Escolar Documentos
Profissional Documentos
Cultura Documentos
Rev. 1.0
This document and all information and expression contained herein are the property of ASU Department of Industrial Engineering, and may not, in whole or in part, be used,
duplicated, or disclosed for any purpose without prior written permission of ASU Department of Industrial Engineering. All rights reserved.
ASU Department of Industrial Engineering 2012
Define Phase
Paul Sandell (paul.g.sandell@intel.com)
Geetha Rajavelu (geetha.rajavelu@bankofamerica.com)
ASU Department of Industrial Engineering 2012
2
Objectives
What is Six Sigma
Pre-Define: Project ideation and Prioritization
Identify the elements of the Define phase
Discuss tollgate elements completed in Define
Discuss some Define tools
ASU Department of Industrial Engineering 2012
3
What is Six Sigma?
Six Sigma
DMAIC Lean
DMADV
Six Sigma is a problem solving and process improvement methodology that helps
improve our products, processes, services, and people, by reducing variation and
eliminating defects, and waste!
In completing Process Improvement Projects, Six Sigma uses three approaches
DMAIC (Define-Measure-Analyze-Improve-Control) When we have an existing process that
is not meeting customer requirements
DMADV (Define-Measure-Analyze-Design-Verify) When we are designing a new process, or
completely re-designing an existing process
Lean Principles to reduce waste and accelerate the velocity of a process
ASU Department of Industrial Engineering 2012
4
Start Date:
_________________
Project Charter
Problem Statement
Goal Statement
In/Out of Scope
Team & Time
Commitments
Timeline / Milestone
Estimate Financial
Benefits
Risks, Constraints &
Compliance Issues
Identified
SIPOC
High Level Process Map
Start Date:
_________________
Y & Defect Defined
Performance Spec for Y
Data Collection Plan
Measurement System
Validated
Collect data for Y
Process Capability for
Y
Improvement Goal for Y
Detailed Process Map
Tollgate Review Date:
________________
Start Date:
________________
Brainstorm all possible
Xs
Prioritize list of Xs
ID gaps
Refine benefits
estimate
Tollgate Review Date:
_________________
Start Date:
______________
Develop and select
solutions
Perform pilot of
solutions
Confirm improvements
Confirm prioritized Xs
Map new process
Develop
implementation plan
ID controls
Implement
improvements
Start Date:
_____________
Process documentation
/ Controls in place
Measure performance
Confirm sustainable
solution
Transfer ownership
Evaluate translation
opportunities
ID other improvement
opportunities
Project documentation
complete
Financial benefits
verified and approved
Leveragability
Tollgate Review Date:
_________________
Define Control Improve Analyze Measure
Not Started In Progress Complete
DMAIC Tollgate Checklist
ASU Department of Industrial Engineering 2012
5
Life Before Define
One of the most critical aspects of a successful Six Sigma
deployment is to select the right project(s)
Effective project ideation, prioritization and selection leads to Six Sigma
project results
Many companies fail before they even start Define
ASU Department of Industrial Engineering 2012
6
Generate
Ideas
Prioritize Launch
2 1 3
The high level process.details to follow!
Project Selection Roadmap
ASU Department of Industrial Engineering 2012
7
Generate
Ideas
Prioritize Launch
CTQ flow from Strategic
Plan
Financial Analysis
Performance
Metrics
Things that keep you up at
night (Organic)
Potential project ideas Projects
Project Ideation Methods
ASU Department of Industrial Engineering 2012
8
CTQ Flow Down Process
What A process in which strategic goals of the organization are used and a statistical relationship is
determined to describe how the strategic goal is improved by completing the project.
How and Who A trained MBB or BB would partner with the key business leaders (Champions) and
process owners, to establish the linkage from strategy to project ideas.
How Often Should be completed at least annually, and updated as business strategies change
Issues This process can take from week to months to adequately complete!
This is our most essential voice of the customer linkage to the business!
Strategy 1 Strategy 2
Business Need 1 Business Need 2 Business Need 3
Project Idea 1 Project Idea 2 Project Idea 3 Project Idea 4 Project Idea 5
ASU Department of Industrial Engineering 2012
9
Financial Analysis
What A process that reviews keys financial indicators for the business to identify project
opportunities
How and Who A financial leader would partner with the key business leaders (Champions) and
process owners, to establish the linkage from strategy to project ideas. An MBB or BB can be used to
help facilitate the process.
How Often Completed at least annually, and as frequent as quarterly
Issues Potential introduction of variation
This is a voice of the business process to generate project ideas
ASU Department of Industrial Engineering 2012
10
Performance to Plan
What A process that reviews metrics of existing performance to the business plan, and develops
project ideas based on performance gap to the plan
How and Who Process owners and key business leaders review the gaps, primarily during operational
reviewsactions (projects) are typically an output of the process
How Often Quarterly
Issues Potential introduction of variation
Another voice of the business process to generate project ideas
ASU Department of Industrial Engineering 2012
11
Organic Project Path
What A process that uses structured brainstorming to bubble up project ideas at all business levels
How and Who Process owners (with MBB or BB assistance as necessary) facilitate their work teams
through the process
How Often Quarterlyuntil the process becomes a natural part of the culture
Issues Can be great for team buildingdont let the process become a complete whining session
A creativity process, based on business pain points
ASU Department of Industrial Engineering 2012
12
Yes
Six Sigma Project checklist
Key driver of Customer Satisfaction focused
Narrow scope
Available metrics or measurements that can be developed
quickly
Control of the process owned by the Champion
Recurring events
Linked to corporate or business unit objectives
Financial benefits are traceable
Solution unknown
ASU Department of Industrial Engineering 2012
13
How Do We Prioritize?
Ideas Prioritize Launch
Ensure projects are linked to a
companys businesses?
With ideas generated through multiple
methods, we are ready to score the
ideas against one another.so what is
the process?
Ideas
Become
Projects
ASU Department of Industrial Engineering 2012
14
Prioritization of Projects
Now that we have a list of projects (project pipeline) how do we
decide to do which ones first??
Prioritization A system is needed to gauge the relative customer, business, team
member and time impact of each project idea.
Best practice organizations develop filters or criteria to complete this assessment,
with a numerical importance value attached to the criteria.
Panning for the gold nuggets in our business!
ASU Department of Industrial Engineering 2012
15
Proposed criteria will be used to prioritize identified projects
Criteria Description Worse Better
Potential
Impact on
Employee
Satisfaction
Relative impact to employee satisfaction
Potential Impact
on Customer
Metrics
Relative benefits and impact to key
business drivers, when compared
customer requirements
1 3 9
No change Improve < 20% Improve > 20%
Time to
Implement
Expected time required to fully implement
the project
Savings or
Revenue
Approximate savings or revenue obtained
Potential
Impact on
Experience
Relative impact to customer experience
Weight
20%
30%
10%
20%
20%
C
u
s
t
o
m
e
r
F
i
n
a
n
c
i
a
l
E
m
p
l
o
y
e
e
P
r
o
c
e
s
s
1 3 9
No change Improve < 20% Improve > 20%
1 3 9
< $50k > $50k < $200k > $200k
1 3 9
No change Improve < 20% Improve > 20%
1 3 9
9+ months 3-9 months 0-3 months
ASU Department of Industrial Engineering 2012
16
Project Launch
Ideas Prioritize Launch
Identification of Project Champion
Champion drafts charter
. Champion selects belt and attend pre-training
Team members are identified
Pre-Launch Checklist
Output is Project Kickoff!
ASU Department of Industrial Engineering 2012
17
Black Belt vs Green Belt Project
Black Belt Project
Full time BB resource
Project scope is broad
Typically more complex
Large business impact
3-6 months to complete
Green Belt Project
Part time GB resource
Project scope is narrow
Typically less complex
Localized business impact
3 months to complete
Champion supports the determination of BB Vs GB project!
ASU Department of Industrial Engineering 2012
18
Review of the Process
Lets take a few minutes to review our process
1. Business strategy and operation plans established
2. Ideas generated based on key processes identified
3. Prioritization completed
4. Business drafts charters and determines project methodology
5. Projects launched
Our projects are established, and we are on our way to Define!
ASU Department of Industrial Engineering 2012
19
What Happens In Define?
In the Define phase of DMAIC there are three
key elements that we seek to understand
Clarify the business problem (i.e. opportunity)
Identify and validate customer requirements
Begin documentation of the process
ASU Department of Industrial Engineering 2012
20
Define for Design Approach (DMADV)
Objectives
Define a vision and project strategy to achieve the vision
Review your Design Charter
Opportunity, goal, scope
Validated by leadership
Identify initial design requirements
based on Voice of the Customer
Document the process where design
will focus
ASU Department of Industrial Engineering 2012
21
Define: DMAIC Vs DMADV
Define in DMAIC
Define the business
problem/goals and customer
deliverables
Documents the existing process
and existing customer needs
Define in DMADV
Define the business
problem/goals and customer
deliverables
Determines unknown customer
needs and links to future design
ASU Department of Industrial Engineering 2012
22
Charter Elements Buttoned Down
We have (definitely) completed a draft charter before the belt
attends class, and we want to refine and clarify as appropriate
Problem statement
Goal statement
Scope
Team & time commitments
Project plan (timeline)
Estimated benefits
Define: Measure: Analyze: Improve: Control:
Start End
Start End Start Revised
One time Impact Annual One time Impact Annual
0.00 0.00 0.00 0.00 0.00 0.00
Other
Other
Other
Defect Definition
Revenue Growth
Other
Materials
Other
Other
PROJECT OVERVIEW
Problem Statement:
Identify Business Strategic Linkages & Secondary Business Impacts:
Project Goal:
Project Milestones
Project Title:
Project Leader: Champion:
BENEFITS SUMMARY ($000)
Indirect (Soft) Type B
Productivity
Cost Avoidance
Benefits Category
Direct (Hard) Type A
PROJECT DETAILS
In / Out of Scope
Team Members & Resources
Opportunity Definition
Sign-offs: Open Close
Coaching MBB: _____________________ __________________________
Champion: ____________________ __________________________
Business Leader: ____________________ __________________________
Finance Rep. ____________________
Other
One Wells Fargo Customer Impact
You know me
TOTAL BENEFITS:
(For example: $ per resource requirements)
Other
Other Key Project Measurements:
Overall Benefits: 0.00
Baseline Who is the Customer & What is the Customer Impact?:
Process Owner:
Idea Document Completed?:
Six Sigma Project Charter
MBB / Coach:
Location: Project Start Date:
Project End Date:
Planned Completion Dates, by project phase:
PROJECT MILESTONES
Project Type: Project ID No.:
Process:
Goal
`
Project Metrics (can include secondary metrics)
Rev. 2.2
ASU Department of Industrial Engineering 2012
23
Lets Look At A Real Green Belt Project
Problem statement: The current lack of procedures for completing
emergency installs to production systems creates a risk of user and customer
impacts, down-time and inadequate communication.
Project objective / goal: Design and implement an emergency installs
process that will allow for quick and accurate resolution of high severity
production issues and create objective tracking of results.
ASU Department of Industrial Engineering 2012
24
Takeaway
What is your assessment of this Problem and Goal? Why?
ASU Department of Industrial Engineering 2012
25
Project Plan
The roadmap on the journeywe want all belts to complete a
project plan!
Six Sigma Project Plan
Scheduled
Start
Scheduled
Finish
Actual
Start
Actual
Finish
Estimated
Duration
Tool or
Task Define Day 1 Day 35
Task Champion Identifies business oportunity linked to Business strategy On going On going
Tool Champion drafts rough cut charter Day 1 Day 2
Task Champion selects belt candidate Day 2 Day 2
Task Belt is assigned to training wave Day 2 Day 2
Task Belt assigned MBB coach Day 3 Day 3
Tool Champion/Belt review and modify charter Day 4 Day 9
Task Problem statement definition complete Day 4 Day 4
Task Project goal definition complete Day 4 Day 4
Task Project defect definition complete Day 4 Day 4
Task Project scope complete Day 4 Day 4
Task Key output metric and customer benefits complete Day 4 Day 4
Task Project benefits estimated Day 4 Day 9
Task Review benefits estimate with finance Day 5 Day 5
Task Finalize benefits and obtain finance signoff Day 9 Day 9
Task Champion and Belt determine project resources Day 5 Day 9
Task Final project signoff with Champion and MBB coach Day 10 Day 10
Task Meeting schedule determined with MBB coach Day 10 Day 10
Tool Project plan complete Day 10 Day 15
Task Kickoff meeting held with team and customer Day 11 Day 11
Task Roles clarified Day 11 Day 11
Tool Issue/Action/Risk log initiated Day 11 Day 11
Task Customer requirements obtained Day 12 Day 15
Tool SIPOC completed Day 15 Day 15
Tool Survey completed Day 15 Day 35
Tool High level "as is" process map complete Day 16 Day 16
Measure Day 11 Day 44
ASU Department of Industrial Engineering 2012
26
Project Benefits
A critical element in definehelps clarify business value!
