, quality
control and quality assurance,

with the emphasis on applied


research and direct applications as opposed to the development of theory. A broad
scholarship program description has been written, but
details must be fleshed out.
Those interested in assisting
in this formative effort are
encouraged to contact Lynne
Hare. His address is:
National Institute of Standards
& Technology
Statistical Engineering
Division
Bldg. 101/Room A337
Gaithersburg, MD 20988-0001
email: lynne.hare@nist.gov

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

15

CQEs Needed
Statistics Division members who are
also Certified Quality Engineers
(CQEs) are needed to assist in the
development of CQE exams. A key
element in ensuring the quality of this
examination is having the participation of top quality professionals who
are also CQEs. Since statistical methods constitute such a large part of the
exam, the participation from the Statistics Division is particularly critical.
As part of the process of developing new exams each year, volunteers
are needed to participate in any one
of three workshops held at ASQC
headquarters. The three workshops
are:
1. Item Writing - developing
exam questions and answers;
each workshop consists of 30
volunteers, at least 7 are needed
from the Statistics Division,
2. Item Review - critiquing items
produced during the Item
Review Workshop; each workshop consists of 12 volunteers, at
least 3 are needed from the
Statistics Division, and
3. Exam Review - reviewing the
exam prior to its administration;
each workshop consists of 12
volunteers, at least 3 are needed
from the Statistics Division.
The Exam Review Workshop is
held twice each year, the others are
held annually.
The next opportunity is an Exam
Review Workshop scheduled for
March 22-23, 1996. An Item Writing
Workshop is tentatively scheduled for
September 13-15, 1996, followed by
an Item Review Workshop December
6-7, 1996. Reasonable expenses for
travel to Milwaukee, meals and lodging are covered by ASQC. The cycle
of workshops is repeated annually
and you may participate once, or as
often as you like.
You must be a CQE to participate
and be willing to sign a non-disclosure contract. For additional information and details on how to volunteer,
please contact Statistics Division
Certification Chair Nick Martino by
phone at (508) 534-2556 between 7
AM and 4 PM EST; or by mail, Nick
Martino, Novacor Chemicals Inc., 31
Fuller St., Leominster, MA 01453; or by
E-mail, martinnv@mail.novacor.com.

16

1996 AQC Technical Session


Sponsored by Statistics Division
The Statistics Division will sponsor
one technical session at the 1996
Annual Quality Congress in Chicago.
The session will take place on
Tuesday, May 14 from 8:00 - 10:00 am.
The session is entitled In Search of the
Future: Models and Methods for
Whole Systems Change.
In a world of increasing complexity
and change, new models and methods
are needed to rapidly adapt. This session will present a change model,

using the seven new management and


planning tools, which can be used in
a wide variety of applications.
The session will be presented by
Tom Swails. Tom is the quality manager for the Taping Systems Business
Unit of 3M Company. He has facilitated numerous planning session within
3M, non-profit organizations and
industry. Tom led the development of
the most recent Statistics Division
long-range plan.

Annual Quality
Congress Activities
Saturday, May 11
Tactical Planning Meeting

Sunday, May 12
Tactical Planning Meeting
Preconference Tutorial:

12:00 pm - 5:00 pm

8:00 am - 12:00 pm

Committee Meetings
Council Meeting
STAT Division Hospitality Suite

1:00 pm - 5:00 pm
8:00 pm - 10:00 pm
10:30 pm - 11:30 pm

Monday, May 13
Annual Meeting
STAT Division Hospitality Suite

5:30 pm - 7:30 pm
9:00 pm - 11:30 pm

Tuesday, May 14
Technical Session: In Search of the
Future: Models and Methods for
Whole Change Systems
STAT Division Hospitality Suite

8:00 am - 10:00 am

9:00 pm - 11:30 pm

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

AQC Preconference Tutorial


Effective Application of Statistical Thinking
to Business Processes and Systems
The workshop is directed towards
individuals who want to encourage
and support managers in the use of
control charts in non-manufacturing
applications. This workshop is directed primarily towards the statistical
practitioner who will guide managers
in the proper setup of the charts.
Non-manufacturing applications
of control charts refers to the activity
of manufacturing a product rather
than the manufacturing function. The
traditional applications of control
charts have been in the manufacture
of products (e.g. control of peel
strengths, coating weights or percent
additives). By contrast, non-manufacturing applications may include management of labor hours, safety, sales,
computer uptime, logistics cycle time.
manufacturing costs, etc. To avoid
confusion, we have adopted the terminology of business process charts.
Applications of business process
charting are motivated by the perspective known as Statistical Thinking,
which states that:
All work is done by means of a
process.
All processes exhibit variation in
performance.
Variation is of two types: systemic from common causes and
sporadic from special causes.
Improvement means the reduction of variation from the
desired performance, and must
by approached differently
depending on the type of variation present.
Control charts are an effective
way to distinguish the two types
of causes.
While SPC methods and concepts
are clearly applicable in any process
application, there are some important
differences for business processes.
Most business processes are less tangible than manufacturing processes, are

