The Modern Practice of Statistics in Business and Industry

1
The Modern Practice of Statistics in Business
and Industry

• Douglas C. Montgomery
• Professor of Engineering Statistics
• Arizona State University
• doug.montgomery_at_asu.edu

2
Background

• Todays statistician lives and works in
  different/changing times
• Widespread availability/use of statistical
  software by nonstatisticians
• The democratization of statistics (six-sigma)
  everybodys doing it
• Expanding scope of problems in which statistics
  plays a role
• These changes cannot be ignored
• How to play a leadership role?

3
The New Environment

• Lots of people use statistics the techniques are
  no longer exclusively the province of
  statisticians
• Applications in distribution systems, financial,
  and services are becoming at least as important
  as applications in manufacturing and RD
• Statistical Thinking in management decision
  making is becoming just as important as the
  actual use of statistical methods
• Data-driven decision-making
• In God we trust, all others bring data

4
The New Environment

• Statisticians are needed
• Sometimes even wanted, respected (loved?)
• But not just to analyze data, design experiments,
  etc
• Non-statisticians often do that for themselves
• The scope of professional practice is changing,
  expanding
• So the options are lead, follow, or get out of
  the way

5
Some Contrasts
Then Now
Narrow (operational) focus Broad, strategic focus
Consultant Team leader, facilitator
Design experiments, analyze data Help define problems, tools to be employed
Teach statistics to small groups Develop/implement broadly based systems (six sigma)
Technical clients Work with managers
Narrow application of professional skills Broader application of an expanded skill set is expected
Limited accountability Great accountability
Low visibility (under radar), few opportunities High visibility, potentially many opportunities
6
Business/Industry Drivers

• Flattening (delayering) of organizations
• Less staff, fewer consultants technical experts
• More operational accountability
• Shift from manufacturing to service economy
• Impacts even traditional manufacturers
• Supply chain management critical (domestic
  content issues)
• Drive to create value for stakeholders
• More broad application of basic tools
• Perhaps fewer applications of advanced tools

7
Business/Industry Drivers

• Data-rich, highly automated business and
  industrial environment
• Semiconductor manufacturing process
• Fabrication process typically has 200 steps
• Assembly and test required to complete product
• 1000s of wafers started each week
• In-process, probe, parametric, functional test
  data available

8

• Taxonomy of methods
• data collection
• data analysis/manipulation
• data storage
• data warehousing
• data mining
• data drilling leading to
• data blasting, and finally
• data torturing

Traditional statistics courses
9
The multivariate nature of process data

• We dont recommend one-factor-at-a-time
  experiments, why do we use lots of univariate
  control charts?
• This has implications for academic programs, what
  we teach students
• Emphasis on small sample sizes, hypothesis
  testing, P-values, etc

10
Business/Industry Drivers

• Extend use of statistical methods into
  engineering design and development
• Methods for reliability improvement continue to
  be of increasing importance - driven by customer
  expectations
• Reliability of software, process equipment
  (predictive maintenance) are major considerations
• Reducing development (cycle) time
• Robustness of products and processes are still
  important problems
• DFSS a growing emphasis

11

• Traditionally the industrial statistician has
  been an internal consultant
• Often viewed primarily as a manufacturing
  person
• This perspective is changing as statistical
  methods penetrate other key areas, including
• Information systems
• Supply chain management
• Transactional business processes
• The statistician's role is changing as well
• Six-sigma activities have played a part in this

12

• Its important to be a team member (or
  facilitator, leader) and not just a consultant
• The mathematics orientation of many statistics
  programs does not make this easy
• Quote from Craig Barrett (INTEL)
• To be successful at INTEL, the statisticians
  need to be better engineers
• Statisticians still often
• Do not share in patent awards/recognition, other
  incentives
• Not viewed as full team members
• Regarded as merely data technicians