There are two types of project benefits.
Customer satisfaction (includes internal customersteam member
satisfaction)
Financial
A Six Sigma project should have at
least one if not both of these benefits!
Both benefit types are the result of improved PROCESSES
ASU Department of Industrial Engineering 2012
27
Project Benefits
Customer Satisfaction A measurable result of the belt project,
would be higher levels of satisfaction.
Financial
Productivity
Cost or Growth
Cost Avoidance
Materials
Physical or vendor costs
Best practice organizations measure annualized benefits
ASU Department of Industrial Engineering 2012
28
Risks, Constraints & Compliance Issues
Risk Impact Probability Risk Score Mitigation Actions
Six Sigma Project: Risk, Constraints, Compliance Issues
Why do we discuss this as part of Define??
ASU Department of Industrial Engineering 2012
29
Document the Process
Two critical tools that belts use to document the process (and
which Leaders should understand are.
SIPOC
Process Map (high level)
Lets take a look at both!
ASU Department of Industrial Engineering 2012
30
A Process Is Defined As...
...A series of tasks or activities whereby one thing (the
input) is changed or used to create something else (the
output)
ASU Department of Industrial Engineering 2012
31
The SIPOC is a tool that documents a process from suppliers to
customers. Once completed, it is used to:
Identify and balance competing customer requirements.
Aid in identification of data collection needs.
See process connected to the Customer
Avoids getting stuck in detail
A simplified view of entire process visible at a glance
Help provide scoping direction on projects
SIPOC Defined
Suppliers Inputs - Process Outputs - Customers
ASU Department of Industrial Engineering 2012
32
Steps to Diagram a SIPOC
1. Identify the Process to be diagrammed and name it
Write that in the Process Name
Complete other information at top of form
2. Define the Outputs and Inputs (boundaries):
Start at the END: Customer(s) and key Output(s)
Supplier(s) and key Input(s)
3. Clarify the Requirements (optional, but recommended)
What are key features/characteristics of the Output for each Customer?
4. Establish ~2-5 high-level Process Steps
Brainstorm major process activities on sticky notes
Group or organize activities into similar categories or
major steps in the process (Suggestion: use Affinity method)
Place major steps in most appropriate order
ASU Department of Industrial Engineering 2012
33
SIPOC Form
ASU Department of Industrial Engineering 2012
34
SIPOC Example
Project Title: Project Champion:
Process Owner: Project Belt:
Core Process: Project Number:
SUPPLIERS INPUTS OUPUTS
(Providers of the
required resources)
(Resources required by
the process)
(Deliverables from the
process)
Requirements Requirements
Knowledgeable Baked Cookies Soft/chewy Kids
Spouse
Spouse
Spouse
Soft/chewy
Warm
Clean
Baked Cookies
Baked Cookies
Messy Kitchen
Avaliable
Avalible/quality
Available
Available/Working
Cook
Recipe/Book
Ingredients
Utinsils
Oven
Food store
Retail store
Appliance store
Family
Amazon
SIPOC
Cookie
Betty Crocker
Cookie Baking
Martha Stewart
Rachael Ray
1
(Top level description of activity)
(anyone who receives a deliverable from
the process)
PROCESS CUSTOMERS
Timer Dings
Bake Cookie
Dough
Obtain
Ingredients
1
1
9
5
8 6 7 4 3 2
ASU Department of Industrial Engineering 2012
35
Graphical representation of a process
Identifies Key Process Input Variables (KPIVs, also called your little xs)
Identifies Key Process Output Variables (KPOVs, also called your little
ys)
First process map should be as is
Ensure process is walked. A business process can be walked, by
representing information transfer and modification points. Team should not
assume they know the process well enoughwalk it.
The result should encompass a process map that identifies KPIVs and
KPOVs. Critical KPOVs should be linked to customer CTQs
Process map can have other information identified on it as well as
information the team feels is appropriate (ie. Data collection points)
Take advantage of tribal knowledge held by those who work the process
What is a Process Map?
ASU Department of Industrial Engineering 2012
36
High Level Process Map
The high level process map builds upon our SIPOC by
seeking to show the primary sequence of events in the process
A high level go to ASU process
ASU Department of Industrial Engineering 2012
37
A graphic representation of process that details decision points, lists
and classifies KPIVs (little xs) and lists KPOVs (little ys)
Detailed Process Mapping
Rep answers
phone
Rep greets
customer
Rep determines
product need
Cust identify need
date
Rep obtains
customer info and
amount
Rep obtains
internal
information
Rep determine
terms
Rep verifies
information
Rep completes
request worksheet
Rep inputs order
entry info
Rep prints order
confirmation
Rep determines
ship date
Rep reviews order
Rep faxes
confirmation to
customer
Rep verifies
manufacturing
receipt
Phone - SOP
CSR - N
Greeting - SOP
CSR - N
Customer - N
Product Infomation - C
CSR - N
Customer - N
Answred phone
Customer greeted Prod. need obtained
Date obtained
Cust info obtained
Internal info obtained
Terms completed
Info validated Completed worksheet
Order entered
Confirmation printed
Ship date
Order reviewed
Confirmation faxed
Receit verified
System - SOP
CSR - N
Customer - N
System - SOP
CSR - N
Customer - N
System - SOP
CSR - N
Customer - N
System - SOP
CSR - N
Customer - N
Term info - N
Fax machine- SOP
CSR - N
Customer - N
Order - C
CSR - N
Customer - N
Order - C
CSR - N
Customer - N
Order - C
Printer - SOP
Customer - N
Order - C
CSR - N
CSR - N
Customer - N
Order - C
System - SOP
CSR - N
Customer - N
Worksheet - C
CSR - N
Order - C
CSR - N
Receipt - SOP
Inputs
Outputs
ASU Department of Industrial Engineering 2012
38
Define Completed
With these elements completed, the Define phase is essentially
complete.why do we say essentially?
Project Charter
Problem Statement
Goal Statement
In/Out of Scope
Team & Time Commitments
Timeline / Milestone
Estimate Financial Benefits
Risks, Constraints & Compliance Issues Identified
SIPOC
High Level Process Map
ASU Department of Industrial Engineering 2012
39
Define
"What we think..."
Purpose: Properly define project in terms of Project
purpose, scope, objective & customer CTQ's are
stated. Processes or product to be improved
identified.
Key Outputs:
Customer / Business CTQ's
Project Charter
SIPOC
Process Map
To Measure
Corporate Vision &
Objectives Set
Dept. A Dept. A
Dept. B Dept. B
Dept. C Dept. C
Dept. D Dept. D
Dept. E Dept. E
Cycle Time Cycle Time
2 days 2 days 4 days 4 days 3 days 3 days 3 days 3 days 2 days 2 days
1 1
2 2
3 3
4 4
10 6 8
Project
Home Mortgage Defects 9 3 9 180
0
0
0
C
T
Q
C
u
s
t o
m
e
r I m
p
a
c
t
F
i n
a
n
c
i a
l I m
p
a
c
t
E
m
p
l o
y
e
e
I m
p
a
c
t
C
T
Q
Total
C
T
Q
Cause and Effect Matrix Project Prioritization
Business Groups Establish Objectives
supporting Corporate Objectives
Belt Candidate Selected
Projects Selected & Prioritized
Champion or Champion & Belt
complete Project Charter. Customer &
Business CTQ's become part of
project.
High Level Process Flow Diagram to
begin understanding the process
Customer Requirements & CTQ's
determined
Complete SIPOC which helps to
scope project & ID measurement
points as well as customers
process inputs & outputs
Detailed Process Map with
inputs and outputs identified
Answer phone Greet customer Determine product
need Identify need date Obtain customer
info and amount
Obtain internal
information Determine terms
Verify information Complete request
worksheet
Input order entry
info
Print order
confirmation
Determine ship
date Review order Fax confirmation
to customer
Verify
manufacturing
receipt
Phone - SOP
CSR - N
Greeting - SOP
CSR - N
Customer - N
Product Infomation - C
CSR - N
Customer - N
Answred phone Customer greeted Prod. need obtained Date obtained Cust info obtained Internal info obtained Terms completed
Info validated Completed worksheet Order entered Confirmation printed Ship date Order reviewed Confirmation faxed
Receit verified
System - SOP
CSR - N
Customer - N
System - SOP
CSR - N
Customer - N
System - SOP
CSR - N
Customer - N
System - SOP
CSR - N
Customer - N
Term info - N
Fax machine- SOP
CSR - N
Customer - N
Order - C
CSR - N
Customer - N
Order - C
CSR - N
Customer - N
Order - C
Printer - SOP
Customer - N
Order - C
CSR - N
CSR - N
Customer - N
Order - C
System - SOP
CSR - N
Customer - N
Worksheet - C
CSR - N
Order - C
CSR - N
Receipt - SOP
ASU Department of Industrial Engineering 2012
40
Summary
Reviewed and discussed the elements in the Define
Phase
Demonstrated appropriate applications of the Six Sigma
tools in the Define Phase
ASU Department of Industrial Engineering 2012
41
Appendix Completed Charter
Home and Consumer Finance Group
Rev. 1.0
This document and all information and expression contained herein are the property of ASU Department of Industrial Engineering, and may not, in whole or in part, be used,
duplicated, or disclosed for any purpose without prior written permission of ASU Department of Industrial Engineering. All rights reserved.
ASU Department of Industrial Engineering 2012
Six Sigma Define
2012
This document and all information and expression contained herein are the property of ASU Department of Industrial Engineering, and may not, in whole or
in part, be used, duplicated, or disclosed for any purpose without prior written permission of ASU Department of Industrial Engineering. All rights reserved.
ASU Department of Industrial Engineering 2004
IEE 581
Six Sigma Methodology
DMAIC Measure Phase
Dr. Harry Shah
President and Master Black Belt
Business Excellence Consulting LLC
harry.shah@bizxlnc.com
www.bizxlnc.com
ASU Department of Industrial Engineering 2004
DMAIC - Process Improvement Roadmap
What is
important?
How are
we doing?
What is
wrong?
What
needs to
be done?
How do we
guarantee
performance?
1.0
Define
Opportunities
2.0
Measure
Performance
3.0
Analyze
Opportunity
4.0
Improve
Performance
5.0
Control
Performance
ASU Department of Industrial Engineering 2004
Measure Performance
Key Deliverables
Input, Process, and
Output Indicators
Operational
Definitions
Data Collection
Formats and Sampling
Plans
Measurement System
Capability
Baseline Performance
Metrics
Process Capability
DPMO
PLT
PCE
Yield/Scrap
Others
Productive Team
Atmosphere
Inputs
Team Charter
Business case
Goal statement
Project scope
Project plan
Team roles and responsibilities
Prepared Team
Critical Customer Requirements
Process Maps
Quick Win Opportunities
2.0 Measure Performance
Determine What to
Measure
Manage Measurement
Evaluate Measurement
System
Determine Process
Performance
ASU Department of Industrial Engineering 2004
Determine What to Measure
ASU Department of Industrial Engineering 2004
Determine What to Measure
SIPOC Diagram
Common Elements to All Processes
Supplier
Input
Process
Output
Customer
ASU Department of Industrial Engineering 2004
Case Study - Coffee Example
A fast food restaurant conducted an annual customer
survey. There was an overwhelming response from
customers. A good percentage of the responses were
favorable. The customers liked their service and food.
Other customers complained that the coffee served by
the restaurant was not consistent in taste. As a result
some customers stopped patronizing the restaurant.
Owners son, is enrolled in Six Sigma Methodology
course at ASU. He decided to tackle the problem. The
process consisted of Coffee Brewing.