less easily and less frequently measured than manufacturing processes,


and have greater variation (versus
requirements) than the manufacturing
processes. In addition, for some business processes, some patterns in the
data are expected and/or desirable,
such as growth in sales or cycles in
energy use. For these, the straightforward application of the Shewhart control chart may prove unsatisfactory. In
these cases, certain modifications will
be needed as discussed in the workshop.
Examples will be used to explore
data issues specific to business
process charts. Topics will include:
Initial assessment of data
Dealing with limited data
Trends in the data
Autocorrelation
Normalization
Seasonality

Standards
Committee Report
The Statistics Division is responsible for four ASQC Standards: B1, B2,
B3 and S1. These standards are in
need of review for eventual reaffirmation of a five year cycle. The standards cover the following areas:
ASQC Standard B1 (ANSI Z1.1) Guide for Quality Control Charts
ASQC Standard B2 (ANSI Z1.2) Control Chart Method for Analyzing
Data
ASQC Standard B3 (ANSI Z1.3) Control Chart Method for Controlling
Quality during Production
ANSI/ASQC S1 - An Attribute SkipLot Sampling Program
A Standards Committee meeting
was held at the Fall Technical
Conference. A task-force was established to help determine the initial
approach to B1, B2 and B3 before formally setting up a writing group.
Individuals who might be interested in
writing or reviewing standards should
complete the member interest form.

Tuning sensitivity of a measure


Aggregate numbers
Low frequency events
Related measures
Resetting control limits

About the speaker:


Andrew Kirch is an Advances
Statistical Specialist for the 3M
Company. He has an M.S. degree in
statistics from the University of
Wisconsin and an M.A. in applied
mathematics from the University of
Massachusetts.
Andy has worked with 3M plants
and laboratories worldwide for the
past twelve years on a wide variety of
products and processes, from electronics to abrasives. He also has
responsibilities in the areas of statistical software, education and methods
research.

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

17

Membership Report
As of September 30, 1995, the Statistics Division total membership stands at 12,410 (10,469 Regular, 683 Senior, 126
Fellow, 1 Honorary, 283 Student, 688 Sustaining, and 160 New). This current total is down considerably from our membership peak of 15,985 in September 1992.
The Statistics Division membership is shrinking by approximately 100 members per month.

Being a Not-for-Profit organization it is not necessarily the Divisions goal to grow at some given rate; rather, we seek
to continuously anticipate and meet our customers needs. Our goal is to serve our customers well and to provide value.
It is somewhat disconcerting that we continue to lose more members than we gain despite our customer-focus orientation. Recent Long Range Planning sessions have focused on developing the Divisions Mission, Vision, Values, Strategy,
and Principles. The Special Edition Newsletter (Vol. 13, No. 2) published in Winter 94 offers a summary of the
Divisions Long Range Planning efforts. Several Tactical Plans have emerged as the result of our strategic planning:
Enabling the broad application of Statistical Thinking, Integrating Statistical Thinking into Education, Establishing a
Publications Committee, Assessing Members Needs, and several tactical plans involving improvements to the How
To... Series. In addition, the Division has recently named Mark Kiel as BBS Administrator to manage Statistics Division
forums on the ASQCNet electronic bulletin board. Two newly assigned Tactical Plans are the development of a Division
Scorecard (a dashboard to monitor the health of the Division) and to Define the Role of the Regional Councilor. A
change to the Division by-laws was approved in October by the Statistics Division Council which will enable the Division
to appoint more than one Councilor per region. The Regional Councilors serve as Statistics Division ambassadors to the
ASQC local Sections.
In an attempt to better understand why we are continuing to lose members, the Assessing Member Needs tactical
plan seeks to develop a customer satisfaction measurement for the Statistics Division. Member surveys are being conducted by telephone and via mail to gain feedback on our direction and insight to member satisfaction. Exit surveys are being
developed to probe the reasons for lost members. Results from these survey instruments will be fed into the
Division tactical planning process to ensure customer focus. A pilot Member Needs telephone survey was recently
conducted by the Regional Councilors. Though, admittedly, a small sample size, results of this survey indicate an overall
member satisfaction rating of 97.9%. Most members place a high degree of importance, in order, on the Division
Newsletter, the Statistics Divisions involvement in the CQE Body of Knowledge, Short Courses, maintenance of
ANSI/ASQC Standards, and the How to... Series. Perceived lower value benefits of Statistics Division membership
include co-sponsorship of the Applied Statistics Conference, Nomination of Fellows, and the development of Division
awards. Areas where most survey respondents would like the Division to concentrate its efforts are: keep membership
current on new tools and techniques; promote Statistical Thinking everywhere; more Case Studies in the Newsletter; and,
more technical Mini Papers. Two of the least desired potential activities are the development and administration of a
Certified Quality Statistician exam, and software reviews.
The pilot survey was PDCAd during the Tactical Planning Session in October and the improved Member Needs survey
will be mailed to a statistically representative sample of our membership during the month of November. The Member
Needs survey will be conducted quarterly and the results published in the Newsletter.