13
Some Must Background/Courses for Modern
Industrial Statisticians

• Preparation for professional practice
• Design of Industrial Experiments
• Emphasis on factorials, two-level designs,
  fractional factorials, blocking
• Random effects, nesting, split plots
• Response Surface Methodology
• Traditional RSM, philosophy, methods, designs
• Mixture Experiments
• Robust design, process robustness studies

14
Some Must Background/Courses for Modern
Industrial Statisticians

• Reliability Engineering
• Survival data analysis, life testing
• RAM principles
• Design concepts
• Modern Statistical Quality Control
• Analysis of Massive Data Sets
• Traditional multivariate methods
• CART, MARS, other data mining tools
• Categorical Data Analysis, GLM

15
Some Must Background/Courses for Modern
Industrial Statisticians

• Forecasting, Time Series Analysis Modeling
  (should overview a variety of methods, include
  system design aspects)
• Discrete Event Simulation
• Principles of Operations Research
• Basic optimization theory
• Linear nonlinear programming
• Network models

16

• I have just outlined about 27 semester hours of
  graduate work!!
• Most MS programs require 30 hrs beyond the BS
  (non-thesis option), 24hrs with thesis
• PhD programs require a minimum of 30 hrs of
  course work beyond the MS
• Academic programs would need to be significantly
  redesigned if a serious effort is going to be
  made to educate industrial statisticians

17

• Where do graduates go?
• Lots of places business and industry,
  government, academia
• But few of them will be theorists or
  teach/conduct research in theory-oriented
  programs
• So why do many graduate programs operate as if
  all of them will?
• More flexibility is needed

18

• Most PhD programs require a minor (sometimes two,
  sometimes out-of-department)
• Require that this be in engineering,
  chemical/physical science, etc.
• Most departments will be interested in setting
  these up
• Could also work at MS level
• Certificate programs

19

• Recruit engineers/scientists/ORMS majors for
  graduate programs in statistics
• But graduate programs had better be meaningful!
• Significant program redesign will be required
• Alternative develop joint graduate
  (degree/certificate) programs with engineering
  departments, business schools

20
The ASU Graduate Certificate Program in Statistics

• Students take five approved courses
• Certificate can be pursued as part of a graduate
  degree or as a stand-alone program
• Emphasis area in industrial statistics and
  six-sigma methods is available

21
Industrial Statistics Six-Sigma

• Design of Experiments
• Regression Analysis
• Statistical Quality Control
• Shewhart control charts
• Measurement systems analysis
• Process capability analysis
• EWMAs, CUSUMs, other univariate techniques
• Multivariate process monitoring
• EPC/SPC integration

22
Industrial Statistics Six-Sigma

• Six-Sigma Methods
• How to use tools (case studies, illustrations)
• DMAIC framework
• Non-statistical skills
• Design for six-sigma, lean concepts
• Taught by six-sigma black belts from industry
• Six-Sigma Project
• 150 hour duration
• Typical industrial BB project
• Must use DMAIC approach, statistical tools
• Supervised by faculty industrial sponsor

23
Project Examples

• Develop web-based decision system for deployment
  of statistical tools
• Reduce average internal cycle time of instrument
  calibration lab
• Develop prediction model for rate of customer
  returns to quantify benefits of yield and test
  coverage improvements, and to identify parts
  within a technology that do not fit the model

24
Increasing the Power of Statistics
A force F acting through a distance s performs
work W Fs
F
s
25
Increasing the Power of Statistics
F
s
Power is a measure of how fast work is done
26
Increasing the Power of Statistics
More force more power More distance more
power Shorter time more power
How well can we apply force to this
opportunity? How much leverage (distance) can we
generate? How quickly can we apply it?
27
Statistics in Business and Industry

• Use of statistical methods (thinking?) is routine
• Statisticians can be leaders, change agents
• Logistics/service/financial applications are
  growing rapidly
• This requires a different type of professional
  with different skills
• There are significant challenges in preparing
  these individuals for profession practice
• Statisticians are valued and needed