ASU Department of Industrial Engineering 2004
SIPOC Diagram
Process
Supplier
Input
Output
Customer
Case Study - Coffee Example
Coffee
Mfg
Filter Mfg
Water
Supplier
Coffee
Filter
Water
Brewed
Coffee
Patron
ASU Department of Industrial Engineering 2004
Determine What to Measure
SIPOC Diagram
Common Elements to All Processes
Supplier
Input
Process
Output
Customer
Input Indicators Process Indicators Output Indicators
ASU Department of Industrial Engineering 2004
Determine What to Measure
Input Indicators
Measures that evaluate the degree to which the inputs
to a process, provided by supplier, are consistent with
what process needs to efficiently and effectively
convert into customer satisfying outputs.
ASU Department of Industrial Engineering 2004
Determine What to Measure
Process Indicators
Measures that evaluate the effectiveness, efficiency,
and quality of the steps and activities used to convert
inputs into customer satisfying outputs.
ASU Department of Industrial Engineering 2004
Determine What to Measure
Output Indicators
Measures that evaluate the effectiveness of the output.
ASU Department of Industrial Engineering 2004
Case Study - Coffee Example
Input
Indicators
Coffee
Manufacturer
Filter
Manufacturer
Type of Water
(Tap vs Bottle)
Process
Indicators
Amount of
Coffee
Amount of
Water
Age of Coffee
Output
Indicators
Coffee Temp.
Coffee Color
Coffee Flavor
Customer
Satisfaction
Index (Taste)
ASU Department of Industrial Engineering 2004
Determine What to Measure
Input
Indicators
Process
Indicators
Output
Indicators
X
Y
Y = f(X)
ASU Department of Industrial Engineering 2004
Tools
Functional Process Map
Brainstorming (Cause & Effect Diagram)
Failure Modes and Effects Analysis (FMEA)
Cause and Effect Matrix
Selecting and Prioritizing Input, Process and
Output Indicators
ASU Department of Industrial Engineering 2004
Empty
Coffee Pot
Put Coffee
Filter
Put Coffee
in Filter
Fill Water
Jug
Turn Coffee
Maker On
Pour Water
in Coffee Maker
Coffee
Ready
Receive
Customer
Order
Fill Coffee
in Cup
Serve
Customer
Get
Payment
Coffee Maker Sales Associate
Functional Process Map - Coffee Example
ASU Department of Industrial Engineering 2004
MEOPLE MACHINE
VARIATION IN
COFFEE TASTE
MATERIAL METHOD
Amount of Coffee
Age of Coffee
Caffeine Content
Amount of Water
Water Type
Coffee Mfg
Training
Cause & Effect Diagram - Coffee Example
Age of Brewed Coffee
Heater
ASU Department of Industrial Engineering 2004
Failure Modes and Effects Analysis
Process
=3
Process
=0.333
Measurement
=0.4
Measurement
=0.0167
Total
=3.4
Total
=0.3334
+
=
ASU Department of Industrial Engineering 2004
Measurement Systems Properties for Continuous
Data
Discrimination
Accuracy (Bias)
Stability
Linearity
Gauge Capability (GR&R)
MSA for Continuous Data
ASU Department of Industrial Engineering 2004
Property: Discrimination
Capability of the measurement system to detect and
faithfully indicate even small changes of the measured
characteristic
1 2 3 4 5
Good Discrimination
1 2 3 4 5
Poor Discrimination
ASU Department of Industrial Engineering 2004
Discrimination contd.
A general Rule of Thumb:
A measurement tool will have adequate discrimination
if the measurement unit is at most one-tenth of the six
sigma spread of the total process variation,
Measurement Unit < (6*
Total
)/10
ASU Department of Industrial Engineering 2004
Property: Accuracy or Bias
Bias is the difference between the observed average
and the reference value
Accurate
Not
Accurate
ASU Department of Industrial Engineering 2004
Obs Avg = 101.63 Ref Value = 100 Bias
Accuracy or Bias contd.
ASU Department of Industrial Engineering 2004
The distribution of the measurements should be
constant over time
Average
Standard deviation
No drifts, sudden shifts, cycles, etc.
Evaluated with control charts of standard/golden
unit(s) measurements
Xbar/R, Xbar/S, X/MR, etc.
Property: Stability
ASU Department of Industrial Engineering 2004
Stable
Gage
Time 1 Time 2
Not
Stable
Gage
Stability contd
ASU Department of Industrial Engineering 2004
Stability contd
3 Reference Units on 1 Metrology Tool
ASU Department of Industrial Engineering 2004
50 40 30 20 10 0
4250
4240
4230
Reading No.
P
o
l
y
T
h
i
c
k
n
e
s
s
(
A
n
g
s
t
r
o
m
s
)
6/9/xx
6/22/xx
7/11/xx
Stability -- Example
Trend chart for polysilicon thickness measurements in a Chemical
Vapor Deposition system.
On 6/22, something apparently happened to the process.
The change on 6/22 was traced to a faulty measurement tool.
ASU Department of Industrial Engineering 2004
Property: Linearity
Linearity is the difference in the bias values through the
expected operating range of the gauge
Good Linearity
Not Good
Linearity
Low
High
Range of
Operation
ASU Department of Industrial Engineering 2004
Bias and Linearity Example
(File: gauge study.mtw)
ASU Department of Industrial Engineering 2004
Property: Gauge Capability (GR&R)
Gauge Capability is made up of two sources of
variation or components
Repeatability & Reproducibility
2 2 2
Repeatability Reproducibility Measurement
+ =
2 2 2 2
Repeatability Reproducibility Process Total
+ + =
ASU Department of Industrial Engineering 2004
Repeatability
The inherent variability of the measurement system.
The variation that results when repeated measurements are made
of the same parameter under as absolutely identical conditions
as possible:
same operator.
same set up procedure.
same test unit.
same environmental conditions.
during a short interval of time.
ASU Department of Industrial Engineering 2004
Repeatability
True
Value
Mean
Poor
Repeatability
Good
Repeatability
Mean
6
6
ASU Department of Industrial Engineering 2004
2
Measurement
=
2
Repeatability
+
2
Reproducibility
Reproducibility
The variation that results when different conditions are used to
make the measurement:
different operators.
different set up procedures, maintenance procedures, etc.
different parts.
different environmental conditions.
During a longer period of time.
ASU Department of Industrial Engineering 2004
Reproducibility
True
Value
Good
Reproducibility
Poor
Reproducibility
Operator 1 Operator 2
Operator 3
Operator 2
Operator 3
Operator 1
ASU Department of Industrial Engineering 2004
Gauge Capability Metrics
Measurement
Total
% R&R 100
=
Measurement
6
% P/T *100
USL - LSL
=
ASU Department of Industrial Engineering 2004
Requirements for Gauge Capability Metrics
Guidelines for %R&R and %P/T:
Under 10% Acceptable
10% - 30% May be Acceptable
Over 30% Not Acceptable
To find %R&R and %P/T we must estimate
Measurement
and
Total
ASU Department of Industrial Engineering 2004
Example: ANOVA Method (File: gauge study.mtw)
3 Operators, same 10 Parts, 2 Readings/Part
Operators & Parts are crossed
USL = 2 and LSL = 1
Gage R&R
%Cont r i but i on
Sour ce Var Comp ( of Var Comp)
Tot al Gage R&R 0.0012892 11. 44
Repeat abi l i t y 0. 0004033 3. 58 Er r or
Repr oduci bi l i t y 0. 0008858 7. 86
Oper at or 0. 0002584 2. 29
Oper at or *Par t 0. 0006274 5. 57
Par t - To- Par t 0. 0099772 88. 56 Pr oc ess
Tot al Var i at i on 0.0112664 100. 00
ASU Department of Industrial Engineering 2004
ANOVA Method contd - Minitab Output
=
ASU Department of Industrial Engineering 2004
T L S L U S L
F o u r P r o c e s s e s w i t h C p k = 1 . 5
C p = 6 . 0
C p = 3 . 0
C p = 2 . 0
C p = 1 . 5
A :
B :
C :
D :
Cpk alone is not sufficient to indicate the capability of a process
ASU Department of Industrial Engineering 2004
Cpm Alternative to Cpk
Cpm is considerably
more sensitive to
deviations from target
than Cpk
ASU Department of Industrial Engineering 2004
A hotel provides room service meals to its guests. It is
hotel policy that the meal is delivered at the time
scheduled by the guest.
The hotel Six Sigma team has found from the Voice of
the Customer that a breakfast delivered too early will
inconvenience the guest as much as a late delivery.
Research indicates that guests require that breakfast
be delivered within 10 minutes of the scheduled
delivery time.
Example: Hotel Breakfast Delivery
(File: HotelMeals.mtw)
ASU Department of Industrial Engineering 2004
Example: Hotel Breakfast Delivery
24 18 12 6 0 -6 -12
LSL Target USL
Process Data
Sample N 725
StDev(Within) 7.20201
StDev(Overall) 7.16405
LSL -10.00000
Target 0.00000
USL 10.00000
Sample Mean 6.00357
Potential (Within) Capability
CCpk 0.46
Overall Capability
Pp 0.47
PPL 0.74
PPU 0.19
Ppk
Cp
0.19
Cpm 0.36
0.46
CPL 0.74
CPU 0.18
Cpk 0.18
Observed Performance
PPM < LSL 13793.10
PPM > USL 268965.52
PPM Total 282758.62
Exp. Within Performance
PPM < LSL 13138.34
PPM > USL 289479.68
PPM Total 302618.02
Exp. Overall Performance
PPM < LSL 12745.81
PPM > USL 288475.05
PPM Total 301220.86
Within
Overall
Process Capability of Delivery Time Deviation
ASU Department of Industrial Engineering 2004
Defects per Million Opportunities
D = Total # of defects counted in the sample
Must be at least 5 defects and 5 non-defects to calculate
DPMO
N = # of units of product/service
O = # of opportunities for a defect to occur per unit of
product/service
M = million
DPMO = 1M Defects .
Units Opportunities
ASU Department of Industrial Engineering 2004
Sigma DPMO
2 308770
2.25 226716
2.5 158687
2.75 105660
3 66811
3.25 40060
3.5 22750
3.75 12225
4 6210
4.25 2980
4.5 1350
4.75 577
5 233
5.25 88
5.5 32
5.75 11
6 3.4
Defects per Million Opportunities vs Process Sigma
ASU Department of Industrial Engineering 2004
D = 205
N = 725
O = 1
Example: Hotel Breakfast Delivery
DPMO = 1M 205 .
725 1
= 282,758
ASU Department of Industrial Engineering 2004
Process Lead Time (PLT)
Littles Law
Customer
Orders
Order Entry Credit Check Schedule Orders
Order Take
Exit Rate =
20 units/day
WIP = 100
PLT= 100/20 = 5 days
Exit Rate (ER)
Work IN Process (WIP)
=
Process
Lead Time
(PLT)
ASU Department of Industrial Engineering 2004
Definitions
Process Lead Time (PLT) The time taken from the
entry of work into a process until the work exits the
process (which may consist of many activities).
Work-In-Process (WIP) The amount of work that has
entered the process but has not been completed. It
can be paper, parts, product, information, emails, etc.
Exit Rate (Average Completion Rate or Throughput)
The average output of a process over a given period
of time (usually a day) (units/time).
ASU Department of Industrial Engineering 2004
Value is Defined by the Customer
Customer Value-Added (CVA)
An activity adds value for the customer only if:
The customer recognizes the value
It changes the service/product toward
something the customer expects
It is done right the first time
ASU Department of Industrial Engineering 2004
Process Cycle Efficiency (PCE)
Process Lead Time
Customer Value Added Time
=
Process
Cycle
Efficiency
Customer
Orders
Order Entry Credit Check Schedule Orders
Order Take
Exit Rate =
20 units/day
WIP = 100
CVA=0.4 hrs CVA=0.4 hrs
CVA=0.3 hrs
CVA=0.4 hrs
PCE = 1.5 hrs/5 days = 1.5 hrs/40 hrs = 3.75%
Assuming 1 day = 8 hrs
ASU Department of Industrial Engineering 2004
Measure Performance
Inputs
Team Charter
Business case
Goal statement
Project scope
Project plan
Team roles and responsibilities
Prepared Team
Critical Customer Requirements
Process Maps
Quick Win Opportunities
2.0 Measure Performance
Determine What to
Measure
Manage Measurement
Evaluate Measurement
System
Determine Process
Performance
Key Deliverables
Input, Process, and
Output Indicators
Operational
Definitions
Data Collection
Formats and Sampling
Plans
Measurement System
Capability
Baseline Performance
Metrics
Process Capability
DPMO
PLT
PCE
Yield/Scrap
Others
Productive Team
Atmosphere
ASU Department of Industrial Engineering 2004
DMAIC - Process Improvement Roadmap
What is
important?