Robert Mitchell
Membership Committee Chair
18

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

Bill Lawton Receives Statistics


Divisions Hunter Award

Bill Lawton
The 1995 William G. Hunter Award
was presented to William H. Lawton at
the Fall Technical Conference in St.
Louis, MO. The ASQC Statistics
Division established the Hunter Award
in memory of the Divisions founding
chair. The purpose of the award is to
promote, encourage, and acknowledge outstanding accomplishments
during a career in the broad field of
applied statistics. Bill Lawton follows
Bill Hunters model of statistical leadership as a communicator, consultant,
educator and innovator, with the ability to integrate statistical thinking into
many disciplines.
Bill Lawton is a Fellow of both
ASQC and the American Statistical
Association (ASA) and is a past chair
of ASAs Section on Physical
Engineering Sciences (SPES). He
received his Ph.D. in Statistics from
the University of California at
Berkeley. Bill made outstanding contributions during his 28 years with
Eastman Kodak, first as a statistical
consultant to manufacturing and R&D,
and later as a leader who championed
the use of statistical thinking and
methods throughout the company. In
the 1980s he served on Kodaks
Quality Advisory Council, a council of
senior managers reporting to the
President. During the 1990s, Bill was

a Senior Research Associate with


Joiner Associates and Professor of
Marketing at the William E. Simon
School of Business Administration.
During this period, Bill did pioneering
work on quality-based customer
research involving the linking of product development to the voice of the
customer.
Bill is the author of over 20 papers
in the application of statistical
methods in the physical and engineering sciences, business forecasting and
marketing. He received the Wilcoxon
Award in 1971 and 1974 and the
Shewell Award in 1970 and 1980. He
played significant roles as editor of
Technometrics and in helping establish the Fall Technical Conference as a
premiere meeting of practicing statisticians, scientists and engineers. Bill
made these remarks in accepting the
award at the FTC:
I am genuinely grateful and honored to be the 1995 William G. Hunter
Award recipient. As Ted Jackson
noted last year, the citation for this
award characterizes the sort of person
Bill was, and I am pleased to have my
contributions recognized in this manner. Bill was a good friend and professional colleague. We worked
together for over 20 years in the
Chemical Division (now CPID), SPES,
Gordon Research Conference and
Statistics Division. These organizations had a major impact on my
growth as an applied statistician. The
organizations and their associated
conferences provided a valued forum
for the exchange of ideas on how to
further the field of applied statistics.
Bill was master in the use of this type
of forum. There are two aspects of
Bills contributions Id like to mention
today.
First, Bill had intrinsic recognition
that statistics is a tool and the value
created by statistics occurs only in the
use in an application. Hence, Bills
overriding concern was in keeping
any discussions of statistics tied to a

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

field of application. Let me use an


analogy. It may be nice to have a
crescent wrench on a desert island, it
may be beautiful, intellectually satisfying -- but of little value. But when
that same wrench is used to close the
open seacock on a sinking boat, its
value is clear.
Second, Bill was not just a communicator, in my opinion; he was a
master of networking. His personal
contacts from the forum included
application disciplines in the
University, as well as broad industry
and government practitioners of
applied statistics. He used this network tirelessly to promote various
opportunities for the exchange of
ideas in statistical application. This
was demonstrated by his roles in the
formation of the Statistics Division and
the Center for Quality and Productivity
Improvement, and his constant
involvement in the FTC.
We sorely miss his rare skill in this
broad, multidisciplinary networking.
Yet, to some extent, his networking
legacy remains with us today in the
form of the FTC with its program
derived by the networking of CPID,
the Statistics Division, SPES and both
the industrial and academic communities. Keep up the good work, Bill
would be proud. Again, thank you
for this recognition.
Nomination forms for the 1996
award can be obtained from the
William G. Hunter Award Committee
Chair:
Steven P. Bailey
DuPont Engineering
Quality Management &
Technology Center (QM&TC)
Nemours Bulding, Room 6543
1007 Market Street
Wilmington, DE 19898
phone: 302-774-2375
fax:
302-774-2458
e-mail:
baileys@engg.dnet.dupont.com
Nominations must be received no
later than Monday, July 1, 1996.

19

Division Activities
with ASQCs Chemical and
Process Industries Division and
the American Statistical
Associations Section on
Physical and Engineering
Sciences. The conference features new tools and opportunities for applications of statistical
and quality technologies.
1996 FTC - October 23 25, Scotsdale, Arizona

The Statistics Division engages in


many activities in order to achieve
their mission.

Publishing the quarterly


Statistics Division Newsletter
containing technical and nontechnical mini-papers and
basic-tools articles, and information on division activities
and upcoming events.
Editing the Glossary and Tables
of Statistical Quality Control
and the ASQC Basic References
in Quality Control: Statistical
Techniques. The glossary and
16-volume How To series are
available through Quality Press
in Milwaukee.

Sponsoring a technical session


at the Annual Quality Congress.
1996 AQC - May 13 - 15,
Chicago, Illinois

Offering short courses prior to


the Fall Technical Conference
and Annual Quality Congress.

Co-sponsoring the annual Fall


Technical Conference, jointly

Sponsoring young researchers


to attend the Gordon Research

Conference on Statistics in
Chemistry and Chemical
Engineering and students to
attend the Fall Technical
Conference.

Providing a speakers list and


otherwise sponsoring technical
sessions at local conferences,
upon request.

Maintaining and updating


ANSI/ASQC standards for quality programs and statistical procedures.

Presenting the William G.


Hunter Award annually to recognize solid records of achievement in the development and
creative application of statistics.

Statistics Division Job Openings


The Statistics Division is seeking
members to work in various capacities. The job descriptions are printed
below. If you have an interest in any
of these openings or any other activities of the Division, complete the
Member Interest Record Form and
return it to Rick Lewis. Ricks address
is included on the page with the form
itself.

1997 AQC Short Course Chair


Purpose:
To coordinate with ASQC national
in setting up a Statistics Division sponsored short course at the 1997 AQC in
Orlando.
Responsibilities:
1) Identify short course(s) presenter.
20

2) Submit a proposal to ASQC by


August 1996.
3) Serve as the contact between
ASQC and the presenter(s).
4) Moderate the short course at
AQC.

1997 Fall Technical Conference


Program Committee
Representative

Responsibilities:
1) Work on a 3-member FTC
Program Committee to plan a 2day technical program. This
includes issuing a Call for
Papers, inviting speakers and
moderators, accepting or rejecting submitted papers and
preparing the final program.
2) Attend the 1996 and 1997 FTC.
3) Submit items for publication in
the Division newsletter.

Purpose:
The Fall Technical Conference provides a unique forum to present current and emerging quality technologies to a national audience of quality
professionals. It is jointly sponsored
by ASQCs Statistics and Chemical &
Process Industries Divisions and the
American Statistical Associations
Section on Physical and Engineering
Sciences.
ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

Fall Technical Conference &


Tactical Planning Meeting

; Walter
Awardee
t
n
a
r
G
t
den
tive
o Lee - Stu
r - Execu
ht: Chulh ditor; Ray Walle Bailey - Hunter
ig
r
to
ft
eries E
; Steve
From le
ow-To S
ast Chair
Liggett - HSA; Galen Britz - P
A
Director, ir
a
h
C
d
r
a
Aw

John MacGregor
- Youden
Memorial Addr
ess speaker

ies
w-To Ser
oing Ho
tg
u
o
a
kyth
ber
or; Ed My
ade - Edit t - interested mem
h
S
e
ic
n
r
Ja
e
k
ic
tr
on S
Editor; D

Bill Lawton - Hunter Award Winner;


Steve Bailey - Hunter Award Chair

Nancy Belunis; Nick Martino, 1995 FTC Short Course Chair

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

21

Statistics Division Job Openings


The Statistics Division has several job openings for which we are seeking members willing to do some work for the
Division. The openings are as follows:
1.
2.
3.
4.
5.