How are
we doing?
What is
wrong?
What
needs to
be done?
How do we
guarantee
performance?
1.0
Define
Opportunities
2.0
Measure
Performance
3.0
Analyze
Opportunity
4.0
Improve
Performance
5.0
Control
Performance
IEE 581 Six-Sigma Methodology
DMAIC The Analyze Phase
Fall 2012 Class 6
Cheryl L. Jennings, PhD, MBB
c.jennings@asu.edu
1
More on Process Capability Analysis
Previous Measure lecture, measures of process performance included:
Cp, Cpk, Cpm for continuous data
DPMO vs PPM for discrete data
Typically used as a goodness measure of a process performance
In the Measure phase to baseline performance
During the Analyze phase to identify suspect equipment, suppliers, etc., and
provide direction to the project
In the Improve phase to compare before performance to after
In the Control phase to monitor ongoing performance
Underlying assumptions are normality, and that the process is in statistical control
2
Relationship Between Cp and Cpk
3
* From Montgomery, D. C. (2009), Introduction to Statistical Quality Control 6
th
edition, Wiley, New York
Motorola Definition of
Six Sigma Quality
Cp and Cpk The Usual Equations
4
Arent these just Point Estimates?
{ }
USL-LSL
6
USL LSL
min , where
3 3
P
PK PU PL PU PL
C
s
X X
C C C C and C
s s
=
= = =
Confidence Interval for Cpk
Key Points
The Cpk metric is routinely used.
Recall that the Cpk that we calculate are based on statistics. Therefore our
calculated Cpks are used to estimate the TRUE Cpk.
Rarely (if ever) is the confidence interval on a Cpk considered.
Black Belts should consider CIs.
5
2 2
1 1 1 1
1 1.96 1 1.96
9 2 2 9 2 2
PK PK PK
PK PK
C C C
nC n nC n
| | | |
+ s s + +
| |
\ . \ .
For a 95% confidence interval:
/2 /2
2 2
1 1 1 1
1 1
9 2 2 9 2 2
PK PK PK
PK PK
C Z C C Z
nC n nC n
o o
| | | |
+ s s + +
| |
\ . \ .
Example
Based on a sample size of n = 13 and an estimated Cpk = 1.11, a 95% confidence
interval for Cpk is:
2 2
2
2
1 1 1 1
1 1.96 1 1.96
9 2 2 9 2 2
1 1
1.11 1 1.96
9(13)(1.11 ) 2(13) 2
1 1
1.11 1 1.96
9(13)(1.11 ) 2(13) 2
0.63 1.59
PK PK PK
PK PK
PK
PK
C C C
nC n nC n
C
C
| | | |
+ s s + +
| |
\ . \ .
| |
+ s
|
\ .
| |
s + +
|
\ .
s s
6
What if Data is not Normally Distributed?
7
Example
n = 200
Values range from 1001.68 to 2891.49
Histogram shows clearly that data are
skewed right and not normal
With LSL = 900 and USL = 2700
Assuming normal data, the usual Cpk
estimate would be 0.46
However non-normal Cpk = 1.15
8
Yet Another Process Performance Measure
9
Are these indices really useful?
* From Montgomery, D. C. (2009), Introduction to Statistical Quality Control 6
th
edition, Wiley, New York
Key Points
The Cpk metric is routinely used
Rarely (if ever) is the confidence interval on a Cpk considered
Black Belts should consider using Confidence Intervals
10
The DMAIC Process
* From Montgomery, D. C. (2009), Introduction to Statistical Quality Control 6
th
edition, Wiley, New York
Analyze Opportunity
12
3.0 Analyze Opportunity
Identify and Validate
Root Causes
Basic Tools
Advanced Tools
Inputs
Input, Process, and Output
Indicators
Operational Definitions
Data Collection Formats and
Sampling Plans
Measurement System Capability
Baseline Performance Metrics
Process Capability
Cost of Poor Quality (COPQ)
Time
Yield
Other
Productive Team Atmosphere
Outputs
Data Analyses
Validated Root Causes
Potential Solutions
* From Montgomery, D. C. (2009), Introduction to Statistical Quality Control 6
th
edition, Wiley, New York
The Primary DMAIC Six Sigma Tools
Three Ways to Obtain Data for Analysis
1. A retrospective study using historical data
A lot of data
But generated under what conditions?
Data quality issues
2. An observational study
Planned data collection, under known conditions in a production mode
Typically a short period of time, may not see all variation or be able to see
changes in key variables
3. A designed experiment
Also planned data collection, with deliberate manipulation of controllable
process inputs
The only way to prove cause-and-effect relationship
Requires commitment of resources
14
Identify Potential Root Causes Basic Analyze Tools
Cause & Effect Diagram*
FMEA*
Cause & Effect Matrix*
Histogram
Scatter Plot
Box Plots
Pareto Diagram
*Discussed in Measure lectures
15
Pareto Diagram
16
80%
Top three complaint categories comprise
80% of problem.
Other teams are working on 1 & 2. Your
team is tasked with cabin-related
complaints.
Cabin accommodations generated most
complaints related to aircraft cabins; most
complaints were about room for carry-on
baggage.
In the last year, 65% of airline passenger complaints about aircraft cabin interior baggage accommodations
concerned insufficient stowage in overhead bins for carry-on luggage.
Complaints
N
u
m
b
e
r
o
f
D
e
f
e
c
t
s
Cost Sched Cabin Bags Rgs Tix Etc.
Cabin-related
Complaints
Accom. Food Bevs Ent Sound Other
50%
N
u
m
b
e
r
o
f
D
e
f
e
c
t
s
Cabin Physical
Accommodations
Bag
Room
Leg
Room
Seat
Width
Head
Room
Rest
Room
Other
80%
50%
Bag Accommodations
(Storage)
Ovhd
Bin
Under
Seat
Garment
Rack
Other
65%
Identify Potential Root Causes Advanced Analyze Tools
Statistical Process Control (SPC)
Comparative Methods: Hypothesis tests, Confidence intervals
ANOVA
Source of Variation (SOV) Studies
Regression Analysis
Screening Experiments (Designed Experiment, DOE)
Nonparametric Methods
17
Phase I and Phase II Control Chart Application
Phase I Process Taming
Process is likely out of control; as in Measure, Analyze and Improve phases
Use of control charts is to bring process into state of control, with the
identification of out-of-control signals and investigation for root cause
Shewhart control charts are suited to Phase I because
Easy to construct & interpret
Effective at detecting both large, sustained process shifts as well as outliers,
measurement errors, data entry errors, etc.
Patterns are often easy to interpret and have physical meaning
Also suited to use of sensitizing or Western Electric rules
Phase II Process Monitoring
Process is relatively stable, causes of larger shifts have been identified and
permanently fixed; as in Control phase
18
SPC to Identify Potential Causes
In Phase I, control limits are typically calculated retrospectively
Data is collected, say 20 or 25 subgroups
Trial control limits are calculated
Out-of-control points are investigated for assignable causes and solutions
Control limits are recalculated from points within the trial control limits
New data is collected, compared with the revised trial control limits, and the
analysis is repeated until the process is stabilized
In Phase II, control limits are calculated from the stabilized process
19
Shewart 3-sigma limits
Why do we often use 3 sigma limits?
... Experience indicates that t = 3 seems to be an acceptable economic value. ...
Economic Control of Quality of Manufactured Product, W.A. Shewhart,
Commemorative Issue published by ASQ in 1980, p. 277.
Wider control limits decrease the risk of a type I error, the risk of a point falling
beyond the control limits indicating an out-of-control condition when no assignable
cause exists
For 3-sigma limits, the probability is 0.0027 (27 out of 10,000 plot points), or
0.0135 in one direction
Wider control limits also increase the risk of a type II error, the risk of a point falling
between the control limits when the process is really out of control
20
Comparative Methods
Comparison Type Analysis Tests
Single sample one-to-standard (fixed
value)
Z-test
t-test
_
2
-test
Sign/Wilcoxon
Two samples
Paired two-sample
one-to-one Z-test
t-test
F-test
Paired t-test
Sign test
Wilcoxon Rank Sum (also called
the Mann-Whitney test)
Multiple samples multiple ANOVA
Kruskal Wallis
Use of ranks
_
2
-tests
21
Parametric Inference Methods
We will look at three tests, but fundamentals apply to all tests
The one-sample Z-test
The one-sample t-test
The two-sample t-test (also the pooled t-test)
Assumptions for these three tests are
Random samples
From normal populations
And for two-sample tests, the two populations are independent
Checking for random, independent samples
Best approach is to use a sound sampling plan
Statistical approaches for time-oriented data include runs tests and time series
methods
22
Checking Normality
Probability Plot
Boxplot
Goodness-of-fit tests: chi-square, Anderson-Darling
H0: The form of the population distribution for characteristic is Normal.
23
7-Step Hypothesis Testing Procedure
24
1. Parameter of Interest
2. Null Hypothesis
3. Alternative Hypothesis
4. Test Statistic
5. Reject H0 if:
Test statistic approach (fixed significance)
P-value approach
Confidence Interval approach
6. Computations
Includes checking assumptions
7. Conclusions
The One-Sample Z-Test
25
We Could Also Use a P-Value Approach
26
27
An Example of the Z-Test
28
3
rd
Approach:
Confidence
Intervals
The One-Sample t-Test
29
An Example of the t-Test
30
MINITAB 1-Sample t-Test
31
When doing the t-test
manually, it is usually necessary
to approximate the P-value
Approximating the P-value with a t-Table
32
Approximating the P-value with MINITAB
33
The Two-Sample t-Test
34
Testing Hypotheses on the Difference in Means of Two Normal
Distributions, Variances Unknown
An Example
35
MINITAB 2-Sample t-Test
36
Other Comparative Tests for Normal Distributions
The Paired t-Test
2 samples, paired data
If analyzed incorrectly as a 2-sample test, the variance estimate may be inflated
and give misleading results
_
2
-test
Variance of a normal distribution
F test
Variances of two normal distributions
37
What if the Distribution is Not Normal?
Comparative methods discussed are based on assumption of random sample from a
normal distribution
Most of the comparative methods based on the normal distribution are relatively
insensitive to moderate departures from normality
Two exceptions are the _
2
and F tests for variance
Options for more severe departures from normality are
1. Transform the data to normal, for example using logarithm, square root or a
reciprocal, and use a method based on the normal distribution
See Montgomery, DOE, Selecting a Transformation: The Box-Cox Method
2. Utilize a nonparametric or distribution-free approach
38
More Than Two Populations?
For more than two populations or two factor levels aka a single-factor experiment
ANOVA can be used for comparing means
39
40
Assumptions can be checked by analyzing residuals
Normality
Independence
Equal variance
Sources of Variation Studies
Sources of Variation (or SOV) studies are used to understand and characterize
process variability
Often described as a process snapshot, the process is observed in a production
mode without adjustment or manipulation
A sampling plan is designed to encompass what are thought to be the major
contributors to process variability
Data is collected over a sufficient period of time to capture a high percentage of the
historical process variation
Often suited to analysis as a nested design
May be a precursor to a designed experiment (DOE)
41
Solder Paste Example
A process engineer is interested in determining where the majority of the variability is
coming from in the raw material being supplied to a screen-printing process. Three
lots of solder paste are randomly selected. From each lot, four tubes of solder paste
are selected at random. Three boards are printed for each tube of solder paste.
42
For more on Nested Designs, see Chapter 14 in Montgomery, D. C. (2009),
Design and Analysis of Experiments, 7
th
edition, Wiley, New York.
2 3 1 Lot:
1 2 3 4 Tube:
Board:
1
2
3
1 2 3 4 1 2 3 4
Tree Diagram
Volume
Measurement:
28
23
23
1
2
3
1
2
3
1
2
3
1
2
3
1
2
3
1
2
3
1
2
3
27
25
24
MINITAB Analysis
Examining p-values, conclude there is
no significant effect on Volume due to
Lot, but the Tubes of solder paste from
the same Lot differ significantly.
Knowing that the major source of
variability is the Tube-to-Tube
variation within a Lot points gives
direction for solving the problem.
Unfortunately, also note that the
Within-Tube (Error, or Board-to-Board)
variability is the largest source of
variation, suggesting improvement in
the screen-printing process.
43
Regression Analysis
Recall that two ways to obtain data for analysis included
A Retrospective study using historical data
An Observational study resulting from planned data collection
Regression can be used for both, with care on Retrospective data
Abuses of Regression include
Selection of variables that are completely unrelated in a causal sense a strong
observed relationship does not imply that a causal relationship exists. Designed
experiments are the only way to determine cause-and-effect relationships.
Extrapolation beyond the range of the original data
We will study logistic regression in a later class lecture on Categorical data analysis
44
Design of Experiments
Types of
Experiments
Screening Optimization Comparison Robust Design
Full Factorial Medium Medium High Medium
Fraction Factorial High Low Medium Low
Response Surface
Methodology (RSM)
Low High Medium High
Plackett-Burman High Low Low Low
45
The table below lists four types of experiments and the degree of suitability (High,
Med, or Low) for each experimental objective
Screening and Comparison experiments are suited for use in the DMAIC Analyze
phase
46
Step 4.
Perform Residual Diagnostics
Step 1.
View the Data
Step 5.
Transformation
Required?
Make
Confirmation Runs
Yes
Yes
No
Step 9.
Stop
Experimentation?
Run RSM
Yes
No
No
Step 8.
Interpret Chosen Model
Step 7.
Choose Model
Step 6.
Reduce Model?
Step 3.
Fit the Model
Step 2.
Create the Model
Analysis and Interpretation
of Factorial Experiments
Tips for Designed Experiments
Plan Experiment (Use Engineering and
Statistical Knowledge)
Objective
Selection of Responses & Input
Variables (Operating Range, Levels,
Interactions etc.)
Blocking
Replication
Dont forget Center Points!
Conduct Experiment
Randomization
Data collection and Comments
Statistical Analysis
Analyze Experiment
Sparsity of Effects
Statistical Model
Residual Diagnostics
Interpret Results
Results match with engineering
intuition
Confidence Interval on Predictions
Confirmation Tests
47
One Tip on How NOT to Design an Experiment
A Designed Experiment is NOT a Retrospective or Observational study
The variables and variable levels are deliberately manipulated in a random
manner
A DOE cannot be retro-fitted to data collected retrospectively or through passive
observation
48
References
Montgomery, D. C. (2009), Design and Analysis of Experiments, 7th edition, Wiley,
New York.
Montgomery, D. C. (2009), Introduction to Statistical Quality Control, 6th edition,
Wiley, New York.
Montgomery, D. C. and Runger, G. C. (2011), Applied Statistics and Probability for
Engineers, 5th edition, Wiley, New York.
Upcoming
Analyze dataset is posted on Blackboard (both MINITAB and Excel)
Read two case studies posted on Blackboard
Goodman et al, Six Sigma Forum Magazine, November 2007, When Project
Termination is the Beginning
Tong et al, Intl Journal AMT, January 2004, A DMAIC approach to printed circuit
board quality improvement
How to contact me
E-mail: c.jennings@asu.edu
Cell: 602-463-5134
50
51
IEE 581 Six-Sigma Methodology
DMAIC The Analyze Phase
Fall 2012 Class 7
Cheryl L. Jennings, PhD, MBB
c.jennings@asu.edu
1
Fisher 1 in 20
Why do we often use = 0.05 as significance level?
http://psychclassics.asu.edu/Fisher/Methods/chap3.htm, Statistical Methods for
Research Workers By Ronald A. Fisher (1925), Chapter III, Distributions
we can find what fraction of the total population has a larger deviation; or, in other
words, what is the probability that a value so distributed, chosen at random, shall exceed a
given deviation. Tables I. and II. have been constructed to show the deviations
corresponding to different values of this probability. The rapidity with which the probability
falls off as the deviation increases is well shown in these tables. A deviation exceeding the
standard deviation occurs about once in three trials. Twice the standard deviation is
exceeded only about once in 22 trials, thrice the standard deviation only once in 370 trials,
while Table II. shows that to exceed the standard deviation sixfold would need [p. 47] nearly
a thousand million trials. The value for which P =.05, or 1 in 20, is 1.96 or nearly 2 ; it is
convenient to take this point as a limit in judging whether a deviation is to be considered
significant or not. Deviations exceeding twice the standard deviation are thus formally
regarded as significant. Using this criterion, we should be led to follow up a negative
result only once in 22 trials, even if the statistics are the only guide available. Small effects
would still escape notice if the data were insufficiently numerous to bring them out, but
no lowering of the standard of significance would meet this difficulty.
2
How robust is the t-test to the normality assumption?
One assumption for using the t-test for
means is that the data is normally
distributed
While the test is somewhat robust to
this assumption, consider the test
statistic calculation
Two key things about this statistic
When sampling from the normal
distribution, are
independent
The denominator is distributed as
Lets look at an example
Consider a cycle time problem, say
the time it takes to process a loan
from receipt of application to wiring
of funds. Cycle times are often
exponentially distributed.
Select a random sample of ten loans
and test the hypothesis that the
mean cycle time is 10 days
To study the impact of cycle time
distribution on the t-test statistic,
randomly generate 50 samples of 10
loans each, from an exponential
distribution with a mean of 10 days
3
( )
0
X
t
S n
=
and X S
2
~ df S n _
4
Recall that for an exponential
distribution, = o, so clearly the
independence assumption is violated
Histograms of the 50 samples show
the skewness of cycle time
A histogram of the 50 t statistics is
clearly skewed in comparison to a t
distribution with 9 degrees of
freedom
Using p-values based on the t
distribution could lead to erroneous
conclusions
What if the distribution is not normal?
Comparative methods discussed were based on assumption of random sample from
a normal distribution
Most of these procedures are relatively insensitive to moderate departures from
normality
Options for more severe departures from normality are
1. Transform the data to normal, for example using logarithm, square root or a
reciprocal, and use a method based on the normal distribution
See Montgomery, DOE, Selecting a Transformation: The Box-Cox Method
2. Utilize a nonparametric or distribution-free approach
5
Non-Parametric Inference Methods
We will look at two types of tests, tests based on Signs and tests based on Ranks
Distribution-free, or no underlying parametric distribution assumption
However each test does have other assumptions
Why not always use nonparametric methods?
In general, nonparametric procedures do not use all the information in a sample,
and as a result are less efficient, requiring larger samples sizes to achieve same
power as the appropriate parametric procedure
6
Comparison Type Analysis Tests
Single sample
Paired two-sample
one-to-standard (fixed value) Sign
Wilcoxon Signed-Rank
Two samples one-to-one Wilcoxon Rank Sum (also called
the Mann-Whitney test)
Multiple samples multiple Kruskal Wallis
Use of ranks
The Sign Test for One Sample
7
* From Montgomery, D. C. and Runger, G. C. (2011), Applied Statistics and Probability
for Engineers 5
th
edition, Section 9-9 Nonparametric Procedures, Wiley, New York
8
9
10
Sign Test Example
11
Calculating P-value in Minitab
12
P-value = 2 x Pr(R
+
14) = 2 x Pr(R
+
13) = 2 x (1 0.942341) = 0.1153
13
Or, Minitab 1-Sample Sign Test
14
15
16
17
18
19
20
The Wilcoxon Signed-Rank Test for One Sample
21
* From Montgomery, D. C. and Runger, G. C. (2011), Applied Statistics and Probability
for Engineers 5
th
edition, Section 9-9 Nonparametric Procedures, Wiley, New York
22
23
Wilcoxon Signed-Rank Example
24
are shown to the left.
Minitab 1-Sample Wilcoxon
25
Uses maximum sum of
ranks instead of
minimum
26
Comparison to the t-Test
27
Median Tests for Paired Samples
Both the sign test and the Wilcoxon signed-rank test can be applied to paired
observations.
In the case of the sign test, the null hypothesis is that the median of the differences
is equal to zero.
The Wilcoxon signed-rank test is for the null hypothesis that the mean of the
differences is equal to zero.
The procedures are applied to the observed differences as described previously.
28
29
* From Montgomery, D. C. and Runger, G. C. (2009), Applied Statistics
and Probability for Engineers 4
th
edition, Wiley, New York
30
1. Parameter of Interest: The parameters of
interest are the median fuel mileage performance
for the two metering devices.
2. Null Hypothesis: H
0
: Median
1
= Median
2
, or
equivalently, H
0
: Median
D
= 0
3. Alternative Hypothesis: H
1
: Median
1
Median
2
, or equivalently, H
1
: Median
D
0
4. Test Statistic: We will use Appendix Table VIII
for the test, so the test statistic is r = min(r+, r).
5. Reject H
0
if: For = 0.05, n = 12, two-sided
test, Table VIII gives the critical value as r*
0.05
=
2. We will reject H
0
in favor of H
1
if r 2.
6. Computations: Table 15-2 shows differences
and their signs, r+ = 8 and r = 4. So r = min (8,
4) = 4.
7. Conclusion: Since r = 4 is not less than or
equal to the critical value r*
0.05
= 1, we cannot
reject the null hypothesis that the two devices
provide the same median fuel mileage
performance.
EXAMPLE 15-3
An automotive engineer is investigating two different
types of metering devices for an electronic fuel
injection system to determine whether they differ in
their fuel mileage performance. The system is
installed on 12 different cars and a test is run with
each metering device on each car. The observed fuel
mileage performance data, corresponding
differences, and their signs are shown in Table 15-2.
We will use the sign test to determined whether the
median fuel mileage performance is the same for
both devices using = 0.05.
* From Montgomery, D. C. and Runger, G. C. (2009), Applied Statistics
and Probability for Engineers 4
th
edition, Wiley, New York
Minitab 1-Sample Sign with Paired Data
31
Minitab 1-Sample Wilcoxon with Paired Data
32
The Wilcoxon Rank-Sum Test for Two Samples
33
* From Montgomery, D. C. and Runger, G. C. (2011), Applied Statistics and Probability
for Engineers 5
th
edition, Section 9-9 Nonparametric Procedures, Wiley, New York
34
35
36
Wilcoxon Rank-Sum Example
37
38
Minitab Mann-Whitney (Wilcoxon Rank Sum)
39
40
41
More Than Two Means
Weve looked at methods that can be used for experiments with two levels: the t-
test and Wilcoxon-Rank Sum test
For more than two levels, ANOVA can be used for comparing means
The single-factor analysis of variance model for comparing a population means is
In this model the error terms c
ij
are assumed to be normally and independently
distributed with mean zero and variance o
2
. The assumption of normality leads
directly to the F-test.
The Kruskal-Wallis test is a nonparametric alternative to the F-test; it requires only
that the c
ij
have the same continuous distribution for all factor levels i = 1, 2, ..., a.
42
* From Montgomery, D. C. and Runger, G. C. (2009), Applied Statistics and
Probability for Engineers 4
th
edition, Wiley, New York
1,2,...,
1,2,...,
ij i ij
i
i a
Y
j n
t c
=
= + +
=
. Model for
improving and appraising the performance of
development in software organizations and related fields.
JCHAO Accredits health care organizations and programs
NCA Accredits learning institutions in 19 states
MBNQA
Performance
Excellence Criteria
Basis for the Malcolm Baldrige National Quality Award, in
three areas: Business/Nonprofit, Health Care, Education.
Not a certification.
ISO 9000 Family
Addresses what an organization does to meet a Customers quality requirements, meet
applicable regulatory requirements, enhance Customer satisfaction, and achieve continual
performance improvement
Comprises
ISO 9000, Quality Management (Definitions, fundamentals and vocabulary)
ISO 9001, Quality management systems Requirements
ISO 9004, Quality management systems Guidelines for performance improvements
ISO 9001:2008 did not introduce additional requirements ISO 9001:2000, which
Combined twenty elements of original standard into four major processes
Management Responsibility
Resource Management
Product and/or Service Realization Management
Measurement, Analysis and Improvement
Added requirements to
Establish a system-level procedure to facilitate continual improvement
Measure customer satisfaction and dissatisfaction
Use of appropriate statistical techniques
25
Sample of ISO 9001:2000* Language
26
* Note: ISO 9001:2008 does not introduce additional requirements or
change intent, but clarifies existing requirements
The Baldrige Criteria for Performance Excellence
The Baldrige Performance Excellence Program and the Performance Excellence
criteria play a role in US Competitiveness
Managed by NIST
Help improve organizational performance practices, capabilities, and results
Facilitate communication and sharing of best practices
Serve as a working tool for understanding and managing performance
Are the basis for conducting organizational self-assessments, giving feedback to
applicants, and making the MBNQA
Provide the framework for an organization to
Measure organizational performance
Plan in an uncertain environment
Decide on approaches to performance improvement
Lean, Six Sigma, ISO, Balanced Scorecard, etc.