1997 AQC Short Course Chair - Please see job description elsewhere in the Newsletter
997 FTC Program Committee Representative - Please see job description elsewhere in the Newsletter
Authors and Reviewers for Basic Tools and Mini Paper Articles for the Newsletter
Standards Committee Members - Please see article elsewhere in Newsletter
Authors for How To booklets related to Process Improvement

The Job Descriptions are printed in this Newsletter. If you have an interest in any of these openings, please fill out the
form below and return it to Rick Lewis.
Monsanto Co., Mail Zone 04B, 800 N. Lindbergh Blvd., St. Louis, MO 63167
----------------------------------------------------------------------------------

ASQC STATISTICS DIVISION


MEMBER INTEREST RECORD FORM
Name ______________________________________________________________DATE________________________________
COMPANY __________________________________________________________POSITION____________________________
ADDRESS __________________________________________________________
__________________________________________________________
__________________________________________________________
PHONE (WORK) _____________________________________________________(HOME) _____________________________
FAX _____________________________________________________
MEMBER NUMBER_________________________________ STATUS (MEMBER/SENIOR)______________________________
MEMBER AREAS OF INTEREST
1997 AQC Short Course Chair ______
How-To-Series Author ______
Standards Committee ______
Other ______

1997 FTC Program Committee Rep. ______


Basic Tools/Mini Paper Author ______
Basic Tools.Mini Paper Reviewer ______

RELEVANT EXPERIENCE/EDUCATION
_________________________________________________________________________________________________________
_________________________________________________________________________________________________________
MEMBER TIME AVAILABILITY/COMPANY SUPPORT/TRAVEL ETC.
_________________________________________________________________________________________________________
_________________________________________________________________________________________________________
DATE _________________
OTHER COMMENTS
_________________________________________________________________________________________________________
_________________________________________________________________________________________________________
_________________________________________________________________________________________________________
_________________________________________________________________________________________________________
22

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

ASQC STATISTICS DIVISION


1995-1996
OFFICERS AND COMMITTEE CHAIRS
Chair
Nancy Belunis
Merck & Company, Inc.
One Merck Drive
P.O. Box 100, WS1E-45
Whitehouse Station, NJ 08889-0100
Phone: (908) 423-3423
Fax: (908) 735-1107
belunis@merck.com
Chair-Elect
Beth Propst
1507 East 53rd Street #332
Chicago, IL 60615
Phone: (312) 288-4468
Fax: (312) 288-4468
Secretary
Don Emerling
3M Center
235-3C-23
St. Paul, MN 55144-1000
Phone: (612) 737-2606
Fax: (612) 736-7616
Treasurer
Don Williams
Process Improvement Consultants
2515 Jamestown Lane
Denton, TX 76201-2212
Phone: (817) 382-5992
Fax: (817) 382-5992
Past Chair
Rick Lewis
Monsanto Company
Mail Zone O4A
800 N. Lindbergh Blvd.
St. Louis, MO 63167
Phone: (314) 694-7735
Fax: (314) 694-5614
Membership Chair
Bob Mitchell
3M Consumer Products Plant
915 Highway #22 South
Hutchinson, MN 55350
Phone: (612) 234-1864
Fax: (612) 234-1629
Newsletter Editor
Janice Shade
R.M. Schaeberle Technology Center
200 DeForest Avenue
P.O. Box 1944
East Hanover, NJ 07936-1944
Phone: (201) 503-4915
Education Committee Chair
Chris Ayers
Hamilton Beach/Proctor-Silex, Inc.
4421 Waterfront Drive
Glen Allen, VA 23060
Phone: (804) 527-7158
Fax: (804) 273-9825
Publications Committee Chair
Vacant

Acquisitions Coordinator
Falguni Sharma
8858 South Court
Apt. 303
Allison Park, PA 15101
Phone: (412) 364-0717
Briefings Editor
Rick Lewis
Monsanto Company, Mail Zone O4A
800 N. Lindbergh Blvd.
St. Louis, MO 63167
Phone: (314) 694-7735
Fax: (314) 694-5614
How-To Series Editors
Walter Liggett
Statistical Eng. Division
Computing and Applied Mathematics
Administration 101, Rm. 339
National Institute of Standards &
Technology
Gaithersburg, MD 20899
Bob Brill
Monsanto Company
Mail Zone T1B
800 N. Lindbergh Blvd.
St. Louis, MO 63167
Phone: (314) 694-1684
Glossary Editor
Jim Bossert
5650 Alliance Gateway
Fort Worth, TX 76178
Phone: (817) 490-7147
Awards Committee Chair
Lynne Hare
21212 Chrisman Hill Terrace
Boyds, MD 20841
Phone: (301) 975-2840
Fax: (301) 990-4127
William G. Hunter Award Chair
Steve Bailey
DuPont Engineering, QMTC
Nemours Building, Room 6543
1007 Market Street
Wilmington, DE 19898
Phone: (302) 774-2375
Fax: (302) 774-2458
Standards Committee Chair
Ed Schilling
Rochester Institute of Technology
Center for Qual. & Applied Stat.
1 Lomb Memorial Drive, Bldg. 14
P.O. Box 9887
Rochester, NY 14623-0887
Phone: (716) 475-6129
Examining Committee Chair
Bob Perry
Pillsbury Company
330 University Avenue S.E.
Minneapolis, MN 55414
Phone: (612) 330-8144
Fax: (612) 330-8294

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

Certification Chair
Nick Martino
Novacor Chemicals Inc.
31 Fuller Street
Leominster, MA 01453
Phone: (508) 534-2556
Fax: (508) 840-0112
ASA Q&P Liaison
Joe Voelkel
Rochester Inst. of Technology
Ctr. for Quality and Applied Stat.
1 Lomb Memorial Drive
Bldg. 14, P.O. Box 9887
Rochester, NY 14623-0887
Phone: (716) 475-2231
1996 Fall Technical Conference Program
Jacob Van Bowen
Statistics and Computer Science
University of Richmond
Richmond, VA 23173
Phone: (804) 289-8081
Fax: (804) 287-6444
1996 Fall Technical Conference - Short
Course
Bill Bleau
Picker International Inc.
1130 Stonecrest Drive
Tallmadge, OH 44278
Phone: (216) 473-2385
Conference on Applied Statistics Program
Frank Alt
University of Maryland
College of Bus. & Management
College Park, MD 20742
Phone: (301) 405-2231
1996 Annual Quality CongressProgram
Lori Coons
Eastman Kodak Company
QTAS/MQAQ
Building 6, 7th Floor
Kodak Park
Rochester, NY 14652-4608
Phone: (716) 722-5217
1996 Annual Quality Congress Short Course
Carol Meeter
3M Center
224-4S-19
St. Paul, MN 55144-1000
Phone: (612) 736-6297
Bulletin Board Administrator
Mark Kiel
Acme Steel Company
13500 S. Perry Avenue
Riverdale, IL 60627-1182
Phone: (708) 841-8383 ext. 216
Fax: (708) 841-0661

23

STATISTICS DIVISION
AMERICAN SOCIETY FOR
QUALITY CONTROL
c/o Janice Shade
200 DeForest Avenue
P.O. Box 1944
East Hanover, NJ 07936-1944

Non-Profit Org.
U.S. Postage

PAID
Cedarburg, WI
Permit No. 199

The ASQC Statistics Division Newsletter


is published quarterly by the Statistics
Division of the American Society for
Quality Control.
All communications regarding this publication, EXCLUDING CHANGE OF
ADDRESS, should be addressed to:
Janice Shade, Editor
ASQC Statistics Division Newsletter
Nabisco, Inc.
200 DeForest Avenue
P.O. Box 1944
East Hanover, NJ 07936-1944
Phone: (201) 503-4915
Fax:
(201) 503-4884
Other communications relating to the
Statistics Division of ASQC should be
addressed to:
Nancy Belunis
Merck & Company, Inc..
One Merck Drive
P.O. Box 100, WS1E-45
Whitehouse Station, NJ 08889-0100
Phone: (908) 423-3423
Fax:
(908) 735-1107
Communications regarding change of
address should be sent to ASQC at:
American Society for Quality Control
P.O. Box 3005
Milwaukee, WI 53201-3005
This will change the address for all
publications you receive from ASQC
including the newsletter. You can also
handle this by phone (414) 272-8575 or
(800) 248-1946.

UPCOMING NEWSLETTER
DEADLINES
Issue

Vol. No. Due Date

Summer 96
Fall 96
Winter 97

16
16
16

3
4
5

May 17, 1996


Aug. 16, 1996
Nov. 18, 1996

E
24

Printed on Recycled Paper

ASQC STATISTICS DIVISION NEWSLETTER, VOL. 16, NO. 2

También podría gustarte