Improve communication, productivity and effectiveness
Achieve strategic goals
27
28
1. Leadership
1. Senior Leadership
2. Governance and Societal Responsibilities
2. Strategic Planning
1. Strategy Development
2. Strategy Implementation
3. Customer Focus
1. Voice of the Customer
2. Customer Engagement
4. Measurement, Analysis, and Knowledge Management
1. Measurement, Analysis, and Improvement of Organizational Performance
2. Management of Information, Knowledge, and Information Technology
5. Workforce Focus
1. Workforce Environment
2. Workforce Engagement
6. Process Management
1. Work Systems
2. Work Processes
7. Results
1. Product and Process Outcomes
2. Customer-Focused Outcomes
3. Workforce-Focused Outcomes
4. Leadership and Governance Outcomes
5. Financial and Market Outcomes
D
20112012 Business Criteria for Performance Excellence*
C
* 2011-2012 Business Criteria for Performance Excellence, BNQP, NIST
M,A,I
Sample of PE Criteria Language
29
Lean? Six Sigma?
P-D-C-A?
30
What Makes the Difference?*
Although all three are quality management systems, Six Sigma, ISO 9001 Registration and the
Baldrige Criteria for Performance Excellence each offer a different emphasis in helping
organizations improve performance and increase customer satisfaction.
Six Sigma
concentrates on measuring product (and service) quality and improving process engineering
drives process improvement and cost savings
ISO 9001 Registration
is a product/service conformity model for guaranteeing equity in the marketplace
concentrates on fixing quality system defects and product/service nonconformities
Baldrige Criteria for Performance Excellence
focus on performance excellence for the entire organization in an overall management
framework
identify and track all-important organizational results: customer, product/service, financial,
human resource, and organizational effectiveness
* Baldrige, Six Sigma, & ISO: Understanding Your Options, CEO
Issue Sheet, Baldrige National Quality Program, Summer 2002.
Control Tollgate Questions
Was the solution tested on a small scale? How
representative was the test? How are your
learnings from the pilot integrated into the
implementation plan?
What is the implementation plan? Who is
responsible for implementation? What are the
potential problems? What are the contingency
plans?
How has the process been standardized? How have
you documented the process changes?
How has training been conducted to assure
understanding of the process changes? How
effective was this training? What continuing issues
does your team need to address in the area of
training?
What is the communication plan for
implementation? How will your team use
communications to manage this change, minimize
resistance and mobilize stakeholders?
What controls are in place to assure that the
problem does not recur? What is being measured?
What evidence do you have that would indicate the
process is in-control? How well and consistently
is the process performing? Is a response plan in
place for when the process experiences out-of-
control occurrences?
Who is the process owner? How will the
responsibility for continued review be transferred
from the improvement team to the process owner?
Have key process metrics been integrated into the
management review process? How frequent are
the reviews?
What gains or benefits have we realized from
implementation? How can we replicate the
improvements elsewhere in the organization?
What did the team learn from the project? Where
are some other areas of the business that can
benefit from your learnings? When will the
learnings be shared?
31
The DMAIC Process
From Montgomery, D. C. (2009), Introduction to Statistical Quality Control 6
th
edition, Wiley, New York
32
For Six Sigma to be Successful
As described by Mikel Harry (class lecture video), three keys for successful Six Sigma
implementation are:
1. Six Sigma must be owned at the top, driven by executive management
throughout the organization
2. There must be a company-wide infrastructure to support Six Sigma
3. Six Sigma projects must have demonstrated success in terms of money, not just
cost avoidance
3a. Compensation must be tied to performance results
33
Caterpillar is One Company with a Successful Six Sigma Initiative
2001: Launched 6 Sigma.
2002: 6 Sigma is process improvement and people
engagement.
2003: Increased emphasis on 6 Sigma in New
Product Introduction and Continuous Product
Improvement initiatives.
2004: Leadership team commits to encoding 6
Sigma into CATs DNA and extends deployment to
dealers and suppliers.
2005: Ongoing effort to encode 6 Sigma discipline
into daily work. Used 6 Sigma to develop content
for first Sustainability Report.
2006: Refocused 6 Sigma human resources on
quality, to drive common processes, metrics and
simplification.
2007: Continued focus on 6 Sigma to make
significant improvements in quality, availability,
inventory turns and cost performance.
2008: Caterpillar Production System is powered by
6 Sigma.
2010: Caterpillar Production System and 6 Sigma
methodologies remain core operating principles,
with references in Annual and Sustainability
Reports.
2011: 6 Sigma is embedded in Caterpillars
Worldwide Code of Conduct and Sustainability
Report.
34
And their progress in level of maturity is evident in annual reports
Reading / References
Reading assignment for both Control lectures:
BALDRIGE, SIX SIGMA, & ISO: Understanding Your Options, Baldrige National Quality Program, CEO Issue
Sheet, Summer 2002. (http://www.baldrige.nist.gov/PDF_files/Issue_Sheet_SS.pdf)
George, Michael L., Lean Six Sigma, Chapters 10 & 11
McTurk, Laurie and Wilson, Dennis (2002). Presentation on the Instability Index.
(www.pro.enbis.org/members/documentation/Nissanmotorolapres.ppt)
Montgomery, Douglas C. (2009), Introduction to Statistical Quality Control, 6th edition, Wiley, New York
NIST-SEMATECH, Engineering Statistics Handbook, (http://www.itl.nist.gov/div898/handbook/index.htm)
Simon, Kerri. Poka Yoke Mistake Proofing. (http://www.isixsigma.com/library/content/c020128a.asp)
Snee, Ronald D. and Roger W. Hoerl, Leading Six Sigma, Chapters 5, 6 & 7
The Toyota System web site, Poka-Yoke examples and definition By Greg, May 12, 2009.
(http://www.thetoyotasystem.com/lean_inventions/poka_yoke-you-can%E2%80%99t-go-wrong.php)
35
Up Next
Be sure to watch the video History of Six Sigma by M. Harry
Next three classes with Dr. Cathy Lawson
Upcoming Office hours : Wednesday, 10/3 in BYAC TBD, from 5pm to 7pm. Please
email or call if you need to make other arrangements.
36
This document and all information and expression contained herein are the property of ASU Department of Industrial Engineering, and may not, in whole or
in part, be used, duplicated, or disclosed for any purpose without prior written permission of ASU Department of Industrial Engineering. All rights reserved.
ASU Department of Industrial Engineering 2004
IEE 598 Six-Sigma Methodology
The role of teams in Six Sigma
Dr. Cathy Lawson Medtronic
cathy.a.lawson@medtronic.com
ASU Department of Industrial Engineering 2004
Attributes of Teams
Made up of individuals working towards a shared goal
Cross-functional
Short term
Dynamic
Six Sigma projects are best accomplished using cross-
functional teams
Six Sigma projects are best accomplished using cross-
functional teams
ASU Department of Industrial Engineering 2004
Types of Six Sigma Teams
Process Improvement Teams
Problem Solving Teams
Self-managed Work teams
Steering Committees and Councils
ASU Department of Industrial Engineering 2004
Discussion
Think of a team that you have participated on in the past.
What was your role on the team?
Was it a successful team?
What were the attributes that either made it successful or not?
ASU Department of Industrial Engineering 2004
Attributes of Effective Teams
Informal, comfortable atmosphere
Lots of discussion everyone participates
Tasks and objectives well defined
Members listen to each other
There may be healthy conflict
The decision making process is understood
People feel free to express their ideas
Clear assignments are made for required actions
Leader does not dominate unnecessarily
The team is self-monitoring
Members are committed to the goal
ASU Department of Industrial Engineering 2004
Black Belt Roles on a Team
Leader
Facilitator
Member
Consultant
Expert
Mentor
Black Belts serve in many roles when interacting with teams
ASU Department of Industrial Engineering 2004
Black Belt Team Skills
Planning and goal setting
Meeting management
Listening
Resolving conflict
Presenting
Risk taking
Mentoring
Influencing
Giving and accepting feedback
Dealing with ineffective team members
Monitoring and evaluating
ASU Department of Industrial Engineering 2004
Common mistakes teams make
Not defining the problem/task well
Falling in love with tools
Jumping to a solution without defining the problem
Looking busy while the problem goes away
ASU Department of Industrial Engineering 2004
Common mistakes Management makes with teams
Dont scope team projects well Achieve world
peace
Putting a hero on the team
Charter and walk away
ASU Department of Industrial Engineering 2004
A problem well stated is a
problem half solved.
Charles F. Kettering
ASU Department of Industrial Engineering 2004
Getting the Team Started
Who
What
Where
When
Why
How
Definition (66%)
Cause (17%)
Solution (17%)
ASU Department of Industrial Engineering 2004
What Questions
What do we know as fact?
What do we need to know?
What kind of data do we need to have to tell us what
we need to know?
What are the symptoms we are observing?
What is the problem statement?
ASU Department of Industrial Engineering 2004
Case Study
You are asked by the program manager to work with a
project team to develop and implement a Statistical
Process Control plan for his area. No one on the team
has had any SPC training. At the first meeting, the
program manager is absent and the factory manager
does not hide the fact that he does not want to
participate in this initiative.
What is your role on this team?
How do you make this successful?
ASU Department of Industrial Engineering 2004
Case Studies
You are brought in to consult with a highly visible project that is
having a serious technical problem. It has halted production and
is costing the company thousands of dollars each day the
production line is on hold. When you are brought in, the team
believes they have identified the red X but they want to run an
experiment to verify their conclusions. You learn through
questioning that they have tested two units and found the red
X in one of them.
What is your role on this team?
How do you make this successful?
ASU Department of Industrial Engineering 2004
Case Studies
You are asked to participate on a team working on an
improvement project for a new product design. The team is
following the DMAIC approach. You are asked to design an
experiment to look at the effect of five variables on a particular
response. You present a 2
5-1
fractional factorial design to the
team. After the meeting, you hear from the technician that the
process engineer (who was not at the meeting) has reviewed the
experiment design and said that 5 of the treatment combinations
are not going to work and that he is unwilling to provide
experimental units for those runs.
What is your role on this team?
How do you make this successful?
ASU Department of Industrial Engineering 2004
Case Studies
You are mentoring a new Black Belt candidate who has recently
been assigned an improvement project to lead. You have been
asked to sit in and observe this team and this candidate as they
work on this project. The team has been given 8 weeks to work
on this issue. Four weeks into the project, the team is
significantly behind schedule and it is obvious that the candidate
has no idea how to get this team back on track.
What is your role on this team?
What do you do to make this successful?
This document and all information and expression contained herein are the property of ASU Department of Industrial Engineering, and may not, in whole or
in part, be used, duplicated, or disclosed for any purpose without prior written permission of ASU Department of Industrial Engineering. All rights reserved.
ASU Department of Industrial Engineering 2004
IEE 598 Six-Sigma Methodology
Communicating Results and
Consulting
Dr. Cathy Lawson Medtronic
cathy.a.lawson@medtronic.com
ASU Department of Industrial Engineering 2004
Whos the audience?
Management
Customers
Peers
Team
Hourly Employees
One of the keys to a successful presentation is to know
your audience
ASU Department of Industrial Engineering 2004
Formats for presenting results
Formal presentation
Executive Summary
Journal article
Informal briefing
Written report
Memo
Minutes
ASU Department of Industrial Engineering 2004
Good Etiquette for Formal Presentations
Keep to your time limit
Allow time for questions
Have the number slides you present be approximately
equal to 1/3 the number of minutes you have to
present
Make sure your slides are readable from all areas of
the room
Dont assume everyone knows your acronyms
Be aware of the audience reception of your
presentation and make adjustments as necessary
ASU Department of Industrial Engineering 2004
Audience Analysis - Management/Customers
Summarize concisely
Prepare an executive summary
Make practical recommendations
Relate subject to the bottom line
Identify 1-3 key takeaways
Be prepared to discuss how you performed within the
given constraints.
ASU Department of Industrial Engineering 2004
Audience Analysis - Peers/Teammates
Prepare technical reports with an abstract
Use graphical techniques to convey your results
Translate the statistical analysis into the specific
actions the team needs to take
Document results as you go
ASU Department of Industrial Engineering 2004
Audience Analysis Hourly Employees
Use communication as a form of training
Keep it simple
Relate to their job
Acknowledge their expertise
Be aware that the hourly employee may have some
fear about what you are doing
ASU Department of Industrial Engineering 2004
The Art of Consulting
The practice of consulting is to provide guidance,
knowledge and assistance to the individual or
institution making the request on a topic(s) in which
the consultant is a recognized expert.
The art of consulting is being able to provide
meaningful guidance, knowledge and assistance to
the individual or institution which enables them to
perform better.
ASU Department of Industrial Engineering 2004
Black Belts as Consultants
Black Belts within an organization become an internal
resource with whom people will consult.
Technical expertise
Statistical applications
Use of continuous improvement tools
Problem solving/prevention strategies
ASU Department of Industrial Engineering 2004
Consulting Skills
Listening
Questioning
Framing
Advising
Mentoring
Problem Solving
Influencing
ASU Department of Industrial Engineering 2004
Case Study
An engineer comes to you asking for help in
designing an experiment to help solve a
problem.
What do you do next?
ASU Department of Industrial Engineering 2004
Case Study
A process engineer comes to you asking for
advice. He says that the factory has just
solved a technical problem and put a fix in
place. They want to verify that the fix is
successful. How many samples do they need
to collect in order to prove that?
What do you do next?
ASU Department of Industrial Engineering 2004
Case Study
A project is having numerous quality
problems. The manager of the area wants you
to come in and make sure the team is using
the proper statistical tools to help them solve
the problems.
What do you do next?
ASU Department of Industrial Engineering 2004
Case Study
A team needs to run an experiment to
investigate the effect of a proposed design
change on a new product design. There are
two variables of interest under investigation.
The team says they can only provide 5
experimental units.
How can you help them?
IEE 581 Six Sigma Methodology
Categorical Data Analysis, Part 1
Fall 2012 10/11/2012
Cheryl L. Jennings, PhD, MBB
c.jennings@asu.edu
1
Categorical Data Analysis
Analysis of Categorical, or Discrete, RESPONSE data
Can be encountered throughout the DMAIC process
Define
VOC
Measure
MSAs
Process capability studies
Analyze
Hypothesis testing, regression
Improve
Designed experiments
Control
SPC charts
2
Categorical Data Analysis Techniques
Attribute Agreement Analysis
Agreement tables (Measure lecture)
Kappa (Measure lecture)
Interclass Correlation
Kendalls Coefficient of Concordance
Process Capability
DPMO for discrete data (Measure
lecture)
Contingency Table Analysis
Test for Independent classifications
Test for Homogeneity across
categories
Logistic Regression Analysis
Binary
Nominal
Ordinal
Attribute Control Charts (Control
lecture, SPC course)
3
Measurement System Analysis Properties
The properties analyzed for a system measuring continuous data are
Discrimination
Accuracy (or Bias)
Stability
Linearity
Gauge Repeatability & Reproducibility (GR&R)
There are many situations where the output of a measurement system or an
evaluation is an attribute, for example pass/fail, classification or rating.
Some of the usual MSA properties cannot be evaluated, but they do suggest
properties that should be considered.
4
Attribute Agreement Analysis
An Attribute Agreement Analysis can be used to evaluate
Within Appraiser agreement, or Repeatability
Between Appraiser agreement, or Reproducibility
Appraiser agreement versus a standard if available, or Bias
The Attribute MSA shown in the Measure lecture included
Agreement tables
Kappa statistics
The Intra-correlation coefficient and Kendalls Coefficient of Concordance are
additional methods for evaluating agreement.
5
Small Business Loan Example
Small business loans are reviewed to approval parameters in a manual underwriting
process, and the decision is made to either approve or decline the loan application.
Consistency in underwriting decisions is desired at several levels:
Consistency with decision of a subject matter expert
Consistency when provided with the same application
Consistency between underwriters
6
Agreement Tables for Loan Decisioning
7
Within Appraisers
Assessment Agreement
Appraiser # Inspected # Matched Percent 95% CI
1 14 14 100.00 (80.74, 100.00)
2 14 11 78.57 (49.20, 95.34)
3 14 14 100.00 (80.74, 100.00)
# Matched: Appraiser agrees with him/herself across trials.
Each Appraiser vs Standard
Assessment Agreement
Appraiser # Inspected # Matched Percent 95% CI
1 14 11 78.57 (49.20, 95.34)
2 14 9 64.29 (35.14, 87.24)
3 14 10 71.43 (41.90, 91.61)
# Matched: Appraiser's assessment across trials agrees with the known standard.
Between Appraisers
Assessment Agreement
# Inspected # Matched Percent 95% CI
14 8 57.14 (28.86, 82.34)
# Matched: All appraisers' assessments agree with each other.
All Appraisers vs Standard
Assessment Agreement
# Inspected # Matched Percent 95% CI
14 6 42.86 (17.66, 71.14)
# Matched: All appraisers' assessments agree with the known standard.
*Minitab output, data is in categorical data sets.xlsx
Kappa Statistic
Kappa is defined as the proportion of agreement between Appraisers after
agreement by chance has been removed. The formula is:
Values range from -1 to +1. The higher the value, the stronger the agreement
between the rating and standard.
If K = 1, is perfect agreement. K > 0.9 is excellent; K < 0.7 needs improvement.
If K = 0, then agreement is the same as would be expected by chance.
If K < 0, then agreement is weaker than expected by chance.
Hypotheses are:
H0: Level of agreement is the same as that expected by chance alone (K = 0).
HA: Level of agreement is significantly different (stronger or weaker) than
expected by chance.
8
1
Observed Expected
Expected
P P
K
P
= + = + =
= + =
= =
Between Appraisers
Assessment Agreement
# Inspected # Matched Percent 95% CI
14 11 78.57 (49.20, 95.34)
# Matched: All appraisers' assessments agree with each
other.
Fleiss' Kappa Statistics
Response Kappa SE Kappa Z P(vs > 0)
A 0.550802 0.267261 2.06091 0.0197
D 0.550802 0.267261 2.06091 0.0197
*Minitab output, data is in categorical data sets.xlsx
Kappa Statistics for Loan Decisioning
10
Within Appraisers
...
Fleiss' Kappa Statistics
Appraiser Response Kappa SE Kappa Z P(vs > 0)
1 A 1.00000 0.267261 3.74166 0.0001
D 1.00000 0.267261 3.74166 0.0001
2 A 0.55080 0.267261 2.06091 0.0197
D 0.55080 0.267261 2.06091 0.0197
3 A 1.00000 0.267261 3.74166 0.0001
D 1.00000 0.267261 3.74166 0.0001
Each Appraiser vs Standard
...
Fleiss' Kappa Statistics
Appraiser Response Kappa SE Kappa Z P(vs > 0)
1 A 0.569231 0.188982 3.01209 0.0013
D 0.569231 0.188982 3.01209 0.0013
2 A 0.492949 0.188982 2.60844 0.0045
D 0.492949 0.188982 2.60844 0.0045
3 A 0.416667 0.188982 2.20479 0.0137
D 0.416667 0.188982 2.20479 0.0137
Between Appraisers
...
Fleiss' Kappa Statistics
Response Kappa SE Kappa Z P(vs > 0)
A 0.628361 0.0690066 9.10581 0.0000
D 0.628361 0.0690066 9.10581 0.0000
All Appraisers vs Standard
...
Fleiss' Kappa Statistics
Response Kappa SE Kappa Z P(vs > 0)
A 0.492949 0.109109 4.51795 0.0000
D 0.492949 0.109109 4.51795 0.0000
*Minitab output,
data is in
categorical data
sets.xlsx
Intraclass Correlation Coefficient*
If a measurement system is needed to simply classify objects in a non-quantitative
manner, then the Kappa statistic is appropriate.
If it is possible to rank or order objects in some way, Intraclass Correlation
Coefficients (ICC) should be used to evaluate the agreement of the ratings.
The distance between the ordered categories is perceived as roughly equal.
ICCs are ratios of between rating variance to total variance, comparing the
covariance of the ratings with the total variance.
Variations of the ICC are a function of
How Appraisers are selected
How Ratings are used (individual or averaged)
11
*from Futrell (1995)
Six Types of ICCs*
Variations of the ICC are set up the same way and differ only in treatment of
components of variation. The six types of ICCs are:
1. Each object is rated by a different set of Appraisers, randomly selected from a
larger population, and the reliability of each Appraisers rating is of interest.
2. Each object is rated by a different set of Appraisers, randomly selected from a
larger population, and reliability of the Appraisers averaged rating is of interest.
3. Each object is rated by a single set of Appraisers, randomly selected from a
larger population, and the reliability of each Appraisers rating is of interest.
4. Each object is rated by a single set of Appraisers, randomly selected from a
larger population, and reliability of the Appraisers averaged rating is of interest.
5. Each object is rated by a set of Appraisers, who are the only Appraisers of
interest (there is not a larger population), and reliability of each Appraisers
rating is of interest.
6. Each object is rated by a set of Appraisers, who are the only Appraisers of
interest, and the reliability of the Appraisers averaged rating is of interest.
If ratings are from a single Appraiser, then 1, 3 and 5 apply. If ratings are averaged
across multiple Appraisers, then 2, 4 and 6 apply.
12
*from Futrell (1995)
ICC for Supplier Evaluation
Package Buyer 1 Buyer 2 Buyer 3
1 5 7 6
2 6 5 4
3 4 4 3
4 4 5 4
5 7 6 5
6 6 7 7
7 8 9 8
8 9 8 8
9 5 5 6
10 6 7 8
Supply management wants to evaluate
a supplier based on the completeness
of purchase orders.
Three senior buyers will evaluate ten
purchase orders and rate
completeness from 1 (poor) to 10
(excellent).
If 3 different randomly selected
buyers rate each purchase order,
then 1 & 2 apply.
If the same 3 randomly selected
buyers rate all 10 purchase orders,
then 3 & 4 apply.
If there are only 3 buyers in the
organization and each buyer rates
all 10 purchase orders, then 5 & 6
apply.
13
ICC for Supplier Evaluation (continued)
14
7.32 0.6
ICC1 0.79
( 1) 7.32 (3 1)0.6
7.32 0.6
ICC2 0.92
7.32
ICC3
( 1) ( ) /
7.32 0.62
0.79
7.32 (3 1)0.62 3(0.43 0.62) / 10
ICC4
( ) /
7.32 0
BMS WMS
BMS k WMS
BMS WMS
BMS
BMS EMS
BMS k EMS k JMS EMS n
BMS EMS
BMS JMS EMS n
= = =
+ +
= = =
=
+ +
= =
+ +
=
+
=
.62
0.92
7.32 (0.43 0.62) / 10
7.32 0.62
ICC5 0.78
( 1) 7.32 (3 1)0.62
7.32 0.62
ICC6 0.92
7.32
BMS EMS
BMS k EMS
BMS EMS
BMS
=
+
= = =
+ +
= = =
Kendalls Coefficient of Concordance*
For ordinal data, Kendalls Coefficient of Concordance (KCC) can also be used to
measure agreement in ratings.
It is a measure of total correlation that indicates the degree of association of ordinal
assessments made by multiple Appraisers when evaluating the same objects.
Values range from 0 to +1.
If there is perfect agreement, with objects assigned same rating by all Appraisers,
W = 1.
If there is perfect disagreement among rankings, W will be zero (or very close).
A high or significant Kendall's coefficient means that the Appraisers are applying
essentially the same standard when evaluating the objects (Agreement is not due to
chance).
15 *from Conover (1980)
Minitabs Approach to Estimating KCC
Minitab estimates KCC as
where:
N = the number of subjects.
E R
i
2
= the sum of the squared sums of ranks for each of the ranked N subjects
K = the number of appraisers.
T
j
assigns the average of ratings to tied observation; T
j
=
where t
i
= the number of tied ranks in the i
th
grouping of ties, and g
j
= the
number of groups of ties in the j
th
set of ranks.
To test significance of Kendall's coefficient, use _
2
= k (N - 1) W where W is the
calculated Kendall's coefficient. _
2
is distributed as chi-square with N - 1 degrees of
freedom.
16
W =
KCC for Supplier Evaluation
17
Attribute Agreement Analysis for Buyer1, Buyer2, Buyer3
Between Appraisers
Assessment Agreement
# Inspected # Matched Percent 95% CI
10 0 0.00 (0.00, 25.89)
# Matched: All appraisers' assessments agree with each other.
Fleiss' Kappa Statistics
Response Kappa SE Kappa Z P(vs > 0)
3 -0.034483 0.182574 -0.18887 0.5749
4 0.280000 0.182574 1.53362 0.0626
5 -0.041667 0.182574 -0.22822 0.5903
6 -0.250000 0.182574 -1.36931 0.9145
7 0.040000 0.182574 0.21909 0.4133
8 0.280000 0.182574 1.53362 0.0626
9 -0.071429 0.182574 -0.39123 0.6522
Overall 0.037433 0.081300 0.46043 0.3226
Kendall's Coefficient of Concordance
Coef Chi - Sq DF P
0.858947 23.1916 9 0.0058
None of the
Buyers ratings
matched
Overall suggests
poor
consistency in
ratings between
Buyers
Suggests consistency between
Buyers is acceptable (i.e., not
due to chance)
*Minitab output, data is in categorical data sets.xlsx
Hot Sauce Example from Futrell (1995)
Wilson and Justin are visiting New
Orleans and are overwhelmed by the
varieties of local hot sauces available.
As they taste a few varieties, they
notice that they seem to agree about
how hot each one is.
Since Wilson aspires to become a
psychometrician, he designs a study to
measure their agreement.
They randomly purchase 10 bottles of
hot sauce and independently classify
them into four categories:
Mild(M)
Hot (H)
Very hot (VH)
Makes me suffer (MMS)
Sauce Wilson Justin
1 M (1) MM (1)
2 M (1) H (2)
3 MMS (4) VH (3)
4 VH (3) MMS (4)
5 H (2) VH (3)
6 VH (3) VH (3)
7 H (2) M (1)
8 H (2) H (2)
9 MMS (4) VH (3)
10 M (1) H (2)
18
Kappa for Hot Sauce Ratings
19
Results for: Hot Sauce.MTW
Attribute Agreement Analysis for Wilson, Justin
Between Appraisers
Assessment Agreement
# Inspected # Matched Percent 95% CI
10 3 30.00 (6.67, 65.25)
# Matched: All appraisers' assessments agree with each other.
Fleiss' Kappa Statistics
Response Kappa SE Kappa Z P(vs > 0)
H 0.047619 0.316228 0.150585 0.4402
M 0.200000 0.316228 0.632456 0.2635
MMS -0.176471 0.316228 -0.558049 0.7116
VH 0.047619 0.316228 0.150585 0.4402
Overall 0.047619 0.187155 0.254436 0.3996
*Minitab output, data is in categorical data sets.xlsx
ICC & KCC for Hot Sauce Ratings
20
1.8944 0.3833
ICC5 0.66
( 1) 1.8944 (2 1)0.3833
1.8944 0.3833
ICC6 0.80
1.8944
BMS EMS
BMS k EMS
BMS EMS
BMS
= = =
+ +
= = =
Results for: Hot Sauce.MTW
Attribute Agreement Analysis for Wilson, Justin
Between Appraisers
Kendall's Coefficient of Concordance
Coef Chi - Sq DF P
0.731148 13.1607 9 0.1555
*Minitab output, data is in categorical data sets.xlsx
Summary of Agreement Analysis Techniques
Agreement Tables
Often used for transactional processes to assess whether appraisers are making
the same decisions
Require relatively more data
Interpretation of results can be subjective
Kappa Statistic
Data classification is categorical: pass/fail, attribute, qualitative feature
Results are less subjective than for Agreement tables
All misclassifications are treated equally
Intracorrelation Coefficient and Kendalls Coefficient of Concordance
Data classification is ordinal and hence quantitative
Misclassification is not treated equally across all ratings
The consequence of rating a Make me suffer hot sauce as Mild is much
more serious than rating as Hot
21
Contingency Table Analysis
Test for Independent classifications
Test for Homogeneity across categories
22
r c Contingency Tables
Consider a sample of size N from a
single population, with observations
that can be classified according to two
criteria.
Results are tabulated into r rows
associated with the 1
st
criterion and c
columns associated with the 2
nd
criterion.
This is referred to as an r c
contingency table.
The question is: Are the two methods
of classification statistically
independent?
Contingency tables answer this
question by comparing the observed
frequencies in the cells of the table to
the expected frequencies.
2
nd
Criterion
1 2 . c Totals
1
st
1 O
11
O
12
. O
1c
R
1
Criterion 2 O
21
O
22
. O
2c
R
2
. . . . . .
r O
r1
O
r2
. O
rc
R
r
Totals C
1
C
2
. C
c
N
This is the first application of a
contingency table a test for
Independence, also known as the Chi-
Square Test for Independence.
Suppose that there are only two
categories (columns) success/ failure,
defective/non-defective. Then the
contingency table could be used to test
the equality of r binomial parameters.
23
Contingency Tables Test for Independence*
24
*from Montgomery & Runger (2011)
Is Pension Plan Preference Independent of Job Classification?
25
*from Montgomery & Runger (2011)
26
*from Montgomery & Runger (2011)
Independence Test in Minitab
27
Chi-Square Test: Plan1, Plan2, Plan3
Expected counts are printed below observed counts
Chi-Square contributions are printed below expected counts
Plan1 Plan2 Plan3 Total
1 160 140 40 340
136.00 136.00 68.00
4.235 0.118 11.529
2 40 60 60 160
64.00 64.00 32.00
9.000 0.250 24.500
Total 200 200 100 500
Chi-Sq = 49.632, DF = 2, P-Value = 0.000
*Minitab output, data is in categorical data sets.xlsx
Contingency Tables Test for Homogeneity
The second application is a test for
Homogeneity, also known as the Chi-
Square Test for Differences in
Probabilities.
Consider r samples drawn from
different populations, with
observations classified into c
categories.
The question is: Do the r samples
have the same proportions of
elements in a certain category?
H0: All the probabilities in the same
column are equal to each other.
HA: At least two of the probabilities
in the same column are not equal to
each other.
Computations are the same as for the
test for independence.
The same approach can also be used
to answer the question: Does a
treatment significantly alter the
proportion of objects in each of two
classifications?
The population before & after the
treatment is represented in the
rows.
28
Categories
1 2 . c Totals
Population 1 O
11
O
12
. O
1c
R
1
Population 2 O
21
O
22
. O
2c
R
2
. . . . . .
Population r O
r1
O
r2
. O
rc
R
r
Totals C
1
C
2
. C
c
N
Did the Experimental Method Result in Better Learning?
Sixty newly-hired employees were
divided into two groups of 30 each and
taught how to complete an order
transaction. One group used the
conventional method of learning, and
the other group used a new,
experimental method.
At the end of the courses, each new
employee was given a test that
consisted of completing the order
transaction. The order transaction
was either correct or incorrect.
The data is shown to the right.
Is there reason to believe that the
experimental method is superior? Or
could the differences be due to chance
fluctuations?
Correct
Order
Incorrect
Order
Conventional
Group
23 7
Experimental
Group
27 3
29
Homogeneity Test in Minitab
30
Chi-Square Test: Correct, Incorrect
Expected counts are printed below observed counts
Chi-Square contributions are printed below expected counts
Correct Incorrect Total
1 23 7 30
25.00 5.00
0.160 0.800
2 27 3 30
25.00 5.00
0.160 0.800
Total 50 10 60
Chi-Sq = 1.920, DF = 1, P-Value = 0.166
*Minitab output, data is in categorical data sets.xlsx
Special Application 2 2 Contingency Tables*
In a 2 2 contingency table, two
samples have observations that may
be categorized into one of two classes.
The alternative hypothesis can have
three forms:
Two-sided Test
H
0
: p
1
= p
2
H
A
: p
1
p
2
One-sided Test (upper tail)
H
0
: p
1
p
2
H
A
: p
1
> p
2
One-sided Test (lower tail)
H
0
: p
1
p
2
H
A
: p
1
< p
2
Test Statistics
For the two-sided test, the test
statistic is the same, more simply:
For the one-sided tests,
Decision rules for one-sided tests
Upper Tail
Reject H
0
at the approximate level
if T
1
exceeds Z
Lower Tail
Reject H
0
at the approximate level
if T
1
is less than -Z
31 *from Conover (1980)
2
2 11 22 12 21
0
1 2 1 2
( ) N O O O O
n n CC
_
=
11 22 12 21
1
1 2 1 2
( )
(0,1)
N O O O O
T N
n n CC
= ~
2 2 Contingency Table Example One-Sided Test*
At the US Naval Academy, a new
lighting system was installed
throughout the midshipmens living
quarters. It was claimed that the new
lighting system resulted in poor
eyesight due to a continual strain on
the eyes of the midshipmen.
Consider a study to test this claim:
H
0
: The probability of a graduating
midshipman having 20-20 (good)
vision is the same or greater under
the new lighting than it was under
the old lighting, versus
H
A
: The probability of good vision is
less now than it was.
Good Vision Poor Vision
Old Lights 714 111
New Lights 662 154
32
*from Conover (1980)
2 2 Example in Minitab
This is a one-sided test, with H
A
: p
Old
>
p
New
However Minitab does not have a one-
sided test option
Run as two-sided test in Minitab, and
take square root of calculated chi-
square test statistic
Compare to standard normal
distribution
Reject H
0
if T
1
> Z
For = 0.05, Z = 1.65
Decision: Reject H
0
, the two classes do
differ with respect to the proportions
having poor eyesight. Vision is worse
with the new lights.
33
Chi-Square Test: Good, Poor
Expected counts are printed
below observed counts
Chi-Square contributions are
printed below expected counts
Good Poor Total
1 714 111 825
691.77 133.23
0.714 3.708
2 662 154 816
684.23 131.77
0.722 3.749
Total 1376 265 1641
Chi-Sq = 8.893, DF = 1, P-Value
= 0.003
1
8.893 2.98 T = =
*Minitab output, data is in categorical data sets.xlsx
References
Bower, K.M. Trainers Corner: Measurement system analysis with attribute data.
http://www.minitab.com/uploadedFiles/Shared_Resources/Documents/Articles/measurement
_system_analysis_attribute_data.pdf (Kappa statistic and Kendalls Coefficient of
Concordance)
Conover, W.J. (1980). Chapter 4, Contingency Tables, Practical Nonparametric Statistics, 2
nd
edition. Wiley, New York.
Futrell, D. (1995), When Quality is a Matter of Taste, Use Reliability Indexes, Quality Progress,
May 1995. (Available for purchase or free to members, at www.asq.org)
Montgomery, D.C. and Runger, G.C. (2011). 9-8 Contingency Table Tests. Applied Statistics
and Probability for Engineers, 5
th
edition. Wiley, New York.
NIST/SEMATECH e-Handbook of Statistical Methods,
http://www.itl.nist.gov/div898/handbook/, 10/2010. (contingency tables)
34
Useful Minitab Documentation
Minitab Assistant White Paper
Attribute Agreement Analysis: http://www.minitab.com/en-
US/support/documentation/Answers/Assistant%20White%20Papers/AttributeAgreementAn
alysis_MtbAsstMenuWhitePaper.pdf
Minitab Technical Support Documents
Attribute Agreement Analysis: http://www.minitab.com/en-
US/support/documentation/Answers/AttribAgreeAnalysisTutorial.pdf
Service Quality Example Using Ordinal Data:
http://www.minitab.com/support/documentation/Answers/SQAttributeAgreementAnalysis.
pdf
Manufacturing Example Using Binary Data:
http://www.minitab.com/support/documentation/Answers/MANUAttributeAgreementAnal
ysis.pdf
35
Upcoming
No class on Tuesday 10/16
Next class on Thursday 10/18
Assignment #1 due on Tuesday 10/23
Mid-term exam on Tuesday 10/23
Following four lectures on DFSS and Lean with Dr. Holcomb ENJOY!
36